7.5 Health economics and public health
Oxford Textbook of Public Health
Health economics and public health
M. Christopher Auld, Cam Donaldson, Craig Mitton, and Phil Shackley
Key concepts in micro-economics
Rational choice theory
Opportunity costs and constraints
Techniques of economic appraisal for programme evaluation
Identification of costs and benefits
What question is being asked?
Identifying opportunity costs
Categorizing costs and benefits
How far and how wide should we go?
Measurement and valuation of costs
Counting costs in a base year
Marginal or incremental costing
Patient-based versus per diem costs
Unthinking acceptance of market values
Measurement and valuation of benefits
Measures used in cost-effectiveness analysis
Generic quality-adjusted life years
Condition-specific quality-adjusted life years and healthy years equivalents
The use of cost per quality-adjusted life year data in health-care decision-making
Using willingness to pay data in health-care decision-making
Programme budgeting and marginal analysis
Micro-economic analysis of public health policy: more theory and some examples
A model of consumer behaviour
Demand for safety and the curious effects of airbags
Equivalent annual costs
Economics is the science of choice. In public health, such choice can take place at two levels: for governments or other health purchasers, between different health-care and public health programmes, or, for an individual, in terms of how health is ‘traded’ against other goods in life. The former level of choice requires economic evaluation and the latter requires an understanding of the contribution of economics to the analysis of how individuals respond to changes in health policy or to changes in the (perceived) prevalence of disease.
In the following section, some basic concepts from micro-economics, which underlie all economic theories of choice, are defined. These are ‘rational choice theory’, ‘opportunity cost’, and ‘production’. Following this, the focus will be on what most public health practitioners probably perceive as the main contribution of health economics: the use of economic evaluation in appraising interventions. The third section begins with an introduction to the main techniques of economic evaluation. In the fourth section, a list of costs and benefits, which should be considered for inclusion in economic evaluations, is provided. Principles of measurement and valuation of costs are then outlined in the fifth section. In the sixth section, concentration is on valuing the benefits of health care. In the seventh section, following presentations of economic evaluation methods, a priority setting framework for use by health (care) purchasers, known as programme budgeting and marginal analysis, is outlined. The remainder of the chapter focuses on the growing contribution of economics to analysing how individuals respond to public health and health promotion policies. Individuals, either implicitly or explicitly, weigh up the costs and benefits of their actions, some of which affect health. Any policy aimed at influencing health will, therefore, change these costs and benefits. Therefore, the eighth section extends the opportunity cost concept to a model of consumer behaviour, and demonstrates the usefulness of this framework for predicting the effects of public health policies. The conclusion summarizes the contribution of economics to evaluating public health practice and policy, and includes a checklist for measurement and valuation of costs and benefits in economic evaluation.
Key concepts in micro-economics
Economics is most famously defined as ‘the science which studies human behaviour as a relationship between ends and scarce means which have alternate uses’ (Robbins 1935). In shorter form, economics is about choice: starting with the premise that individuals have ends (for instance, comfort, health, social status, and pleasure) and limited resources to achieve those ends, economics seeks to make positive statements regarding how social systems work and make normative statements regarding which policies can be expected to make people better off.
Rational choice theory
Economists typically model behaviour as resulting from rational choices on the part of individuals. It is important to emphasize that ‘rational’ in this context is a piece of technical jargon only loosely related to its more common meaning. In economics, individuals are rational if they are aware of and can rank the options available to them and have consistent beliefs about the way the world works. The ranking itself can be utterly irrational in the common use of the word: for instance, crack addicts can be perfectly rational in the economic sense, even if economists and everyone else agrees that their behaviour is perfectly irrational in the everyday sense. See Becker and Murphy (1988) for an analysis of rational choice in the context of addictive goods.
What is meant by a ‘consistent ranking of outcomes?’ Suppose there are three outcomes of interest, A, B, and C. Loosely, the individual is said to have rational or consistent preferences over these outcomes if it is the case that (a) they can state they either prefer one outcome to another, or they are indifferent between the two outcomes, and (b) if A is preferred to B and B is preferred to C, then it is not the case that C is preferred to A. These are minimal requirements to allow a model to make predictions about what a person will choose in a given situation. They are further required to make normative statements about various policies: if preferences cycle (that is, C is preferred to A in the example above), then we cannot say what outcome makes the person better off.
If it is the case that a person is rational in the above sense, then, as an analytical convenience, it is possible to assign numbers to various outcomes such that the ranking of the numbers captures the individual’s subjective ranking of outcomes: if A is preferred to B, then the number attached to A is higher than the number attached to B. See Mas-Colell et al. (1995) for an extensive discussion.
Any such assignment of numbers to outcomes is called a utility function. Since the most preferred outcome available in a given situation will be the one selected, and since, by construction, that outcome will have the highest utility, the decision can be recast as one of maximizing utility subject to whatever constraints the individual faces. The tool ‘utility function’ is therefore fundamental to economic analysis.
Opportunity costs and constraints
The key concept of cost in economics is opportunity cost: the value of the next best alternative forgone. Suppose, for instance, that you have an afternoon off and have narrowed your selection of what to do with it to either reading a novel or gardening. The opportunity cost of reading the novel is the forgone opportunity to garden, and vice versa. The observation that whenever a decision is made, regardless of the context, an opportunity cost is incurred leads to both the labelling of economics as ‘the dismal science’ and Milton Friedman’s pithy truism, ‘there is no such thing as a free lunch’. For instance, if a millionaire donates funds to build a new wing on a hospital, the economist will tend to be the curmudgeonly person at the back of the room pointing out the opportunity cost of the new wing was an increase in the nursing staff, which may have better served patients. Notice that the economist’s notion of costs is not necessarily even related to money, as in the novel/gardening example above.
Since costs are defined in terms of forgone opportunities, in order to evaluate interventions and make predictions about behaviour, it is important to specify what options are available to the decision-maker. Given the assumption of rationality and the useful device of a utility function to represent a given individual’s rational preferences over the domains of interest, if the set of choices available are also specified it is possible to predict a choice and aid the evaluation of options. The content of economic analysis is in the implications arising from the different choices available either to individuals or social decision-makers. For example, at the level of programme evaluation, by using resources to meet one need, opportunities to use these resources in other programmes are forgone. Therefore, to ensure the most efficient use of resources, it is necessary to know the resources consumed (that is, costs), and benefits produced, by each programme. Only then is it possible to choose that combination of programmes which maximizes benefits to the community.
In addition to analysing individual behaviour using the above concepts, another useful tool of economics involves analysis of the decisions of governments, non-profit agencies, firms, and individuals which produce some output using a variety of inputs. For instance, an automotive firm uses labour, machinery, land, and time to produce cars. A hospital might use doctors, nurses, and machines to ‘produce’ health in patients, and individuals might use time and market goods to ‘produce’ their own health (for instance, by buying and using an exercise machine).
The basic tools in production theory can be illustrated with an example. Consider a hospital which produces health, H, using only physicians, P, and nurses, N. We can write the relationship between health produced and the inputs as a function H = H(P, N). If the hospital can hire nurses for WN dollars and can hire physicians for WP dollars per unit of time, the hospital’s payroll for that unit of time will total C = WPP + WNN. To produce the most health for any given budget C, the hospital must solve the following problem:
maxP,N H(P, N)
WNN + WPP = C.
The solution gives the number of nurses and physicians hired to make patients as healthy as possible for a given budget. Notice that the economic logic here is based on opportunity cost; the hospital should keep hiring nurses up until the point where the opportunity cost of another nurse, in terms of the health that forgone physician services would provide, is just equal to the health benefits the marginal nurse provides. The problem above is illustrated in Fig. 1. The iso-cost line represents all combinations of nurses and physicians which can be afforded; being on the line represents the limits of affordability. The iso-quant curve, H(), represents different quantities of ‘health’ which can be produced by different combinations of physicians and nurses. The reason it is this shape is because as fewer and fewer nurses are employed, it would take more and more physicians to ensure the same level of production, and vice versa. Theoretically, there are an infinite number of iso-quants in the diagram, although they are not allowed to cross each other. Given that costs cannot exceed the limits set by the iso-cost curve shown, the highest level of output which can be achieved is at the point (N*, P*). Any other point either costs more money (C is higher) or is associated with lower output of health. This point is also cost-effective in that a given amount of health is achieved at least cost, thus providing the basis for one of the economic evaluation techniques discussed in the following section.
Fig. 1 Maximizing production.
This model of a hospital is very simple and is used to illustrate how economists view both organizations and individuals as ‘producing’ outcomes of interest given available resources. Having outlined the basic methods of analysis employed by micro-economists in thinking about health-related matters, it is possible to consider how these concepts are used in economic evaluation and in evaluating behavioural responses to public health policies.
Techniques of economic appraisal for programme evaluation
Following on from the above concepts, there are three related economic appraisal techniques; cost–benefit analysis, cost-effectiveness analysis, and cost–utility analysis. Each technique has advantages and difficulties associated with it. However, as Mooney (1989) has said: ‘ease cannot be allowed to dictate use; it is a question of what is best for which question’. Therefore, which technique to use should be determined by the question to be addressed in a given evaluation.
Principally, economic evaluation is useful in addressing two levels of question—about allocative and technical efficiency. With allocative efficiency, all health-care programmes have to compete with each other for more (or fewer) resources. Concern is with how many of the scarce resources to allocate to each programme. Thus, in Table 1, surgery for tonsillectomy and outpatient clinics for asthmatics would compete with each other for more (or fewer) resources.
Table 1 Questions of allocative and technical efficiency
With technical efficiency, concern is more about how best to deliver a programme or to achieve a given objective. The resources already allocated to the programme are taken as given. Therefore, in the cases outlined in Table 1, technical efficiency questions would address how best to deliver surgery for tonsillectomy and how best to provide asthma clinics.
Generally, with allocative efficiency, one group of patients/clients will gain at the expense of another group. With technical efficiency, the same group will be treated, but the question is how. The distinction is important, although it often becomes blurred in practice.
Cost–benefit analysis is used to address allocative efficiency questions, such as:
Is it worth allocating resources to achieving this goal?
How much more or how much less of society’s resources should be allocated to achieving this goal or to this type of health care?
Thus, cost–benefit analysis can appear to involve looking at one health-care programme in isolation (although the alternative of doing nothing, or current practice, is always implied). Looking at one programme alone, cost–benefit analysis addresses the question of whether its benefits are greater or less than its opportunity costs. Looking at several competing projects, comparisons are made on the basis of the costs and benefits of each. If not all projects can be funded, the combination that maximizes benefits should be chosen. In both cases, costs and benefits still need to be known.
In principle, answers to the above questions require all costs and benefits to be valued in a commensurate unit (such as money). If everything is measured in one unit, comparisons are straightforward. The valuation of benefits in money terms is the most obvious distinguishing characteristic of cost–benefit analysis. Conceptually, cost–benefit analysis has a very wide range of applicability as, with everything valued in commensurate terms, it could be used to compare health-care objectives with each other or with those arising in other sectors of the economy.
In practice, however, the monetary valuation of benefits in cost–benefit analysis is difficult. How can a value be placed on the saving of life or the relief of suffering? Health improvements are traditionally valued in money terms by using one of two approaches: the human capital approach and the willingness to pay approach. Each of these is reviewed in the section below on measurement and valuation of benefits.
Not surprisingly, ideal cost–benefit analyses are rare in the field of health care, despite the name being included in the titles of many articles. In many cost–benefit analyses, the cost of a new intervention is often assessed against ‘benefits’ measured in terms of cost savings. However, this clearly involves only a comparison of costs. No consideration is given to the difficult issue of valuing health improvements and other benefits in monetary terms (Birch and Donaldson 1987). Two exceptions are recent evaluations of public health measures of reducing food-borne risk and water fluoridation (Donaldson et al. 1996; Shackley and Dixon 2000).
Despite problems of valuing benefits, cost–benefit analysis remains a useful tool, particularly in setting out a decision-making problem. By identifying the costs and benefits associated with different health-care programmes and valuing what can be valued, one can explicitly observe the trade-offs between tangible and intangible costs and benefits resulting from a decision to implement, or not to implement, a health-care programme. A good example of this is the work on costs and benefits of introducing child-proofing of drug containers in the United Kingdom in the early 1970s. In 1971 a decision was taken not to introduce such child-proofing on the basis of its cost (£500 000 per annum at the time). Using conservative estimates, Gould (1971) showed that without child-proofing there would be 16 000 hospital admissions per annum at a cost of £30 per admission, resulting in costs of £480 000 per annum. Thus, the extra cost of child-proofing was £20 000 per annum. If only 20 lives could be saved by child-proofing, the decision not to child-proof would imply that the life of a child was worth less than £1000 per annum.
Such trade-offs would almost certainly have remained implicit in the absence of cost–benefit analysis, thereby not allowing them to be subjected to the same degree of scrutiny. In the absence of such scrutiny, important aspects of efficiency may not be recognized. Cost–benefit analysis, by analysing who receives benefits and bears costs, also opens up important issues of the distribution (or equity) of health and health care which would have remained uncovered. Thus, although economics is often criticized for ignoring equity, it can actually highlight such issues. By quantifying some values and making others explicit, cost–benefit analysis can be a very useful decision-making aid. This ‘balance sheet’ approach to cost–benefit analysis has been outlined in more detail by McIntosh et al. (1999). It also forms the basis of the more practical form of economic evaluation, programme budgeting and marginal analysis (see below). Because decisions about what types of health care to provide cannot be avoided, subjecting decisions to the systematic framework that cost–benefit analysis provides, despite its imperfections, is useful. In the absence of cost–benefit analysis, values would remain implicit and, perhaps, be more prone to error. Avoiding the analysis does not avoid the need for a decision to be made.
