Chapter 6. Valuation of the social benefits of public health programmes

CHAPTER 6 Valuation of the social benefits of public health programmes The choice among programmes requires the setting of relative social values on the different effects of such programmes: mortality in different age groups, impairment, disability, and changes in income. Alternative approaches to the establishment of relative social value weights are reviewed. Up to now we have been dealing primarily with the technical questions of estimating the costs and results of public health programmes and the prob- lems of valuing the purely economic effects of such programmes on national income. Although judge- ments are involved in all of this, they are essentially judgements about technical matters. We now turn to that part of the analysis in which value judgements are central: the setting of relative social values on the different effects of health programmes. The analysis presented in the previous chapters has shown how the impact of a health programme can be described in terms of its effects on mortality in different age groups, on impairment, on disability, and on economic gains (i.e., changes in income). The choice between any two programmes or tech- niques should be governed by the effect that each will produce on all these aspects. Of course, in some cases the advantages of one programme or technique will clearly outweigh those of the other -i.e., for the same resource costs the chosen pro- gramme will prevent more deaths in each age class, produce greater reductions in impairment and disability, and yield a larger increase in national income. In general, however, a programme that is preferable with respect to some outputs will be less favourable with respect to others-e.g., it will save more lives in some age ranges but fewer in others. In such cases, to make a choice of programme we need an overall single measure of the social value of the effects of each alternative. It is important to recognize that this overall value is not the economic loss discussed in the last chapter. Although death, impairment, and disability decrease national income, this is only part of the reason for trying to prevent them. It is a fundamental character- istic of our method that it gives explicit consideration to the social value of reduced mortality and mor- bidity as well as to the impact of such benefits on national income. Our approach recognizes that a life has a value of its own over and above the income that it produces. The calculation of the overall social value of a programme requires a relative value scale for the different physical outputs. Such a scale indicates the relative weights to be given to such specific effects as preventing one death in the age group 0-14 years, or obviating one year of bed disability. There is a more operational and explicit way of thinking about these relative weights. Consider two programmes that are identical in all respects except that the first prevents the death of 100 children (between 0 and 14 years of age) but of no young adults (aged 15-44 years), while the second prevents the death of 50 young adults and of no children. If we have no preference for one or the other programme we are implicitly assigning relative weights of 1 and 2 to deaths prevented in these two age groups.1 More generally, if we have no pre- ference either for a programme that prevents the death of 100 children or for one that prevents the death of N young adults, the relative weights are 1 and 100/N. Similar relative weights must be estab- lished for each of the other outputs of health programmes. An alternative operational definition of the relative weights may further clarify their nature. (This approach may also be more compatible with phy- 1 Note that only relative weights matter: we could equally well use 0.5 and 1.0. Note also that this discussion implies that the weights are assumed to be constant and the evaluation is assumed to be additive. -75 - 76 M. S. FELDSTEIN ET AL. sicians' clinical experience than the method of esta- blishing equivalences between numbers of lives saved in different age groups.) The new problem could be posed as follows: " Consider first a situation in which there are two patients, one in each of the age groups 0-14 years and 15-44 years. Each patient has a probability of dying of 0.2 unless treated with a particular drug. There is only enough of the drug to treat one of them. If treatment reduced the probability of death to 0.1 in both cases, to which patient would you give it?" If the answer is that the treatment should be given to the young adult rather than the child, it is implied that the second weight (w2) is greater than the first (w1). This does not, however, establish the relative weights. They may be established by modifying the problem as follows: " If the drug were less effective in treating the older patient, to whom would you give the treatment? The answer clearly depends on how much less effective is the drug in this instance. If w1 exceeds zero (i.e., if there is some value attached to saving the life of a child), then if the drug is suffi- ciently less effective in reducing the probability of death for the older patient, you would treat the child. More specifically, if treatment lowers the probability of death for the child from 0.2 to 0.1 and that for the older