Cost-effectiveness analysis is the most common form of economic evaluation in health care because of its relative simplicity. Its use does not require benefits to be valued in money terms. Cost-effectiveness analysis deals with technical efficiency and seeks to answer the following question. Given that it has been decided that a goal is to be achieved, what is the best way of doing so or what is the best way of spending a given budget? (Remember the point of tangency in Fig. 1.) Thus, cost-effectiveness analysis always involves comparison of at least two options with the same goal.
Cost-effectiveness analysis can take two main forms. In the first, if the health outcomes of the alternatives to be compared are known to be equivalent, only cost differences need be analysed. The least costly alternative is obviously most efficient as resources are saved which can be put to some other beneficial use without reducing the health outcomes of the client group being studied (Russell et al. 1977). This is often referred to as cost-minimization analysis.
In the second form of cost-effectiveness analysis, alternatives may differ in terms of cost and effect. A ratio is produced for each alternative in which the numerator is cost and the denominator is the health effect under consideration. Health effects are measures of final outcome: they might be life years saved, heart attacks prevented, or improved physical function. The cost-effectiveness ratio produced for each alternative is a measure of cost per unit of health effect. The alternative with the lowest cost-effectiveness ratio is best. Within a given budget, more health can be produced by implementing this alternative. For example, Doessel (1978) used a cost per life year saved ratio in comparing alternatives for the treatment of chronic renal failure.
The phrase ‘within a given budget’ is of crucial importance. Often authors produce a ratio of extra costs per extra unit of health effect of a new intervention over current treatment, and call their analysis an ‘incremental cost-effectiveness analysis’ (Boyle et al. 1983; Kristiansen et al. 1991; Mark et al. 1995; Johannesson et al. 1997). However, such studies are not cost-effectiveness analyses. Some judgement is required as to whether such extra costs are worth incurring, that is whether more resources should be allocated to that area of care. The resources to meet these extra costs will inevitably come from some other health-care programme (that is, either from another group of patients within the budget or from another budget altogether). This takes us back to broader comparisons, which is the role of cost–benefit analysis, not cost-effectiveness analysis.
Another limitation of cost-effectiveness analysis is the use of natural measures of outcome. For example, in evaluating different treatments for chronic renal failure, enhanced (or maintained) quality of life can be as important an outcome as improving life expectancy. Therefore, to use ‘life years gained’ as the outcome in a cost-effectiveness ratio will mean that other important aspects of outcome are missed. Also, one should be careful with outcome measures such as cholesterol reduction, blood pressure, and, even more so, cases detected. This is because the relationship between these and final outcome (that is, health) may be either unclear or not constant across different cases. Nevertheless, to reiterate, cost-effectiveness analysis is very useful for addressing issues of technical efficiency.
Cost–utility analysis lies somewhere between cost-effectiveness analysis and cost–benefit analysis. It can be used to assess not only technical efficiency, but also allocative efficiency within the health-care sector. The basic outcome in cost–utility analysis is ‘healthy years’. The difference between cost–utility analysis and cost-effectiveness analysis is that years of life in states less than ‘full health’ (for example, disability) can be converted to healthy years by the use of various techniques. To date, the most common forms of conversion are quality-adjusted life years and healthy years equivalents (Williams 1985; Torrance 1986; Mehrez and Gafni 1989; Gafni and Zylak 1990). Again, each of these techniques, and their challenges, is reviewed in the section below on measurement and valuation of benefits.
Cost–utility analysis may be seen as an improvement on cost-effectiveness analysis as it attempts to combine more than one outcome measure. Cost–utility analysis can also be seen as an improvement on cost–benefit analysis as it permits comparisons of programmes within the health sector without the need to place monetary values on their benefits. However, this advantage is limited when non-health benefits and non-health-care costs enter into analyses. So far, quality-adjusted life years and healthy years equivalents do not consider non-health aspects of benefit (for example, utility from process, such as the care environment in long-term care). Likewise, cost–utility analysis can address allocative efficiency within health services if health-care costs only are included. Once non-health benefits and non-health-care costs are included, a broader framework (cost–benefit analysis) is likely to be required for analysis of allocative efficiency (Gerard and Mooney 1993).
However, cost–utility analysis is valuable when quality of life is an important outcome. In particular, cost–utility analysis is important where there is the possibility that either one alternative evaluated is better than others in terms of effects on survival but worse in terms of quality of life, or one alternative is better than others in terms of some aspects of quality of life but worse in terms of other aspects.
Identification of costs and benefits
The principles of identifying costs and benefits in cost–benefit analysis, cost-effectiveness analysis, and cost–utility analysis are essentially the same for each type, although including a broader range of societal costs is likely to lead one towards a cost–benefit analysis. Identification of relevant costs and benefits involves a listing of all items of resource use and aspects of well being affected by the project, that is, in a comparison of the situations with and without the programme. This ensures that attention is paid to both tangible and intangible costs and benefits. The process of measurement involves the estimation of amounts of resources used and benefits produced by programmes in naturally occurring units. For instance, staffing resources would be measured in terms of time. The final step involves monetary valuation or estimation of quality-adjusted life years/healthy years equivalents where possible.
The principles relating to identification are fivefold: to be aware of the question that is being asked, to keep the principle of opportunity cost in mind when identifying costs and benefits, what this implies in terms of categorizing what is a cost and what is a benefit, how far and how wide the analysis should go, and double counting.
What question is being asked?
As in any research task, it is important to be aware of the exact nature of the question that is being asked before setting out to collect data in detail. Drummond and Mooney (1983) demonstrate this by reference to a particular example. In one study, the answers to a question about the cost of a delivery in a Scottish maternity unit were £540, £510, and £210 (Gray and Steele 1979). The point is that each of the answers is appropriate, depending on the question being asked. To demonstrate, consider the following questions and answers.
What is the current health service unit cost of a delivery in a Scottish maternity unit? Answer: £540.
If we wanted to increase the number of deliveries in Scottish specialist maternity units, assuming that the number of beds is increased, what would be the extra health service unit cost per delivery? Answer: £510.
If we wanted to increase the number of deliveries in Scottish specialist maternity units, assuming that the number of beds is fixed, what would be the extra health service cost per delivery? Answer: £210.
The first cost is simply an average, calculated by dividing the total cost of a unit by the number of deliveries. The other two costs are marginal or incremental, which relate to changes in either the size of maternity units or the number of deliveries that can be undertaken within a unit that remains constant in size. Most often, questions relating to economic evaluation are of the latter type, that is ‘Should we do more or less of this?’ Obviously, it is important to consider marginal costs and benefits. The concept of marginal cost is discussed in more detail below. The important point to remember at this stage is that when the question posed is, ‘What does X cost?’, the reply should be ‘Why do you want to know?’
Identifying opportunity costs
In keeping with the concept of cost introduced above, items to be identified for inclusion in the cost side of an economic evaluation are any resources that have an opportunity cost as a result of being used in the health-care programmes under consideration. This may seem self-evident, but it is not always followed. Instead, analysts sometimes classify any negative effect of a health-care programme as costs and positive effects as benefits. Alchian (1972) has noted this erroneous tendency.
It is the resources used as inputs to health-care interventions that have opportunity costs. Such costs are often measured in terms of money, but what we are really concerned with is the benefits of an intervention compared with the benefits we could obtain if that money could be used in another (the next best) activity. Sometimes cost is used to mean burden, but the two are not synonymous in economic evaluation. This means that, beyond resource use, adverse effects of health care on people’s well being should not be counted as costs.
For example, anxiety is often counted as a cost (even though it is rarely valued). An example of this has been provided by Buhaug et al. (1989). However, anxiety per se does not have an opportunity cost—it is not a resource that could be used in some other beneficial activity. This is not to say that anxiety should be ignored; it has a negative effect on well being and should be counted on the benefit side. Assuming all other effects on well being to be equivalent, if two programmes differ in terms of anxiety and cost, then if one is less costly and incurs less anxiety, it will be more efficient than the other. If one programme is more costly and incurs less anxiety, then a judgement has to be made as to whether the reduction in anxiety is worth the cost incurred. This issue has been the subject of debate in the Medical Journal of Australia (Gerard et al. 1990; Stanford 1990). How to measure such effects on well being is part of the subject matter of the section below on the measurement and valuation of benefits.
Practically, all negative effects on well being resulting from a health-care programme should be netted out on the benefit side in a cost–benefit analysis, the effectiveness side in a cost-effectiveness analysis, and the utility side in a cost–utility analysis. (Strictly, given that cost-effectiveness analysis may use only one-dimensional measures of outcome, this cannot be done if positive and negative effects are in different dimensions.) Likewise, cost savings should be handled as negative costs and netted out on the cost side. In effect, they are savings in resource inputs. This is the only way of ensuring that improvements in well being, on one side, are compared with their resource (or opportunity) costs on the other. Costs will then reflect resource use, which is correct, as it is resources that have opportunity costs.
Categorizing costs and benefits
General guidance as to what costs should be included in an economic evaluation is given in Table 2. All the costs listed relate to resources used. The list is not exhaustive. Neither is it appropriate to include the full range of costs listed in Table 2 in every evaluation. Costs included will depend on the objectives and context of the evaluation. Hence, Table 2 serves as a checklist. Listing costs in this way represents an attempt to ensure that items that are difficult to measure receive as much prominence as easily measurable items.
Table 2 Identification of costs for inclusion in economic evaluations
Health-care resource costs are often classified under four main headings: staffing, consumables, overheads, and capital. The most obvious health-care resource cost is that of professional staff time, which usually forms the largest part of the running costs of any programme. Consumable items are those with a limited life which are used by (or on behalf of) the patient, such as drugs and medical/surgical supplies.
Less obvious costs are overheads, which are often referred to as non-patient-related costs. Overhead costs are those shared by more than one programme. The most common examples are in hospitals where administration, management, heating, lighting, laundry, linen, and cleaning services are provided centrally. The problem is to determine how much of these costs to allocate to the health-care programme being evaluated.
It may be thought that one small programme has no overall effect on such costs, thus presenting an appealing argument for ignoring them. For example, having one less ward is unlikely to change the total financial costs of administration in a large general hospital. However, in the absence of the programme, if people working in centralized departments could have spent their time on some other beneficial activity, there will be an opportunity cost of the programme. This opportunity cost should be valued. Therefore, although the health-care programme sometimes does not add to financial costs, it may still have opportunity costs which should be counted. This demonstrates that accounting costs and economic costs are not the same.
Time-scale is also important here. In the short term, admission of some extra patients may have little impact on staff costs (that is, most costs remain fixed). A fixed cost is a cost that does not vary with activity. In the long run, however, most costs vary with the introduction of the programme. For example, the admission of a large number of patients over time may lead to the employment of extra staff or existing staff working extra hours. What was a fixed cost in one context (that is, the short term) is now a variable cost in another context (that is, the longer term). It is variable opportunity costs (that is, those that change as activity changes) that should be counted in any particular costing context.
Capital items include land and buildings as well as major items of equipment. It may be thought that these items have already been paid for and so should not be included as a cost. This argument is true only when such capital has no alternative beneficial uses (that is, no opportunity cost). If there is an alternative use, there is an opportunity cost which should be valued in some way. Likewise, even if equipment has no obvious and immediate alternative use, perhaps it could be sold and the proceeds used in another beneficial activity. Again, this opportunity cost should be counted. In addition, time-scale is important here. The opportunity costs of many capital items may be fixed in the short run but variable in the longer term.
The cost of other related services includes the staffing, supplies, capital, equipment, and overhead costs associated with community, ambulance, and voluntary services. Clients and families too are often used to provide care (for example, informal care at home) and also incur out-of-pocket expenses, such as transport costs. As described above, the principle of opportunity cost should be invoked when deciding on which of these to cost and to what extent. None of these resources are necessarily ‘free’ in that their use is likely to incur an opportunity cost. If so, they should be counted in an economic evaluation.
All these costs are usually called the direct costs of a health-care programme (Drummond et al. 1997). Indirect costs of a programme are secondary costs related to paid and unpaid productive activities. Productive activities are those arising from participation in the labour force and from housework. Indirect costs arise because treatment could require confinement to hospital or one’s home. This can result in a temporary loss to the community from reduced productive activity. The danger in including such costs is that they may have occurred anyway because of the patient’s illness, and so the actual treatment itself does not add to opportunity costs. Thus it is the cost of inputs to treatment that are relevant and not the cost of the patient’s illness. The benefit of earlier return to work is a welfare (or well-being) gain to society, which, as already established, should be dealt with on the benefit and not the cost side of an analysis.
Benefits for inclusion in economic evaluations are listed in Table 3. This follows directly from the discussion of identifying opportunity costs, including negative as well as positive health effects and negative as well as positive effects in non-health-related well being. The inclusion of production gains is more controversial, not only because of problems with the human capital approach (see below), but also because it is unclear whether including such gains in addition to other benefits (such as quality-adjusted life years, healthy years equivalents, and willingness to pay) involves double counting, that is respondents may have already considered effects on productive activity when giving quality-adjusted life year, healthy years equivalent, or willingness to pay values (Gold et al. 1996).
Table 3 Identification of benefits for inclusion in economic evaluations
How far and how wide should we go?
There comes a time when we have to evaluate our own activities, that is, evaluate the evaluation. Part of this process involves the decision about how extra pieces of information will affect the evaluation. Some costs may not be included in an evaluation, particularly those that are small and difficult to collect. The decision about which costs to include may also depend on the viewpoint of the policy-makers. Often health-care providers and funders are interested only in costs to the health-care system. In other situations, such as evaluation of community care options, it is obviously relevant to examine costs incurred by informal carers or caregivers, such as family and friends. From an economic perspective, the study would ideally take a societal perspective, including as broad a range of costs as possible—although, as pointed out above, once the study goes beyond considerations of health-care (or certainly, public sector) resources and health benefits, it will usually take the form of a cost–benefit analysis.
Another common problem here is how far in the future to look. This is a particular problem for preventive programmes. For instance, breast cancer screening may detect breast cancer at an early stage and therefore lengthen life compared with no screening; however, people who survive may develop other conditions in the future. Should the future costs and benefits of treating these (unrelated) conditions be included?
One argument for not including such costs and benefits is that the decision about whether or not to treat a future condition may be separate from the one being taken now and should be based on the costs and benefits of alternative ways of treating the future disease. If the future costs and benefits of treating heart failure make breast cancer screening less efficient than not screening, should we no longer offer screening? With the hope of living longer, people may say ‘Give me screening but don’t treat me if I get heart failure in the future’, a perfectly rational and reasonable response, and one that would involve an efficient use of resources.
However, analysts should be careful in evaluating preventive programmes, since life tables are often used to estimate survival owing to the lack of ability to follow-up patients over a number of years. These tables permit approximation of survival given certain characteristics, such as age and medical history, and take into account the possibility of such future adverse events as heart failure and the effects (in terms of life years) of treatment for these adverse events. In principle, the costs of treating future conditions should be included in such cases (Meltzer 1997). In practice, however, there are no cost equivalents to data on life tables. This often makes the estimation of future costs difficult. Despite this, discounting (see the section below on the measurement and valuation of costs) of such future costs to present values often reduces their significance dramatically (Bush et al. 1973). Therefore, the best approximation may be to use life tables for survival and to ignore future costs of treating other conditions. As a word of caution, the nearer in time such future costs arise and the more directly related they are to the preventive activity evaluated, the more important it becomes to make some attempt to estimate their magnitude (Meltzer 1997).
As well as the problems on the benefit side referred to above, one should be careful to avoid counting the same cost twice. For instance, assume that the aim of a hypothetical economic evaluation is to compare the costs of general practitioners/family doctors versus hospital doctors for minor surgery. For each of these, it may seem appropriate to count the fee paid to doctors and cost the amount of time taken to carry out the procedure separately and then add the two together to obtain a total cost. However, to count costs in such a way would constitute an example of double counting, as the fee paid to doctors already accounts for time spent on their activities.
Measurement and valuation of costs
After identification comes measurement. In costing, this involves measuring resource used in naturally occurring units. Thus, from Table 4, staffing costs will often be measured in units of time, whilst other items of resource use are measured in various other units (for example grams of drugs and number of diagnostic tests ordered).
Table 4 Measurement and valuation of costs
The cost of any item used in a health-care programme is made up of the unit cost of the item multiplied by the quantity used. The stage of measurement is important as it involves explicit registration of quantities of items used by programmes being evaluated. Quantities of items used are often more relevant to readers of a report than are expressions of the product of price and quantity, particularly if the reader is from another country or another region of the same country. This is because price and quantity data may vary within and between countries. Costings in which the price and quantity of each item used are expressed separately as well as after they have been multiplied together are more useful. Both prices and quantities can then be adjusted to suit other situations.
Many elements of valuing health-care costs are straightforward. For instance, we have already said that staffing costs represent the largest component of health-care costs. Referring to Table 4 again, it is usual to value such costs using wage rates or salaries plus other labour costs. However, although such a procedure may seem straightforward, it is not always advisable simply to accept such monetary figures at face value (see the subsection below on unthinking acceptance of market values). Despite this, market prices are generally accepted as first approximations to the unit cost of most other items within consumable, overhead, and capital categories. Community services, ambulance services, and family/patient expenses should be costed in the same way as health service resources. It should be noted that using patients’ and families’ leisure time also has an opportunity cost. This is difficult to measure. Transport studies have estimated the value of leisure time to be 25 per cent of the local average gross wage rate per hour (Harrison 1974), although a more recent study, specific to health care, has estimated different weights for different types of unpaid activities forgone (Torgerson et al. 1994).
In finding a monetary value (never mind how good a value) to place on a resource, the greatest problems arise in areas of voluntary care and time lost from housework. No readily available market values exist for such occupations and therefore in most cases a value is imputed from an analogous market. For instance, the wage rate for auxiliary nursing staff has been used to cost inputs by volunteers into respite services for mentally handicapped adults (Gerard 1990). Sometimes it is difficult to find truly analogous markets. For instance, it is difficult to cost housework because of its irregular and long hours, which make it untypical of other occupations. In many cases, average female labour costs may be a more accurate reflection of the opportunity cost of housework.
Despite appearing relatively straightforward, there are a number of principles and pitfalls to take into account in the measurement and valuation of health-care costs. These are addressed in the following subsections.
Counting costs in a base year
An obvious but important point about valuation of health-care costs is that they should be counted in a base year, that is, adjusted so as to eliminate the effects of inflation. This should be done because ‘real’ resource use is what is to be measured. If the general inflation rate is running at 5 per cent per annum, in a year’s time £105 will purchase the same amount of resources as £100 now. The costs of £100 now and £105 occurring in a year’s time are equivalent in real terms (that is, in terms of the real amounts of resources that they can fund), although their ‘nominal’ monetary values do not show this to be the case. This is a particular problem when costs of alternative programmes are spread in different proportions across different years, as in the hypothetical example in Table 5. Here, surgical and drug therapy are alternative treatments for a hypothetical condition. Assume that each has the same effects but different cost streams, with the inflation rate at 5 per cent per annum. Thus, a cost of £1050 occurring in a year’s time is equivalent to £1000 now (that is, £1050/1.05), and £1102.5 occurring in two years’ time is also equivalent to £1000 now (£1102.5/1.052). Use of unadjusted costs would result in the conclusion that surgery is more efficient (that is, less costly and equally effective) than drug therapy. However, the costs of drug therapy appear to be higher only because of inflation. After adjusting costs for the rate of inflation (by adjusting costs to year 0 prices), the two therapies are shown to be equally efficient.
Table 5 Adjusting costs to base year
As stated in the previous subsection, not all costs and benefits of health-care programmes occur at the same point in time. The most obvious example is in prevention, where costs are incurred early for the achievement of health benefit later.
The question is, should costs (and benefits) occurring at different points in time be given equal weighting? Most economists would say that they should not. This is because individuals in a society display a tendency to prefer to put off costs to the future rather than pay them now. This ‘time preference’ arises because of the opportunity cost that would arise by allocating funds to paying costs now rather than later and not having those funds available to pursue some other beneficial activity in the meantime. Thus discounting is simply an opportunity cost concept applied over time. This myopia partly accounts for the existence of interest rates (that is, these rates partly reflect the opportunity cost of not being able to use the resources in an alternative way in the meantime).
Other common reasons to explain discounting are diminishing marginal utility of wealth and health. Diminishing marginal utility of wealth refers to the principle that as societies become wealthier over time, the value of an extra £1 in the future is less than the value of an extra £1 now. Likewise, as health improves over time, the value of an extra unit of health improvement in the future is less than that now. At the margin, money and health are worth less over time and therefore future gains and losses in money and health should be discounted.
In summary, a cost arising in the future impinges less than an equivalent cost arising now. As a result, future costs should be discounted (that is, given less weighting) in order to reflect this. Similarly, people prefer to have benefits sooner rather than later, and so future benefits should also be discounted.
Extending the example given in Table 5 to Table 6, those costs occurring now (that is, in year 0) are not discounted. Those occurring in years 1 and 2 are discounted. In this example, year 1 and year 2 costs have been discounted back to present values by multiplying the original cost in each year (that is, £1000) by a discount factor. Usually, discount factors and present values do not have to be calculated as they are often already available in tables, such as that in the Appendix below which provides a list of discount factors and present values annually from 0 to 50 years at a discount rate of 5 per cent. Some government publications and textbooks contain tables of discount factors at various discount rates (HM Treasury 1982; Drummond et al. 1997). The calculation of the discounted costs of our hypothetical drug treatment in years 1 and 2 is explained in the Appendix. From Table 6 it can be seen that, after adjusting for inflation and discounting, drug therapy is now less costly (and therefore more efficient) than surgery.
Table 6 Hypothetical example of discounting
In reality, society’s exact discount rate is not known. It is difficult to observe a rate. Should it be the rate on long-term government bonds or based on some other interest rates (or average of rates) prevailing in the economy? Drummond et al. (1997) recommend a choice of discount rate that is consistent with the following features: it should be consistent with economic theory (between 2 and 10 per cent), include government recommended rates (from 5 to 10 per cent), and be consistent with other published studies and with current practice (from 3 to 10 per cent, but usually 5 per cent). The recommended rates are different in different countries. Given such variation in recommended rates, it is best to test the sensitivity of one’s results to variations in the discount rate.
Marginal or incremental costing
Often decisions relating to health-care programmes are not about whether to introduce a programme but, rather, whether to have slightly more or slightly less of a programme. For example, the question may be ‘Should we change the screening interval for mammography from 3 years to 1 year?’ rather than ‘Should we have a mammography screening programme at all?’
It follows that it is important to calculate the marginal rather than the average costs of a programme. The marginal cost is the cost incurred or saved from producing one unit more or one unit less of a programme, whereas average cost represents the total cost of the programme divided by total units produced up to the point at which the calculation is made. There is no a priori reason to assume that both costs will be equivalent unless total costs rise at a constant rate as the programme is expanded. For instance, although it may cost £25 000 per annum on average to care for an elderly person in hospital, it is unlikely that this amount would be saved if one person less were admitted or that £25 000 would be added to total annual costs if one person more were admitted. This is because certain costs, such as heating, lighting, and some staffing costs, remain fixed and will not vary with small changes in patient load.
To illustrate the principles of opportunity cost and the margin within the context of screening, consider the alternatives for a programme aimed at reducing heart disease in a group of 200 000 men aged 40 to 49 years. One alternative is to promote healthy eating in the population in combination with general practitioner screening for high cholesterol levels (serum concentration 6.0 to 7.9 mmol/l), followed by dietary treatment for those in the relevant range; we shall refer to this as the combined approach. An alternative is to use population promotion of healthy eating on its own; we shall refer to this as the population approach.
It is assumed that health gain can be measured adequately in terms of healthy years. Table 7 shows that the total cost of the combined approach has been estimated at about £40 038 000 with a total benefit of 4200 healthy years, at an average cost of £9530 for each healthy year gained. Whether this represents a reasonable investment is open to question. The question is put beyond doubt, however, when the marginal costs are examined. The second row of data shows that the population approach alone would have yielded 3800 healthy years anyway, at a cost of £38 000 or £10 for each healthy year gained. The additional 400 healthy years are gained by the combined approach at an additional cost of £40 million, a marginal (or, more accurately, incremental) cost of £100 000 for each healthy year gained. The marginal cost for each healthy year gained by the combined programme is over 10 times its average cost (thus highlighting the danger of looking at simple averages and hence the importance of marginal analysis). Reorganization of health-care resources to permit the addition of screening and dietary treatment would presumably be judged as not worthwhile. The opportunity cost is too great or, in plainer language, the resources could be better spent on some other health-producing activity, although, strictly, it is policy-makers, not economists, who should make the judgement.
Table 7 Costs and benefits of strategies to reduce cholesterol
Patient-based versus per diem costs
Ideally, when comparing two groups of patients consuming alternative types of health care, one would like to follow through each individual’s consumption of health-care and other resources. Each individual’s consumption of resources could then be separately valued and a cost per patient calculated.
Valuing opportunity costs is straightforward in some situations (for example, when nursing is done on a one-to-one basis). Such costing is difficult in other situations because of the problem of joint costing. This problem arises when several patients consume a common resource simultaneously and no rule exists as to how to divide the cost between each individual. For example, a nurse may be providing direct care to one patient on a hospital ward whilst supervising several others. In such circumstances, the cost of the nurse’s time cannot be allocated other than arbitrarily.
As a last resort, in such cases it is easier to calculate the per diem cost, or cost per bed day, of providing a particular service and multiply this by the length of stay or lengths of use by the patients concerned. For instance, the cost per bed day per annum of a hospital would be calculated by dividing the total cost per annum of running the hospital by the number of bed days per annum used in the hospital.
The problem with using hospital average costs per bed day is that they may not reflect actual resource use by the client group in question if calculated at too general a level. A cost per bed day that is calculated on the basis of that hospital’s entire caseload is unlikely to be representative of the cost per bed day for specific conditions; a day in intensive care would be credited with the same cost as a day spent in a postnatal ward. Therefore it is advisable to isolate costs per bed day for the specialties that are of interest. This procedure still requires considerable research effort. For non-hospital or non-institutional settings, such as home nursing or outpatient clinics, similar techniques can be used to calculate a cost per visit. For each study patient, the number of visits would then be multiplied by the cost per visit.