patient from 0.2 to p-p being greater than 0.1-then there exists a value of p at which you would be indifferent as to which patient should be treated. The value of p must be determined." In practice it would be easier to replace this formal statement by a series of questions with high and low trial values of p until the point of indifference is found. The higher the value of p (i.e., the smaller the prospective improvement for the older patient), the higher is the relative weight on preventing a death in the older age group. More specifically, the indifference probability p and the values of w, and w2 satisfy (0.2-p)w2 = (0.2-0.l)wl. This implies that the ratio of w1 to w2 is 2-lop.1, 2 A similar type of approach can be used to define the relative weights for comparing death, impair- ment, and disability. How should the relative weights for use in pro- gramme planning be established? It is our view ' For a full discussion of a variety of methods for eliciting relative values, see Fishburn (1964). 2The existence of a unique ratio of w, to w2 does not violate the condition that the Von Neumann utility func- tion is only unique up to an increasing linear transformation because of the linear and additive way in which both types of reduced mortality benefits enter the utility function. The values of w, and w, are essentially marginal utilities and therefore not affected by nonhomogeneous transformations. that they must ultimately reflect the value judge- ments of the responsible government officials or public health administrators. The two operational definitions of relative weights discussed above provide alternative methods of introspection and questioning. The process of assigning weights is obviously difficult, but it is also clearly unavoidable. Any rational method of choosing among programmes must reflect such a set of relative values even if these are never made explicit. We believe that better decisions can be made by explicitly consider- ing the relative values of the different major pro- gramme outputs. There are now a variety of proposed methods of constructing a relative value scale for health pro- gramme outputs. These have generally been proposed as "solutions" to the evaluation problems that replace the use of the preferences and value judge- ments of the responsible officials. We prefer to think of them as sources of information that can guide the ultimate value judgements. The proposals can be classified into two groups. One group of methods has been suggested by health planners and seeks to make use of " natural " relations among the different outputs. The second group has been pro- posed by economists and seeks to make use of the relative values that representative households place on the different outputs. Examples of both types of procedure will now be reviewed. The most common " solution " proposed by health planners is to avoid the difficulty by focusing on one output and ignoring all others. The general mortality rate and the infant mortality rate are the most common single measures. A slightly more general measure is adopted by the CENDES 3 health planners, who use life expectancy as the criterion for health planning decisions. This is equivalent to giving zero weights to reductions in impairment and disability and to income gains while giving weights to the number of deaths prevented that would probably decline continually with age.4 Chiang (1965) proposes an index that takes into account both mortality and disability. He gives equal weight to a day lost through either death or disability. This implies that the extension of an individual's life expectancy by one year is equivalent to preventing 365 days of disability. No allowance 3 Centro de Estudios del Desarrollo (Center for Develop- ment Studies), Universidad Central, Caracas, Venezuela. 4 The weights would decline continually with age if the remaining life expectancy decreased with increasing age. RESOURCE ALLOCATION MODEL FOR PUBLIC HEALTH PLANNING 77 is made for impairment or for the difference between present and future benefits. In an early progress report on the present study, Piot & Sundaresan1 suggested the use of a dis- counted present sum of " years of healthy living ". For this purpose a current day of disability and a current day lost because of total impairment or death are equal. The total value ascribed to a death is the discounted present sum of all the future lost days. Future disability days are also discounted. Partial impairments are evaluated by multiplying the proportional impairment by the discounted period of impairment (typically the remaining life span). This has the effect of making death at any particular age equivalent to total impairment for life at that age or to a 20% impairment of 5 persons of that age. The use of the discount rate is quite important in this approach: depending on the particular discount rate the method can imply that no difference is made between one year of bed disability now and perhaps as much as 5 or 10 years of survival some 40 years in the future. Economists have generally objected to these methods of combining outputs of different kinds and have suggested that the appropriate relative weights should reflect evidence on the preferences of the individuals in the population for which the programme is being planned.2 Two different ap- proaches have been