Overhead costs are costs shared by more than one programme but, unlike joint costs, can be divided amongst programmes in an apparently sensible way. The most common example arises in a hospital environment where services such as administration, management, heating, lighting, laundry, linen, and cleaning services are provided centrally. The problem for the analyst is to determine how much of the costs of these departments to allocate to the programme being evaluated. Whether one decides to allocate such costs at all will depend on the scale of incremental change considered in the analysis (see the subsection above on marginal or incremental costing).
The solution to this problem is to ensure that overhead costs are allocated to programmes within the hospital on a reasonable basis, that is, they should represent differences between overhead opportunity costs with and without the existence of the patients concerned. Examples of simple methods of allocation are listed in Table 8. Sometimes these will suffice, although details of more sophisticated overhead allocation mechanisms can be found elsewhere (Drummond et al. 1997).
Table 8 Simple methods of allocating overhead costs
The costs of capital assets, such as land, buildings, and equipment, usually arise at a single point in time, usually at the inception of the programme being evaluated. Despite this, such capital assets are used over time, and at any point in time an asset may be resold for an amount that will be less than its initial cost and that will decrease as time goes by (except for the special case of property in ‘boom’ periods, which makes no difference to the general approach to dealing with capital costs). Thus, despite an initial one-off outlay, the opportunity costs of capital assets are spread over time. This is accounted for by spreading the opportunity cost of capital assets over the number of years of their life judged relevant in each particular circumstance.
The most common method of doing this is to calculate an ‘equivalent annual cost’. Using this method the initial outlay on a capital asset is converted to an annual sum which, when paid over a number of years (usually an estimated or known life of the asset), would add up to the initial value of the capital asset plus the opportunity cost of resources tied up with the asset (because, for example, they could have earned a certain rate of interest if invested). This principle is best understood by relating it to the principle of a mortgage on a house, whereby the cost of the house plus interest charges are reflected in a series of annual (or monthly) equivalent mortgage repayments. As with discount factors and present values, equivalent annual costs do not have to be separately calculated as they are readily available in tables, such as that in the Appendix to this chapter. This table displays the equivalent annual costs of £1 discounted at a rate of 5 per cent for payback periods of 0 to 50 years. An example of a conversion of a capital cost to an equivalent annual cost is also provided in the Appendix.
One further problem with costing is in determining a cost for some capital items which can then be converted to an equivalent annual cost. This is not a problem for items for which actual (and recent) purchase prices are available. Problems arise with land and buildings that already exist at the start of the programme and may have been paid for many years before the programme commenced. Such items should be costed if their use by the health-care programme prevents their use in some other beneficial activity. The most common ways of costing such items are to obtain an estimate of the price that the item would attract if made available on the open market, to obtain an estimated cost of replacing the item, or to obtain an estimate of the rental or lease value of the item.
Unthinking acceptance of market values
Economists are often criticized for using market values as a reflection of true opportunity cost. However, such values are often used only as baselines, and adjustments can be made as required. For instance, the readiness of economists to impute values for the unpaid labour of volunteers and home-makers demonstrates that market values are not always accepted at face value (Donaldson et al. 1986; Donaldson and Gregson 1989). Market values would clearly result in the assignment of zero costs to inputs to care by volunteers and home-makers, despite the fact that these resources have opportunity costs.
One should also be wary of values arising for resources for which markets do exist. For instance, owing to rigidities in modern markets the cost of a doctor is greater than the cost of a nurse. Therefore, other things being equal, a doctor-oriented health-care programme will appear to be more costly than a nurse-oriented alternative. However, if a shortage of nurses and a glut of doctors existed, it may be that investment in the nurse-oriented programme would impose a greater opportunity cost than investment in the doctor-oriented programme. This is because the nurse-oriented programme may attract nurses from other health-care specialties, potentially sacrificing benefits to other patients. Alternatively, owing to the glut of doctors, the doctor-oriented programme may involve little opportunity cost to patients in other specialties if doctors can be recruited from the dole queue.
Sometimes it is necessary to attempt to work back to what a market value would have been in the absence of some imposed distortion. For instance, taxes imposed on certain commodities bear no relation to their opportunity cost and often distort the true costs of such commodities. These examples demonstrate that, at first glance, market values should be used only as indicators of cost and that analysts should realize that they may not represent true opportunity costs.
Inevitably, a costing exercise will be subject to some degree of uncertainty. Where one has not been able to estimate costs accurately, some assumptions may have been made about possible values that such costs could take. Sensitivity analysis is useful in such situations as it involves testing the sensitivity of the results (or conclusions) of an evaluation to variations in variables about which one is uncertain. Readers of a report or article can then judge which assumptions, and thus results, are more appropriate to their local situation. A sensitivity analysis can also indicate where further research is needed to obtain a more accurate estimate of a variable which is critical to the end result.
As an example of sensitivity analysis, Hundley et al. (1995) examined the extra cost of introducing a midwives’ unit rather than a consultant-led labour ward in intrapartum care. The baseline extra cost of the midwives’ unit was estimated to be £40.71, which would add about 10 per cent to the total costs of delivery. However, there were uncertainties over capital costs and midwife staff costs. For the former, it was unclear whether the conversion of rooms for the midwives’ unit would have occurred anyway. Deducting this from the baseline results in an extra cost of £36.89. Similarly, some geographical areas would not require the upgrading of midwives which was required in the study location. This would result in an extra cost of £25.01. Combining both these assumptions resulted in an extra cost of £21.19 for the midwives’ unit. The midwives’ unit would break even only if no extra staff were required as well as no upgrading of existing staff, an unlikely scenario.
The most probable candidates for sensitivity analysis in a costing exercise are production effects, items that have been excluded because of difficulty of collecting data, imputed values, the discount rate, and the lengths of life for capital items. In some cases assumptions about the possible values that a variable can take are arbitrary. On other occasions it may be possible to base a sensitivity analysis on the confidence limits of a statistical estimate of a variable. The effect of such extreme values on results could be tested using upper and lower confidence limits as extreme values. For example, in studying the effects of the use of prophylactic antibiotics on wound infections in Caesarean sections, Mugford et al. (1989) estimated costs based on assumptions that the odds of infection would be reduced by either 70 or 50 per cent (the approximate limits of the 95 per cent confidence interval of the odds ratio).
These principles also apply to estimates of benefits. The analyst may be unsure about estimates of life years gained or quality of life. If so, the sensitivity of results to variation in such estimates should be tested. The methods of estimating such benefits will now be examined.
Measurement and valuation of benefits
Associated with each evaluative technique are particular methods for measuring and valuing benefits. Each of these methods is summarized in Table 9. The crudest measures of benefit are those used in cost-effectiveness analysis, where benefits are measured in terms of one-dimensional natural units. In cost–utility analysis, benefits are measured not in physical units, but in terms of healthy years. Healthy years are represented by a single utility-based health index, which incorporates effects on both quantity and quality of life. The most common measure is the quality-adjusted life year, of which there are two distinct forms—generic and condition-specific. As the names suggest, the former are applicable over a large range of interventions and conditions, while the latter are applied to specific conditions. In recent years an alternative to quality-adjusted life years, the healthy years equivalent, has been developed (Mehrez and Gafni 1989). The most comprehensive (but also the most technically difficult) form of benefit measurement occurs in cost–benefit analysis, where benefits are measured in monetary terms. The two principal methods for doing this are the human capital approach and willingness to pay.
Table 9 Identification and measurement of benefits in economic evaluation
Measures used in cost-effectiveness analysis
As outlined above, cost-effectiveness analysis is the simplest form of economic evaluation in health care. In its simplest form, in which costs only are compared, it is necessary to know that the outcomes of alternatives are equivalent or that the less costly alternative is no less beneficial. In such situations it does not matter if outcomes are one-dimensional or multidimensional. For example, in a study comparing different methods of providing long-term care for elderly people, no differences in survival and activities of daily living were found between the alternatives (Bond et al. 1989). In view of this, a subsequent study focusing on cost-effectiveness was able to concentrate on cost differences only (Donaldson and Bond 1991).
However, in situations in which cost-effectiveness ratios are used, the outcome (or benefit) is always one-dimensional. The appropriate measure depends upon the programme being compared. For programmes whose major effect is to extend life, life years gained would generally be used. In contrast, if the major effect of the programmes is to improve quality of life rather than quantity of life, then some other measure would be more appropriate. For example, in comparing programmes for the prevention of coronary heart disease, unit reductions in serum cholesterol might be an appropriate measure. Similarly, if two antenatal screening programmes are being compared, cases detected might be chosen. Concentration on such narrow measures of benefit, however, may mean that other important benefits are overlooked. Consider an example in which dialysis is compared with transplantation for the treatment of chronic renal failure. In general, quality of life following a successful transplant will be higher than quality of life associated with dialysis (Klarman et al. 1968; Doessel 1978). However, if life years gained is chosen as the measure of outcome, these effects on quality of life cannot be incorporated explicitly into the evaluation.
Generic quality-adjusted life years
Generic quality-adjusted life years are typically derived from multi-attribute utility scales, which are instruments for estimating health state values. The five most widely used multi-attribute utility scales are the quality of well being scale (Kaplan et al. 1976), Rosser’s disability/distress classification (Rosser and Kind 1978), the 15-dimensional measure of health-related quality of life (Sintonen and Pekurinen 1993), the health utility index versions 1, 2, and 3 (Torrance et al. 1995), and the Euroqol EQ-5D (Euroqol Group 1990). Of these, perhaps the most widely used is the EQ-5D which was developed by a multidisciplinary group of researchers in five different countries (Euroqol Group 1990). The five dimensions of the EQ-5D are mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Each dimension has three levels, thus enabling the EQ-5D to define 243 different health states (see Table 10 for the dimensions and their levels). The inclusion of two extra health states, unconscious and dead, means that there are 245 possible health state descriptions.
Table 10 Dimensions of the Euroqol EQ-5D
The EQ-5D is designed to be administered in the form of a self-completed questionnaire. Subjects are asked to define their own health state in terms of the five dimensions and their levels and then asked to mark on a visual analogue scale how good or bad they think their current health is. The visual analogue scale is presented as a vertical line on a page divided into 100 equal intervals and with the endpoints marked ‘worst imaginable health state’ and ‘best imaginable health state’. The visual analogue scale values are interpreted as the health state weights.
A number of commentators have argued that visual analogue scale values are a poor indicator of individuals’ strength of preference (Loomes and McKenzie 1989; Nord 1991). In view of this criticism the developers of the Euroqol employed the time trade-off technique (see next subsection) to generate a tariff of quality of life weights for each EQ-5D health state (Dolan 1997). The tariff was based on health state values elicited in a large-scale survey undertaken in the United Kingdom by the Measurement and Valuation of Health Group at the University of York (Measurement and Valuation of Health Group 1994). The tariff allows each of the EQ-5D health states (identifiable by a unique five-digit number corresponding to the levels for each of the five dimensions) to be converted into a score which can be used as the quality adjustment weight in the calculation of quality-adjusted life years. The best health state (that is, 11111) is assigned a value of 1.0 and death is assigned a value of 0. The remaining health states are assigned positive or negative scores depending upon whether they are valued as being better or worse than death respectively. It should be noted that different tariffs are available for different socio-economic groups.
In order to illustrate the calculation of quality-adjusted life years using an EQ-5D tariff, consider the following simple example of an individual with varicose veins facing two alternative treatments—compression hosiery or sclerotherapy—and whose remaining life expectancy is 10 years (note that neither treatment affects life expectancy). Table 11 presents the EQ-5D health states that the individual can expect to experience for the rest of their life under each treatment regimen. For compression hosiery, the total expected quality-adjusted life years is 5.65, while sclerotherapy is expected to yield 6.54 quality-adjusted life years. The expected quality-adjusted life year gain from sclerotherapy compared with compression hosiery is simply 6.54 – 5.65 = 0.89 quality-adjusted life years. If the cost of sclerotherapy over and above the cost of compression hosiery is, for instance, £2000, then the cost per quality-adjusted life year gained from sclerotherapy is £2000/0.89 = £2247. It should be noted that the above example does not include any adjustment for the differential timing of the benefits from each treatment. If account is to be taken of this then each of the tariff weights should be adjusted by an appropriate discount factor (see section on discounting above).
Table 11 Calculation of quality-adjusted life years
The Euroqol EQ-5D has been criticized for being too simplistic and insensitive to changes in health status (Gafni and Birch 1993). However, a recent review of the literature pertaining to the five main multi-attribute utility scales concluded that on the basis of a number of factors, such as practicality, reliability, and various dimensions of validity, the EQ-5D and the health utility index should be the scales of choice (Brazier et al. 1999).
Condition-specific quality-adjusted life years and healthy years equivalents
The main difference between condition-specific and generic quality-adjusted life years lies in the way in which health states are described to subjects. With condition-specific quality-adjusted life years the health-state descriptions focus on the characteristics of the condition being evaluated. This is in contrast with generic quality-adjusted life years where the health-state descriptions are more general.
There are two main methods of generating condition-specific quality-adjusted life years: standard gamble and time trade-off. The standard gamble is based directly on the axioms of standard utility theory; it is the classic method of measuring preferences under uncertainty. The technique can be used to measure health-state preferences for chronic and temporary health states. The discussion here focuses on the use of the technique to calculate quality-adjusted life years for a chronic health state preferred to death. For details of how the standard gamble can be applied to temporary health states and health states not preferred to death see, for example, Froberg and Kane (1989) or Torrance (1986).
An example of the standard gamble framework for a chronic health state preferred to death is shown in Fig. 2. To measure preferences for health state i, subjects are asked to choose between two alternatives. One offers the certain outcome of remaining in the chronic health state for the rest of one’s life, whilst the other is a gamble representing a treatment with two possible outcomes. These outcomes are (a) return to full health for the rest of one’s life (with an associated probability P of occurring), and (b) immediate death (which has a probability of occurrence of 1 – P). The probability P of a successful outcome is varied by an iterative process until the subject is indifferent (that is, cannot choose) between the gamble and the certainty. The probability P* at which the subject is indifferent is used to calculate the utility value of the health state as follows. The outcomes ‘return to full health’ and ‘death’ are assigned utility values of unity and zero respectively. At the indifference probability P, the value (measured in terms of expected utility) of the two alternatives is equal, yielding the following simple equation:
U (state i) = P*U (full health) + (1 – P*) U (death).
Because the utility from full health equals 1.0 and the utility from death equals 0.0, this equation can be reduced to
U (state i) = P*.
Therefore, for the case of a chronic health state preferred to death, the utility of the health state is simply P (the indifference probability). The utility of the health state can then be used to calculate the number of quality-adjusted life years from the treatment in a similar way to that described for generic quality-adjusted life years.
Fig. 2 Standard gamble for a chronic health state preferred to death.
Time trade-off was developed by Torrance et al. (1972) as a substitute for the standard gamble technique. The intention was to develop a technique specifically for use in health care, which gave the same (or similar) values as the standard gamble but which was easier for subjects to understand. Two important differences between time trade-off and the standard gamble should be noted. Firstly, time trade-off does not have an axiomatic foundation; secondly, subjects are asked to choose between two certain alternatives rather than between a certain outcome and a gamble. Like the standard gamble, time trade-off can be used to elicit health-state preferences for chronic and temporary health states that may or may not be preferred to death. An example of a time trade-off framework for a chronic health state preferred to death is shown in Fig. 3. Preferences for health state i are established by eliciting from subjects the number of years in full health (Z years) that is equivalent to spending the rest of their life (T years) in the chronic health state. The value Z* at which the subject is indifferent between the two alternatives is used to calculate the value of the health state as follows. Full health is assigned a value of unity and death a value of zero. It is assumed that individuals have a value function of the form V = h1T, where h is the preference value for health state i (0 < h1, < 1) and T is remaining years of life.
A value function assigns a real number to each possible outcome in a choice problem under certainty. Utility functions are appropriate for uncertain choices.
At Z*, the value functions of the two alternatives are equal, yielding the following equation:
V (alternative 1) = V (alternative 2), h1T = 1.0 × Z*.
Therefore, the preference value for the chronic health state i is h1 = Z*/T. The value of the health state can be used to calculate the number of quality-adjusted life years from the treatment in the same way as described above for generic quality-adjusted life years.
Fig. 3 Time trade-off for a chronic health state preferred to death.
Although the authors have, for purposes of exposition, shown how values for chronic health states are derived, there are examples in the literature, using the same basic principles, of the derivation of values for temporary health states (Johnston et al. 1998).
In recent years an alternative to quality-adjusted life years has been developed—the healthy years equivalent (Mehrez and Gafni 1989). Like quality-adjusted life years, healthy years equivalents combine quantity and quality of life into a single measure. The definition of a healthy years equivalent for a chronic health state is as follows (for a more general definition see Mehrez and Gafni (1989)). Let Q represent the chronic health state of an individual and T the individual’s remaining years of life. Let U(Q, T) be the utility function of the individual, which describes the utility of being in health state Q for T years followed immediately by death. Let Q* represent full health and H* the healthy years equivalent of (Q, T). The problem is defined as follows: find H*’ such that U(Q*, H*) = U(Q, T). The main difference between quality-adjusted life years and healthy years equivalents is that, in calculating the former, the utility weight of a health state is multiplied by the time spent in the health state. The utility weight is based on individual preferences but the duration is not. The algorithm for calculating healthy years equivalents allows individuals to express their preferences for both the health state and the duration.
There is controversy within health economics as to the added value of healthy years equivalents over quality-adjusted life years and the reader is referred to articles by the critics (Buckingham 1993; Culyer and Wagstaff 1993; Johannesson et al. 1993; Loomes 1995) and the advocates (Gafni et al. 1993; Mehrez and Gafni 1993) of healthy years equivalents.
In recent years the most common method of measuring health-care benefits in cost–benefit analysis has been willingness to pay. However, the application of willingness to pay to health care is a relatively recent phenomenon, and for many years the dominant measurement technique was the human capital method.
Under the human capital approach, health improvements are valued in terms of future productive worth to society from individuals being able to return to (paid and unpaid) work following the health improvement. Productive worth tends to be proxied by future earnings potential. This immediately raises questions about how to deal with women, elderly people, and the unemployed (Avorn 1984). Furthermore, if a truly societal perspective is to be taken in an economic evaluation, it may be that, in economies with substantial amounts of unemployment, there is little opportunity cost in losing a worker due to premature retirement or death. As many workers can be replaced, preventing their loss may add little to overall productivity. This has led to controversy over whether or not to include only production losses incurred during the period of replacing a worker in economic evaluations (Koopmanschap et al. 1995; Johannesson and Karlsson 1997).
It is generally acknowledged that willingness to pay is a more theoretically correct measure of benefit in that it captures in one measure notions of human capital as well as other benefits of health care. The principle of willingness to pay is very simple—the utility that an individual gains from something is valued by the maximum amount that he or she would be willing to pay for that something. The technique of willingness to pay is often criticized for attempting to assign a monetary value to things that are considered by many to be incommensurate with monetary valuation (for example, the environment, human life, and so on). However, it has to be remembered that such valuations are being made anyway. These valuations arise from the implicit judgements of decision-makers. What is important is not the unit of value per se, but rather the notion of sacrifice embodied in the technique. In economics, something only has value if an individual is prepared to sacrifice something in order to acquire it. In valuing a health-care programme, it is difficult to ask respondents what services they would give up to have more of that programme. It is easier to ask individuals to state the maximum amount that they would be willing to pay for more of the programme and for some possible alternative uses of those resources. Thus it is largely for convenience of comparison that money is the chosen numeraire.
One advantage of willingness to pay over other measures of benefit discussed so far is that it provides the opportunity for individuals to value other potential benefits of health care beyond just health gain. The value of a health-care intervention can be measured by the maximum amount individuals are prepared to pay for the intervention. Exactly how much individuals are willing to pay will partly depend on the perceived benefits to them of the intervention. One of the assumptions of quality-adjusted life years and healthy years equivalents is that the only benefit from health care is improvement in health-related quality of life. However, there is evidence that this is not always the case (Berwick and Weinstein 1985; Mooney and Lange 1993; Donaldson and Shackley 1997). Other possible sources of benefit might include the provision of information (for example, from screening), dignity (for example, in long-term care), and the process of care (for example, invasive versus non-invasive interventions).
Depending upon the context of the evaluation, there are a number of different ways in which individuals can be asked to imagine they have to pay. These include out-of-pocket payments (Miedzybrodzka et al. 1994), one-off extra taxation payments (Olsen and Donaldson 1998), and payments for insurance (Gafni 1991).
There are also a number of different ways of asking willingness to pay questions. These include payment card questions in which subjects are presented with a series of prompts from which to select a value, open-ended questions in which respondents are asked to state their maximum willingness to pay without prompting, and closed-ended questions in which each respondent is presented with a willingness to pay value and asked to indicate whether or not they would be prepared to pay that value. Each method has its advantages and disadvantages, and there is no overall consensus as to which is the best method (Johannesson and Jonsson 1991).
The willingness to pay technique is not without its problems. A frequent criticism is that it is inevitably a function of ability to pay, which, it is argued, could have implications for equity. There are, however, ways of dealing with this (Donaldson 1999). This is similar to the weighting of quality-adjusted life years to reflect concerns other than health gain, such as severity of illness (Nord et al. 1999).
The use of cost per quality-adjusted life year data in health-care decision-making
So far, this chapter has focused on the measurement and valuation of benefits per se. However, it is also important to consider how the different measures of benefit can be used in practice to aid decision-making. In particular, it is useful to discuss the practical applications of cost per quality-adjusted life year and willingness to pay data.
Cost per quality-adjusted life year ratios have two principal applications. The first is to compare alternative interventions for the same condition. Here, the intervention with the lowest cost per quality-adjusted life year ratio is the most technically efficient. This is simply an extension of cost-effectiveness analysis.
The second (and more problematic) use of cost per quality-adjusted life year ratios is to help judge relative priorities across different programmes in health care. Cost per quality-adjusted life year ratios can be calculated for a wide variety of disparate interventions, ranging from general practice advice to give up smoking, chiropody, and so on, to renal dialysis and heart transplantation. Because the measure of outcome is the same for all interventions, across programme comparisons can be made. It is possible to construct cost per quality-adjusted life year league tables in which interventions are ranked in terms of increasing cost per quality-adjusted life year ratios, that is, decreasing relative efficiency. Indeed, such an exercise has been carried out in the state of Oregon in the United States, although not without problems (Dixon and Welch 1991; Tengs 1996). Quality-adjusted life year league tables have been criticized on a number of grounds, and great care should be taken in their construction and interpretation (Gerard and Mooney 1993; Birch and Gafni 1994). Among the many potential problems of quality-adjusted life year league tables, the following are particularly worthy of note. The most serious problem is that each item in the league table has a different comparator. The cost per quality-adjusted life year gained of programme A may have been produced by comparing programme A with programme B. However, if B is inefficient to begin with, A may also be inefficient.
The use of ratios in league tables also hides the fact that programmes of various sizes are being compared. Thus a small programme C, which is further down the table, could be combined with a programme D, which is higher up, to produce more quality-adjusted life years than a larger programme E in the middle.
With respect to transferability of cost per quality-adjusted life year league tables across locations, on the cost side, different locations will tend to have different cost structures and therefore different marginal costs of expansion and contraction. On the benefit side, the capacity of individuals in one location to benefit from a particular intervention will tend to be different from that of individuals in another location, thus leading to different levels of quality-adjusted life years gained. Because of this potential for variations at the margin between locations, it is likely that a cost per quality-adjusted life year league table used in one location will not be applicable to other locations. Ideally, therefore, if (and it is a big ‘if’) cost per quality-adjusted life year league tables are to be used for priority setting, they should be locally based. The correct application of a crude locally based league table is preferable to the misapplication of a more sophisticated league table from another location.
Using willingness to pay data in health-care decision-making
A recent example of using willingness to pay to aid priority setting was an evaluation of public sector health-care programmes in northern Norway (Olsen and Donaldson 1998). Members of the public were asked their willingness to pay in extra taxation for each of the following: the introduction of a helicopter ambulance service to serve remote communities, an expansion in the number of heart operations performed, and an expansion in the number of hip operations performed. Subjects were presented with detailed descriptions of the programmes and were told that only one could be implemented. They were also asked to state the reasons for their willingness to pay responses.
The use of willingness to pay is not restricted to issues of allocative efficiency—it can also be used to address narrower questions. A recent example of this is the study by Donaldson et al. (1998) in which the relative benefits of two alternative locations for delivery of a baby were evaluated. Using the results from a recently completed randomized trial (Hundley et al. 1994), pregnant women were given descriptions of care in midwife-managed units and in traditional labour wards. They were asked for their preference for one over the other and then were asked about their maximum willingness to pay to have their preferred rather than their less preferred option. The results displayed a clear preference for the midwife-managed unit. Despite the usefulness of such a result, there are limitations as to how these data can be used. The midwife-managed unit was also more costly than the labour ward (Hundley et al. 1995). Thus the more beneficial new form of care (the midwife unit) would also be more costly. This may imply the allocation of more resources to maternity care in order to achieve more benefit.
However, those resources would have to come from some other (presumably beneficial) use. Ideally, values would also be required for this alternative use (as in the case of the northern Norway study described above). In studies which find that total willingness to pay elicited from users of a programme is greater than the programme cost, it may be tempting to conclude that such a programme is worthwhile. However, the opportunity cost context is one where resources for such a programme will have to come from some other use. The decision to invest in the programme is really a community decision. To reiterate, when using willingness to pay values to aid priority setting (particularly in a publicly financed health-care system), the relative values of members of the community are of importance (see the northern Norway example above), and not the absolute values of users. The willingness to pay values of users indicate the strength of preference of the users for the services being evaluated and are still useful in that context. For instance, such values could indicate that the preferences of a minority group are particularly strong. If this strength of preference is not sufficient to outweigh that of the majority, such values may still indicate to the decision-maker that providing both types of care evaluated is the fairest option. However, it should be remembered that providing such choice may come at a substantial cost.
Programme budgeting and marginal analysis
Health-care decision-makers cannot conduct full-scale economic evaluations for every decision which they have to make. However, they are still faced with situations of scarcity of resources, whereby they do not have enough resources to meet all of the claims of those resources. The question then is whether the principles outlined above can at least be used as a framework for a more pragmatic evaluatory process to support health-care priority setting in ‘day-to-day’ contexts. One such framework is programme budgeting and marginal analysis. This economic approach to needs assessment has been used in several care settings in the United Kingdom, Australia, New Zealand, and Canada (Mooney 1984; Donaldson and Farrar 1993; Cohen 1994; Ruta et al. 1996; Peacock and Edwards 1997; Scott et al. 1998; Mitton et al. 1999).
The overall aim of programme budgeting and marginal analysis is to provide assistance to health authority managers in directing resources in a manner in which the impact of health care on the health needs of the local population is maximized (Donaldson and Mooney 1991). In principle, programme budgeting and marginal analysis allows this to happen by addressing both technical and allocative efficiency (Mooney et al. 1994). Furthermore, by permitting the analysis of who receives resources, programme budgeting and marginal analysis also allows issues of equity to be addressed (Mooney et al. 1986; Viney et al. 1995). Its starting point is to examine how resources are currently spent before focusing on marginal health gains and costs of changes in that spend, through comparison within or across programmes of care (Donaldson and Farrar 1993).
The first component of programme budgeting and marginal analysis, programme budgeting, is a means for describing the pattern of spending within health authorities and its distribution between groups in the population (Gold et al. 1997), with emphasis not on inputs and activities, but rather on how these inputs and activities contribute to improving the health status of individuals (Pole 1974; Wiseman et al. 1998). A five-step approach for conducting programme budgeting and marginal analysis has been reported (Donaldson et al. 1995), and can be summarized as in Table 12.
Table 12 Five-step approach to a budgeting and marginal analysis programme
The first step is to identify the programme of interest and determine the total resources available for the programme. The programme budget can classify expenditure by programme (that is, disease group), by service inputs grouped by sector of care (that is, primary care, acute care), or by other means such as population demographics (Pole 1974; Mooney 1977; Miller et al. 1997). The next step is to describe the current pattern of activity and spending for the programmes and subprogrammes, using data available through an information system or by collecting data prospectively. The programme budget allows for comparison of the allocation of resources over time (Mooney et al. 1986) or with the allocation of resources in other regions (Craig et al. 1995; Miller et al. 1997), and also can indicate particular areas for marginal analysis by making the resource allocation process more explicit (Cohen 1995; Twaddle and Walker 1995).
Following this, decisions can be made as to which services are candidates for expansion and which services are candidates for reduction. At this third step, marginal analysis begins, and planning groups are often convened. This panel might consist of doctors, other clinical professionals, health authority managers, patients, and other interest groups (Cohen 1995). In the fourth step, scenarios which involve increases and reductions in spending can be presented to the panel, who can then make priority lists of which services should be expanded or reduced. The key at this point is in focusing on marginal benefits per unit of resource spent, thus alleviating the need to assess total needs and overall benefits. Proposed expansions in services can be funded in one of two ways:
by providing some services at the same level of effectiveness, but at less cost (that is, technical efficiency improvements)
or, if this cannot be done,
by taking resources from some areas of effective care, only if, at the margin, they provide less effectiveness per unit of resource spent than the proposed expansion (that is, improving allocative efficiency).
Thus, at this stage, the proposed options (expansions and contractions) have to be evaluated. Local data can be supplemented with evidence on effectiveness and cost-effectiveness from the literature (Donaldson and Farrar 1993; Cohen 1995). These results can also be used in conjunction with other means, such as needs assessments, review of local and national policy, other consumer/public views, and other health professional views, to determine priorities (Cohen 1995; Craig et al. 1995; Miller et al. 1995; Ruta et al. 1996). Finally, health authority decision-makers have to decide whether resource shifts will actually follow the recommendations of the planning group, and specifically address whether any trade-off with equity results from the potential increases in efficiency (Mooney et al. 1993).
Such planning mechanisms can be used either within programmes of care (for example, maternity care) or across programmes of care. The former is known as ‘micro programme budgeting and marginal analysis’ while the latter is known as ‘macro programme budgeting and marginal analysis’. The technique has also been applied at the level of primary care, where groups of general practitioners are given budgetary responsibility for organizing and purchasing primary, secondary, and tertiary services for their registered population (Scott et al. 1998). Such activity could be known as ‘meso programme budgeting and marginal analysis’.
The major challenge with such a framework is that it is data hungry. While costing data may be available at the local level, it is more difficult to obtain data on the benefits of alternatives of care. Programme budgeting and marginal analysis deals with this real problem by obtaining judgements from expert panels, and supplementing this information with available evidence from the literature. Ultimately, no matter how few data exist, decisions still have to be made. As such, this pragmatic approach to programme budgeting and marginal analysis has been suggested (Scott et al. 1999), using a combination of expert opinion and the literature on the benefit side, and locally available data with estimations where required, supplemented by economic evaluations from the literature, on the cost side. Overall, programme budgeting and marginal analysis can serve to inform complex decisions by highlighting, in an explicit manner, the relationship between the costs and benefits of a particular action. International experience to date suggests that this economic pragmatic approach can be an important part of the priority-setting process.
Micro-economic analysis of public health policy: more theory and some examples
Activity in the use of economics in assessing behavioural changes as a result of public health and health-promotion policies is growing. Before illustrating this with examples from the literature, it is important to outline, in more detail, the ‘canonical model of the consumer’, a cornerstone of micro-economic analysis.
A model of consumer behaviour
This model simply extends the notion of opportunity cost, introduced above. It may appear to be quite technical to non-economists, but it is useful in analysing the behavioural responses of individuals to public health and health promotion interventions. The theory is introduced here in the context of utilization of dental care.
Assume individual A has Y dollars to spend on visits to the dentist, V, and all other goods, G, measured in dollars. The cost of purchasing other goods and the cost of dental visits cannot then exceed Y, so we can express the set of options available to the individual as Y ³ pV + G, where p is the price of a visit to the dentist. This formula is called a budget constraint; it gives the choices available as a function of the prices and incomes of the decision-maker. If the person is rational in the earlier economic sense, we also know their behaviour can be analysed as prices and incomes change by examining the properties of the constrained optimization programme:
max D,G U(D, G)
pD + G ³ Y,
where U ( . . . ) is the utility function representing the individual’s tastes over dentist visits and other goods.
Figure 4 displays a graphical representation of the person’s utility maximization problem. The budget constraint in (G, V) space is the straight line in the diagram. Where it meets the V axis, the consumer is spending all of his or her money on dental visits, and, where it meets the G axis, all money is spent on other goods. Moving away from the origin, the axes represent increasing amounts of visits and other goods. Points above the budget constraint cost more than Y, points below the constraint cost less than Y, and points on the constraint cost exactly Y. The most preferred combination of visits and other goods, V* and G* respectively, can be found by finding the (V, G) pair which is both affordable (does not cost more than Y) and associated with the highest level of utility, denoted by the ‘indifference curves’ U in Fig. 4. The indifference curve is this shape because it is assumed that, as more and more dental visits are consumed, they are worth less and less to the consumer, and therefore he or she would give up less of other goods for each increment in dental visits. This is known as diminishing marginal utility. Within (G, V) space there is an infinite number of indifference curves. These curves are not allowed to cross each other, otherwise the analysis would not make any sense.
Fig. 4 Indifference curves and budget constraints.
Those to the northeast of U would represent higher levels of utility whilst those to the southwest would represent lower levels of utility. The preferred point is denoted E in Fig. 4; any other point is either not affordable or puts the consumer on a lower level curve, which would be a less preferred combination.
The solutions to this problem reveal the relationship between other goods and dental visits as functions of the parameters of price and income. The function giving dental visits as a function of price, conditional on income, D(p, Y), is called the demand curve for dental visits. Notice that the opportunity cost of one more dental visit is p units of other goods, and that the cost of one more unit of other goods is 1/p units of dental visits. For example, if the price of a dental visit is £50, then one visit costs £50 worth of other goods, and £1 spent on other goods has an opportunity cost of 1/50 of a dental visit.
Generally, it is possible to determine the opportunity cost of a marginal change in one of the choice variables by computing the derivative of the constraint at the current choice; since the constraint here is linear, opportunity costs are the same regardless of where on the budget constraint they are calculated.
More generally, constraints need not be financial, and the constraint need not be linear. Consider a simple model of social choice over how much health care to provide. Suppose society can allocate its resources to produce either health care, HC, or food, F. The results generalize easily to an arbitrary number of goods. Producing more health care comes at the cost of producing less food, and as more and more health care is produced each successive unit treats fewer and fewer serious cases, since people will tend to ‘pick the lowest hanging apples first’ (that is, recalling the notion of diminishing marginal utility, it is assumed that the patients who need treatment the most will get it first). Hence, the opportunity cost of producing an extra unit of health care increases as we produce more and more health care, so the constraint can only be written as some non-linear function. Imagine that social preferences can be treated like individual preferences and can be summarized by a utility function U(HC, F), it is possible to compute the socially preferred production of health care and food related to a given constraint. Although Arrow’s famous Impossibility Theorem implies such an exercise, it is at best dubious. (For a detailed explanation see Mas-Colell et al. (1995)).
It would then be possible to examine how changes in the economy which move the constraint about would affect optimal provision of health care; for instance, a technological improvement might make more of both food and health care available.
The analytical underpinnings of many models in health economics are based on Grossman’s (1972a,b) research, which in turn extended earlier work on the allocation of time within households by Becker (1965). The application of this model in empirical studies has paralleled similar efforts in epidemiology. Grossman’s model combines the simple analyses of consumers and of firms discussed above: individuals are assumed to have preferences over their own state of health and other desired outcomes (such as, for instance, spending time with their children or reading books). Grossman places individuals with such preferences in a dynamic world in which they can sell their time in labour markets, use both time and money to ‘produce’ both health and other outcomes with various combinations of time and money, and are subject to the biological process of ageing. Notice that individuals are explicitly assumed to be willing to trade off their health status for other goals; the question is not whether they do so, but how changes in incentives change how much health they trade off for other life goals.
The model yields a rich set of interesting results. In Grossman’s framework, for instance, time of death is determined by the individual’s choices; a given individual who chooses to invest more time and money to keep themselves healthy will tend to live longer, and the choice to invest more in health is itself determined by the person’s preferences and resources available. The model predicts that rising labour market incomes will yield greater health investment and thus lower morbidity and mortality at any age. This view, that it is an individual’s own behaviour that leads to observed health outcomes, and that that behaviour is affected by factors such as labour market opportunities, meshes with a large body of empirical research documenting the importance of ‘lifestyle’ and income as predictors of health and the perhaps surprisingly small impacts of changes in the provision of health care itself (McKeown 1976; Newhouse and Friedlander 1980; Brook et al. 1983; Berger and Leigh 1988). Grossman’s model also suggests that health, education, labour market outcomes, and the health-care system form a system with a variety of interacting causal effects, and thus that analysis of how to increase the efficacy of the health-care system or other interventions aimed to increase health must consider the complex interactions jointly to determine outcomes.
Demand for safety and the curious effects of airbags
One theme stressed in the discussion above is the importance of considering how policy changes will tend to change the incentives individuals face and therefore individuals’ behaviour. Consider the effects of a policy which forces consumers to purchase cars in which they are less likely to be seriously harmed or killed in the event of an accident. Such systems include mandatory passive restraint systems (such as airbags), collapsible steering columns, or side-impact reinforcements. Common sense suggests that such policies will necessarily reduce traffic fatalities, but the economics of the situation suggest that we must consider, in addition to the technical and engineering aspects of the situation, how behaviour will change in order to estimate how much fatalities can be expected to be reduced. Common sense may also be mistaken; it is possible that legislation of this sort may actually increase fatalities.
An individual can choose to drive slowly or at speed, to drive with much attention or carelessly, to talk on a mobile phone while driving or not, to drive while under the influence of alcohol or not, and many other choices which affect the likelihood of a collision. For simplicity, all of these choices can be grouped together into one category called ‘intensity’, with higher values of intensity representing actions tending to produce greater injury probabilities. Also assume that, if the probability of an accident is constant, people would prefer more intensity; they would rather get where they are going sooner and with less time and effort exerted to avoid accidents.
It is now possible to cast the situation as an economic model. Individuals can be considered to have preferences over our index ‘intensity’ and over safety, with more of both being preferred outcomes. For a given driver in a given car in a given city, we can associate a level of intensity, I, with a probability of being hurt in accident, P = P (I | F), where F is a set of parameters reflecting the characteristics of the car, driving conditions in the city, and so forth. The individual can be modelled as choosing the optimal trade-off between intensity and safety, subject to the constraint that increases in speed will reduce safety, in the same manner as the optimal trade-off between dental visits and other goods is modelled above, subject to the constraint that more money spent on dental visits will reduce money available for other goods.
Solving the utility maximization problem generates predicted driver behaviour as a function of the factors placed in F. How can changes in these factors affect behaviour? Suppose, for instance, that the driver has airbags fitted in his car. In this model, the probability of being hurt in an accident decreases for any given level of I. Since individuals like both intensity and safety, this shift out of the constraint will tend to cause a change in behaviour such that both intensity and safety increase; the driver goes faster and uses less care, but not by enough to outweigh the increased safety the airbag confers. Therefore there would be fewer driver fatalities but more accidents and more injuries amongst bicyclists and pedestrians. The ‘common-sense’ answer is at best misleading; whether mandatory safety systems are good policy depends not only on their engineering efficacy and cost, but also on how strongly drivers will change their behaviour in response to those systems.
This idea, often called ‘risk homeostasis’ or ‘risk offsetting’, was introduced by Peltzman (1975) and has been employed in a number of settings. Peltzman’s empirical analysis suggested that the mandatory safety improvements imposed on manufacturers in the late 1960s had an alarming effect; the ensuing decrease in driver fatalities was completely offset by an increase in third-party fatalities. Other studies have found that safety legislation tends to reduce fatality rates on net, but that the effect is lower than it would have been in the absence of risk offsetting behaviour (Lave et al. 1977, 1996; Graham 1984). The concept has been investigated in a variety of other settings, for instance child-resistant bottle caps (Viscusi 1984), workplace safety regulation (Viscusi 1979), and mattress flamability standards (Linneman 1980).
As a final example of economic analysis in health settings, a brief survey of the nascent field of economic epidemiology is presented. Economic epidemiology is distinguished by an emphasis on endogenous individual behaviour in the context of infectious disease. Specifically, this field examines how individuals will change their risky behaviour during the course of an epidemic and how that change in behaviour will, in turn, affect the spread of disease.
To underscore the importance of behavioural considerations, consider an individual who is at risk of contracting a sexually transmitted disease. Suppose that this individual has S randomly matched partners per year, and that the probability of contracting the disease P can be approximated as a linear equation in S, P = bS, where b is a transmission coefficient that depends on the biology of the disease and the proportion of partnerships with infected individuals. Suppose that a partially effective vaccine is made available such that the probability of transmission per partner is decreased. Will this vaccine reduce the probability that this individual becomes infected?
Holding the number of partners constant, the vaccine must reduce the probability the individual becomes infected. But the individual will generally change his behaviour if vaccinated; much like the driver in the previous section who drove faster in a car equipped with an airbag, this individual at risk of disease will take more risk (have more partners) when vaccinated. If this behaviour is incorporated into the analysis, we can write S = S(b), such that S(b) represents a sort of ‘demand curve’ for partners as a function of risk of transmission. Differentiating P = bS(b) with respect to b shows that the vaccine will only reduce this individual’s probability of becoming infected if it is the case that a 1 per cent decrease in the transmission coefficient produces less than 1 per cent increase in number of partners. The possibility cannot then be ruled out that introducing the vaccine into the population will spur rather than retard the spread of disease.
Epidemiological models of disease spread, which incorporate economic behavioural models, include those of Auld (1996), Kremer (1996), and Geoffard and Philipson (1996). One important result of these studies includes the notion that the hazard rate from susceptibility to infection may be decreasing in the proportion of the population infected; as more people become infected, those who are still susceptible face more incentive to reduce or cease engagement in risky behaviour. A possibly counterintuitive result is the theoretical possibility of ‘fatalism’, an increase in risk per partner producing increases in risky behaviour. This result will tend to hold for very high-risk individuals who may have lower incentive to avoid risky behaviour as the risk per partner rises since they may believe it is likely they have already been infected. Auld’s (1997) empirical analysis showed that a sample of gay men in San Francisco reduced their number of partnerships per 6-month period by about 17 per cent on average in response to a 10 per cent increase in the proportion of the (gay) population infected with HIV/AIDS. Other empirical analysis of behaviour in an epidemiological context includes the result that ‘assortive matching’—the tendency for infected individuals to match with other infected individuals, and vice versa—reduces HIV positivity/AIDS incidence by 25 to 40 per cent relative to the standard epidemiological assumption of random matching across infection states (Dow and Philipson 1996, Philipson and Dow 1998) and Ahitiv et al.’s (1996) result that condom use is an increasing function of local AIDS prevalence. See Philipson (1999) for a comprehensive review of recent work in economic epidemiology.
The aim of this chapter has been to provide an introduction to how economics methods can be used to assess the impacts of public health policy.
This has been done at the level of the individual, weighing up costs and benefits to themselves of particular courses of action. The main insights of the economics approach here are to highlight that:
individuals may be prepared to trade off health against other ‘goods’ in life
the introduction of a public health policy may change the ‘incentives’ which individuals face, causing them to behave in ways which reduce the overall effectiveness of the original policy
in modelling the spread of disease, it is important to consider the costs and benefits faced by individuals with respect to decisions on whether or not to participate in ‘hazardous’ activities.
A further aim of the chapter has been to provide a step-by-step guide to the principles of costing and measurement of benefits in the economic evaluation of public health programmes, and to use simple examples to illustrate how some of the concepts can be applied. It is important to follow such principles in order to render such work relevant to the decision contexts of economic evaluation.
As a result of providing a general overview, not all the principles outlined in this chapter will be applicable to every evaluation undertaken. For instance, one would not apply the principle of discounting to an evaluation of health-care alternatives whose costs and effects occur over a period of only a few months. However, it is important to use the principles outlined here as a guideline for which costs and benefits should be valued and how this is to be done. The guideline comprises the questions listed in Table 13. These questions should be addressed in advance of an evaluation. The answers to each of them will very much determine the nature of the exercise carried out. However, it is also important to refer to the checklist whilst an evaluation is ongoing and once it has been completed.
Table 13 Checklist for economic evaluation of health care
The task may seem daunting and often is. However, answering each of these questions will render any evaluation not only easier in the long run but also relevant to the decision-making context in which it is to be applied. In the ‘day-to-day’ world of decision-making, where full economic evaluations cannot always be undertaken, a more practical framework to apply might be that of programme budgeting and marginal analysis. The basic framework (or questions to be asked) which follows from this approach is detailed in Table 12. Because of the number of questions to be asked and the detailed nature of many exercises in costing and benefit estimation, it is recommended that an economist be involved in any health services evaluation, particularly in the design stage. If people are to use the techniques of economic evaluation then they ought to be fully understood. It is hoped that this chapter has shed some light on such understanding.
The figure £2859 (from Table 6) comprises:
1000 + (1000 × 1/1.05) + (1000 × 1/1.052) = 1000 + (1000 × 0.95240) + (1000) × (0.9070)
where 0.954 and 0.9070 are discount factors. In general
Dn = 1/(1 + r)n
where Dn is the discount factor, r is the discount rate, and n is the number of years ahead. In the example in Table 6 the discount rate r is 0.05 (or 5 per cent). Discount factors are obtainable directly from tables such as Table 14, which provides a list of discount factors annually from 1 to 50 years into the future at a discount rate of 5 per cent. The first two values in the column headed ‘discount factor’ are 0.9524 and 0.9070. This is because we are interested in discount factors 1 and 2 years from year 0 (which is now).
Table 14 Discount factors, equivalent annual costs, and present values of £1 per year for a discount rate of 5% (base date is year 0)
An alternative way of arriving at £2859 would be as follows:
£2859 = 1000 × (1 + 1.8594).
The sum of our discount factors (0.9524 + 0.9070) is 1.8594, which is the present value of £1 expended in years 1 and 2. Thus, if annual costs are all equal, calculation of discounted costs can be speeded up by calculating the present value of the sum of the discount factors and multiplying the present value by the annual amount expended (in this case £1000). In Table 14, 1.8594 is the second value in the column headed ‘present value of £1 per year’. This is because we are looking up to 2 years from year 0.
Equivalent annual costs
Let us assume that our surgical and drug interventions in Table 5 and Table 6 each incurred capital costs of £100 per patient and that the capital assets are costed only for the duration of each programme. From Table 14, it can be seen that three annual payments of 37 pence (that is, 36.72 pence rounded up) discounted at 5 per cent would sum to £1. Therefore three annual payments of £37 discounted at 5 per cent would sum to £100, that is
In this example, as capital costs are equivalent in each treatment regimen, their inclusion makes no difference to the final result that drug therapy is more efficient.
Ahitiv, A., Hotz, J., and Philipson, T. (1996). Is AIDS self-limiting? Evidence on the prevalence elasticity of the demand for condoms. Journal of Human Resources, 31, 869–98.
Alchian, A.A. (1972). Cost. In International encyclopaedia of the social sciences (ed. D. Stills), Volume 3. Macmillan, London.
Auld, M.C. (1996). Choices, beliefs, and infectious disease dynamics. Queen’s University Institute of Economic Research, Discussion paper 938. Queen’s University, Ontario.
Auld, M.C. (1997). Behavioral response to the AIDS epidemic: structural estimates from panel data. University of Calgary.
Avorn, J. (1984). Benefit and cost analysis in geriatric care: turning age discrimination into health policy. New England Journal of Medicine, 310, 1294–300.
Becker, G.S. (1965). A theory of the allocation of time. Economic Journal, 75, 493–517.
Becker, G. and Murphy, K. (1988). A theory of rational addiction. Journal of Political Economy, 96, 675–700.
Berger, M. and Leigh, J. (1988). The effect of alcohol use on wages. Applied Economics, 20, 1343–51.
Berwick, D.M. and Weinstein, M.C. (1985). What do patients value? Willingness to pay for ultrasound in normal pregnancy? Medical Care, 23, 881–93.
Birch, S. and Donaldson, C. (1987). Applications of cost benefit analysis to health care: departures from welfare economic theory. Journal of Health Economics, 6, 211–25.
Birch, S. and Gafni, A. (1994). Cost-effectiveness ratios: in a league of their own. Health Policy, 28, 133–41.
Bond, J., Gregson, B.A., and Atkinson, A. (1989). Measurement of outcomes within a multicentred randomized controlled trial in the evaluation of the experimental NHS nursing homes. Age and Ageing, 18, 292–302.
Boyle, M.H., Torrance, G.W. Sinclair, J.C., and Horwood, S.P. (1983). Economic evaluation of neonatal intensive care of very-low-birthweight infants. New England Journal of Medicine, 308, 1330–7.
Brazier, J., Deverill, M., Green, C., Harper, R., and Booth, A. (1999). A review of the use of health status measures in economic evaluation. Health Technology Assessment, 3.
Brook, R.H., Ware, J.E., Rogers, W.H., et al. (1983). Does free care improve adult’s health? Results from a randomized controlled trial. New England Journal of Medicine, 309, 1426–34.
Buckingham, K. (1993). A note on HYE (healthy years equivalent). Journal of Health Economics, 11, 301–9.
Buhaug, H., Skjeldstad, F.E., Backe, B., and Dalen, B. (1989). Cost effectiveness of testing for chlamydial infections in asymptomatic women. Medical Care, 27, 833–41.
Bush, J.W., Chen, M.M., and Patrick, D.L. (1973). Health status index in cost-effectiveness: analysis of a PKU program. In Health status indexes (ed. R.L. Berg). Hospital and Educational Research Trust, Chicago.
Cohen, D. (1994). Marginal analysis in practice: an alternative to needs assessment for contracting health care. British Medical Journal, 309, 781–5.
Cohen, D. (1995). Messages from Mid Glamorgan: a multi-programme experiment with marginal analysis. Health Policy, 33, 147–55.
Craig, N., Parkin, D., and Gerard, K. (1995). Clearing the fog on the Tyne: programme budgeting in Newcastle and North Tyneside Health Authority. Health Policy, 33, 107–25.
Culyer, A.J. and Wagstaff, A. (1993). QALYs and HYEs. Journal of Health Economics, 11, 311–23.
Dixon, J. and Welch, H.G. (1991). Priority setting: lessons from Oregon. Lancet, 337, 891–4.
Doessel, D.P. (1978). Economic analysis and end stage renal disease: an Australian study. Economic Analysis and Policy, 8, 21–36.
Dolan, P. (1997). Modelling valuations for Euroqol health states. Medical Care, 35, 1095–108.
Donaldson, C. (1999). Valuing the benefits of publicly-provided health care: Does ‘ability to pay’ preclude the use of ‘willingness to pay’? Social Science and Medicine, 49, 551–63.
Donaldson, C. and Bond, J. (1991). Cost of continuing-care facilities in the evaluation of experimental National Health Service nursing homes. Age and Ageing, 20, 160–8.
Donaldson, C. and Farrar, S. (1993). Needs assessment: developing an economic approach. Health Policy, 25, 95–108.
Donaldson, C. and Gregson, B. (1989). Prolonging life at home: what is the cost? Community Medicine, 11, 200–9.
Donaldson, C. and Mooney, G. (1991). Needs assessment, priority setting, and contracts for health care: an economic view. British Medical Journal, 303, 1529–30.
Donaldson, C. and Shackley, P. (1997). Does ‘process utility’ exist? a case study of willingness to pay for laparoscopic cholecystectomy. Social Science and Medicine, 44, 699–707.
Donaldson, C., Wright, K., and Maynard, A. (1986). Determining value for money in day hospital care for the elderly. Age and Ageing, 15, 1–7.
Donaldson, C., Walker, A., and Craig, N. (1995). Programme budgeting and marginal analysis: a handbook for applying economics in health care purchasing. Scottish Forum for Public Health Medicine, Glasgow.
Donaldson, C., Mapp, T., Ryan, M., and Curtin, K. (1996). Estimating the economic benefits of avoiding food-borne risk: is ‘willingness to pay’ feasible? Epidemiology and Infection, 116, 285–94.
Donaldson, C., Hundley, V., and Mapp, T. (1998). Willingness to pay: A method for measuring preferences for maternity care? Birth, 25, 33–40.
Dow, W. and Philipson, T. (1996). An empirical investigation of the implications of assortive matching on the incidence of HIV. Journal of Health Economics, 15, 735–49.
Drummond, M. and Mooney, G. (1983). Essentials of health economics. Northern Health Economics, Aberdeen.
Drummond, M., O’Brien, B., Stoddart, G., and Torrance, G. (1997). Methods for the economic evaluation of health care programmes. Oxford University Press.
Euroqol Group (1990). Euroqol—a new facility for the measurement of health-related quality of life. Health Policy, 16, 199–208.
Froberg, D.G. and Kane, R.L. (1989). Methodology for measuring health-state preferences—II. Scaling methods. Journal of Clinical Epidemiology, 42, 459–71.
Gafni, A. (1991). Willingness to pay as a measure of benefits: relevant questions in the context of public decision making about health care programmes. Medical Care, 29, 1246–52.
Gafni, A. and Birch, S. (1993). Searching for a common currency: critical appraisal of the scientific basis underlying European harmonization of the measurement of Health Related Quality of Life (Euroqol). Health Policy, 23, 219–28.
Gafni, A. and Zylak, C.J. (1990). Ionic versus non-ionic contrast media: a burden or a bargain? Canadian Medical Association Journal, 143, 475–8.
Gafni, A., Birch, S., and Mehrez, A. (1993). Economics, health and health economics: HYEs versus QALYs. Journal of Health Economics, 11, 325–39.
Geoffard, P. and Philipson, T. (1996). Rational epidemics and their public control. International Economic Review, 37, 603–24.
Gerard, K. (1990). Economic evaluation of respite care for children with a mental handicap: a preliminary analysis of problems. Mental Handicap, 18, 150–5.
Gerard, K. and Mooney, G. (1993). QALY league tables: handle with care. Health Economics, 2, 59–64.
Gerard, K., Salkeld, G., and Hall, J. (1990). Counting the costs of mammography screening: first year results from the Sydney study. Medical Journal of Australia, 152, 466–71.
Gold, M.R., Siegel, J.E., Russell, L.B., and Weinstein, M. (1996). Cost effectiveness in health and medicine. Oxford University Press.
Gold, L., Raftery, J., and Soderlund, N. (1997). Costing diseases at DHA level: a standardised approach? Health Economists’ Study Group, York.
Gould, D. (1971). A groundling’s notebook. New Scientist, 51, 217.
Graham, J.D. (1984). Technology, behaviour, and safety: an empirical study of automobile occupant-protection regulation. Policy Science, 74, 141–51.
Gray, A.M. and Steele, R. (1979). The identification of the costs of maternity care: a programme approach to health service expenditure. Health Economics Research Unit. Discussion paper 03/79. University of Aberdeen.
Grossman, M. (1972a). On the concept of health capital and the demand for health. Journal of Political Economy, 80, 223–55.
Grossman, M. (1972b). The demand for health: a theoretical and empirical investigation. Occasional paper 119. Columbia, New York.
Harrison, A.J. (1974). The economics of transport appraisal. Wiley, New York.
HM Treasury (1982). Investment appraisal in the public sector. HMSO, London.
Hundley, V., Cruickshank, F., Lang, G., et al. (1994). Midwife managed delivery unit: a randomised controlled comparison with consultant led care. British Medical Journal, 309, 1400–4.
Hundley, V., Donaldson, C., Lang, G.D., et al. (1995). Costs of intrapartum care in a midwife managed delivery unit and a consultant led labour ward. Midwifery, 11, 103–9.
Johannesson, M. and Jonsson, B. (1991). Economic evaluation in health care: is there a role for cost-benefit analysis? Health Policy, 17, 1–23.
Johannesson, M. and Karlsson, G. (1997). The friction cost method: a comment. Journal of Health Economics, 16, 249–55.
Johannesson, M., Pliskin, J.S., and Weinstein, M.C. (1993). Are healthy years equivalents an improvement over quality adjusted life years? Medical Decision Making, 13, 281–6.
Johannesson, M., Jonsson, B., Kjekshus, J., et al. (1997). Cost effectiveness of simvastatin treatment to lower cholesterol levels in patients with coronary heart disease. New England Journal of Medicine, 336, 332–6.
Johnston, K., Brown, J., Gerard, K., O’Hanlon, M., and Morton, A. (1998). Valuing temporary and chronic health states associated with breast screening. Social Science and Medicine, 47, 213–22.
Kaplan, R.M., Bush, J.W., and Berry, C.C. (1976). Health status: types of validity and the index of well-being. Health Services Research, 11, 478–507.
Klarman, H.F., Francis, J.O., and Rosenthal, C.D. (1968). Cost-effectiveness analysis applied to the treatment of chronic renal disease. Medical Care, 6, 48–54.
Koopmanschap, M.A., Rutten, F.F.H., van Ineveld, B.M., et al. (1995). The friction cost method for measuring the indrect costs of disease. Journal of Health Economics, 14, 171–89.
Kremer, M. (1996). Integrating behavioral choice into epidemiological models of the AIDS epidemic. Quarterly Journal of Economics, 111, 549–73.
Kristiansen, I.S., Egen, A.E., and Thelle, D.S. (1991). Cost effectiveness of incremental programmes for lowering serum cholesterol concentration: is individual intervention worthwhile? British Medical Journal, 302, 1119–22.
Lave, C., Elias, P., and Robertson, L.S. (1977). A critical analysis of Peltzman’s ‘The effects of automobile safety regulation’. Journal of Economic Issues, 11, 587.
Linneman, P. (1980). The effects of consumer safety standards: the 1973 mattress flammability standard. Journal of Law and Economics, 23, 461–79.
Loomes, G. (1995). The myth of the HYE. Journal of Health Economics, 14, 1–8.
Loomes, G. and McKenzie, L. (1989). The use of QALYs in health care decision making. Social Science and Medicine, 28, 299–308.
McIntosh, E., Donaldson, C., and Ryan, M. (1999). Recent advances in the methods of cost-benefit analysis in health care: matching the art to the science. Pharmaco-Economics, 15, 357–67.
McKeown, T. (1976). The rise of modern populations. Academic Press, New York.
Mark, D.B., Hlatky, M.A., Califf, R.M., et al. (1995). Cost-effectiveness of thrombolytic therapy with tissue plasminogen activator as compared with streptokinase for acute myocardial infarction. New England Journal of Medicine, 332, 1418–24.
Mas-Colell, A., Whinston, M., and Green, J. (1995). Microeconomic theory. Oxford University Press.
Measurement and Valuation of Health Group (1994). The measurement and valuation of health: first report on the main survey. Centre for Health Economics, University of York.
Mehrez, A. and Gafni, A. (1989). Quality adjusted life years, utility theory, and healthy years equivalents. Medical Decision Making, 9, 142–9.
Mehrez, A. and Gafni, A. (1993). Healthy years equivalents versus quality adjusted life years: in pursuit of progress. Medical Decision Making, 13, 287–92.
Meltzer, D. (1997). Accounting for future costs in medical cost-effectiveness analysis. Working paper 5946. National Bureau of Economic Research, Cambridge, MA.
Miedzybrodzka, Z., Shackley, P., Donaldson, C., and Abdalla, M. (1994). Counting the benefits of screening: a pilot study of willingness to pay for cystic fibrosis carrier screening. Journal of Medical Screening, 1, 82–3.
Miller, P., Parkin, D., Lewis, D., and Gerard, K. (1995). Final report of the project ‘Developing programme budgeting and marginal analysis in purchasing organisations’. University of Newcastle upon Tyne.
Miller, P., Parkin, D., Craig, N., Lewis, D., and Gerard, K. (1997). Less fog on the Tyne? Programme budgeting in Newcastle and North Tyneside. Health Policy, 40, 217–29.
Mitton, C., Donaldson, D., and Dean, S. (1999). Developing a framework for priority setting in regionalised settings: the role of economics. Presented at the Canadian Health Economics Research Association, 17–19 August, Edmonton, Alberta.
Mooney, G. (1977). Programme budgeting in an area health board. Hospital and Health Services Review, November, 379–84.
Mooney, G. (1984). Program budgeting: an aid to planning and priority setting in health care. Effective Health Care, 2, 65–8.
Mooney, G.H. (1989). Economic evaluation: the Australian road to health service efficiency. In Health care evaluation. Report of the National Health Care Evaluations Workshop. Public Health Association of Australia, Canberra.
Mooney, G. and Lange, M. (1993). Antenatal screening: what constitutes ‘benefit’? Social Science and Medicine, 37, 873–8.
Mooney, G., Russell, E., and Weir, R. (1986). Choices for health care: a practical introduction to the economics of health provision. Macmillan, London.
Mooney, G., Gerard, K., Donaldson, C., and Farrar, S. (1993). Priority setting in purchasing: some practical guidelines. Research paper no. 6. National Association of Health Authorities and Trusts, Birmingham.
Mooney, G., Jan, S., and Seymour, J. (1994). The NSW health outcomes initiative and economic analysis. Australian Journal of Public Health, 18, 244–8.
Mugford, M., Kingston, J., and Chalmers, I. (1989). Reducing the incidence of infections after Caesarean section: implications of prophylaxis with antibiotics for hospital resources. British Medical Journal, 299, 1003–6.
Newhouse, J.P. and Friedlander, L.J. (1980). The relationship between medical resources and measures of health: some additional evidence. Journal of Human Resources, 15, 200–18.
Nord E. (1991). The validity of a visual analogue scale in determining social utility weights for health states. International Journal of Health Planning and Management, 6, 234–42.
Nord, E., Pinto, J.L., Richardson, J., et al. (1999). Incorporating societal concerns for fairness in numerical valuations of health programmes. Health Economics, 8, 25–40.
Olsen, J.A. and Donaldson, C. (1998). Helicopters, hearts and hips: using willingness to pay to set priorities for public sector health care programmes. Social Science and Medicine, 46, 1–12.
Parsonage, M. and Neuberger, H. (1992). Discounting and health benefits. Health Economics, 1, 71–5.
Peacock, S. and Edwards, D. (1997). Setting priorities in South Australian community health: the mental health program budget. Center for Health Program Evaluation, Monash University, Melbourne.
Peltzman, S. (1975). The effects of automobile safety regulation. Journal of Political Economy, 83, 677–725.
Philipson, T. (1999). Economic epidemiology and infectious diseases. In Handbook of health economics (ed. J.P. Newhouse and A.J. Culyer). North-Holland, Amsterdam.
Philipson, T. and Dow, W. (1998). Infectious disease transmission and infection-dependent matching. Mathematical Biosciences, 148, 161–80.
Pole, J. (1974). Programs, priorities and budgets. British Journal of Preventive and Social Medicine, 28, 191–5.
Robbins, L. (1935). The nature and significance of economic science. Macmillan, London.
Robertson, L.S. (1996). Reducing death on the road: the effects of minimum safety standards, publicized crash tests, seat belts, and alcohol. American Journal of Public Health, 86, 31.
Rosser, R. and Kind, P. (1978). A scale of valuations of states of illness: is there a social consensus? International Journal of Epidemiology, 7, 347–58.
Russell, I.T., Devlin, B., Fell, M., et al. (1977). Day case surgery for hernias and haemorrhoids: a clinical, social and economic evaluation. Lancet, i, 844–7.
Ruta, D., Donaldson, C., and Gilray, I. (1996). Economics, public health and health care purchasing: the Tayside experience of programme budgeting and marginal analysis. Journal of Health Service Research and Policy, 1, 185–93.
Scott, A., Currie, N., and Donaldson, C. (1998). Evaluating innovation in general practice: a pragmatic framework using programme budgeting and marginal analysis. Family Practice, 15, 216–22.
Scott, A., Donaldson, C., and Scott, S. (1999) Program budgeting and marginal analysis: pragmatism and policy (editorial). Journal of Health Services Research and Policy, 4, 1–2.
Shackley, P. and Dixon, S. (2000). Using contingent valuation to elicit public preferences for water fluoridation. Applied Economics, 32, 777–87.
Sintonen, H. and Pekurinen, M. (1993). A fifteen-dimensional measure of health-related quality of life (15D) and its applications. In Quality of life assessment: key issues in the 1990s (ed. S.R. Wallner and R.M. Rosser), pp. 185–95. Kluwer, Dordrecht.
Stanford, M. (1990). Counting the costs of mammography screening: first year results from the Sydney study. Medical Journal of Australia, 153, 175.
Tengs, T.O. (1996) An evaluation of Oregon’s Medicaid rationing algorithms. Health Economics, 5, 171–82.
Torgerson, D., Donaldson, C., and Reid, D. (1994). Private versus social opportunity cost of time: valuing time in the demand for health care. Health Economics, 3, 149–55.
Torrance, G.W. (1986). Measurement of health state utilities for economic appraisal: a review. Journal of Health Economics, 5, 1–30.
Torrance, G.W., Thomas, W.H., and Sackett, D.L. (1972). A utility maximization model for evaluation of health care programmes. Health Services Research, 7, 118–33.
Torrance, G.W., Furlong, W., Feeny, D., and Boyle, M. (1995). Multi-attribute preference functions. Health utilities index. Pharmacoeconomics, 7, 503–20.
Twaddle, S. and Walker, A. (1995). Programme budgeting and marginal analysis: application within programmes to assist purchasing in Greater Glasgow Health Board. Health Policy, 33, 91–105.
Viney, R., Haas, M., and Mooney, G. (1995). Program budgeting and marginal analysis: a guide to resource allocation. NSW Public Health Bulletin, 6, 29–32.
Viscusi, K. (1979). The impact of occupational health and safety regulation. Bell Journal of Economics, 10, 117–40.
Viscusi, K. (1984). The Lulling effect: the impact of child-resistant packaging on aspirin and analgesic ingestions. American Economic Review, 74, 324–7.
Williams, A. (1985). Economics of coronary artery bypass grafting. British Medical Journal, 291, 326–9.
Wiseman, V., Mooney, G., and Stephenson, J. (1998). An economic approach to priority setting in health care. University of Sydney.