suggested for obtaining such evidence. Dreze (1962) suggested that by observing individuals' behaviour in various situations it should be possible to infer their evaluations. More speci- fically, individuals engage in a variety of activities in which there are risks of death, disability, and impairment. Often these risks can be avoided or reduced at some cost to the individual. For example, it may be decided to accept lower pay for a less hazardous job, to acquire equipment that increases safety at home, or to purchase medical care. By studying these decisions it should be possible to infer the relative weights that individuals are using in their evaluations. Although this is a useful approach, there are three problems in applying it. First, any choice (e.g., of employment or whether 1 Piot, M. & Sundaresan, T. K. (1967) A linear pro- gramme decision model for tuberculosis control: progress report on the first test-runs, Geneva (unpublished mimeo- graphed document WHO/TB/Techn. Information/67.55). " Some economists, such as Fein (1958) and Weisbrod (1961), have chosen to ignore all the benefits except the income benefits described in the last chapter. We reject this approach, reiterating that life and limb are of value per se and not just as means of earning an income. to purchase safety equipment) is likely to involve many different effects-i.e., it will change the pro- babilities of death, disability, and impairment. This makes it very difficult to infer a set of relative weights unless many such choices can be studied. Secondly, individuals may be misinformed about the risks associated with different actions, and therefore their decisions will reflect these incorrect probability assessments. If the analyst uses the correct proba- bility assessments in inferring the individuals' pre- ferences, he will draw incorrect inferences. Finally, individuals faced with complex decisions of this sort may not behave rationally-i.e., they may not wish their actions to be used as a basis for public decisions. Despite these shortcomings, the idea is a promising one. It could be quite useful to the responsible officials to know the relative values that the public is currently placing on health effects. Even if the officials do not feel bound to use these particular weights, information is provided that should be of value in forming their own judgements. Schelling (1968) has suggested that a simpler and more direct way of discovering relative values is by eliciting the opinion of representative individuals as to the monetary value they would assign to reducing their probabilities of death, disability, and impairment.3 He would ask an individual a question of the following type: " Suppose that something causes a small increase-of 1 %, say-in your prob- ability of death for this year only. How much would you be willing to pay to remove that increase-i.e., to return to your normal death rate? " Suppose 100 individuals answered that they would each pay US$ 1 000. This implies that together the group would pay US$ 100 000 to save one life. Schelling also suggests a number of additional features, including a willingness to pay for a reduction in the risk of someone else's death. We shall not enter into these details here. The two important features of this method are that (1) it involves a direct questioning of individuals in the population rather than a mere observation of their behaviour, and (2) it seeks to elicit the value of small probability changes. Acton (1970) showed that this method can be applied to a representative population sample in a way that gives consistent and plausible values. It has advantages in comparison with the method of 3Schelling's analysis was actually limited to changes in the probability of death, since he was concerned about the relative weights on mortality and income. There is no logical difficulty in extending this to impairment and dis- ability. 78 M. S. FELDSTEIN ET AL. Dreze-namely: one risk can be considered at a time; the probabilities can be stated in the problem; and the rationality of individual behaviour is irre- levant. It does have the disadvantage, however, that people's statements about their behaviour in hypo- thetical situations may not reflect either their true preferences or the course that they would adopt in an actual situation. Nevertheless, it is clear that this method can provide useful information of help to the responsible official in determining his final preferences. This concludes our summary of the alternative suggestions for determining relative weights. It should be emphasized that, although the decision process must incorporate such a set of relative weights, the choice among alternative programmes or techniques may not be very sensitive to the actual relative weights that are used. If the decisiQn-maker is uncertain about his own preferences, he might ascertain which programme is best for different sets of weights. We would suggest that this should be done in any practical application: a variety of different relative weights should be tried and the resulting optimum plans should be available to the decision-maker. Several such sensitivity tests are discussed in Chapter 9 of this study. REFERENCES Acton, J. (1970) Evaluation of a life-saving program: the case of heart attacks, Cambridge, Mass. (unpub- lished dissertation, Harvard University) Chiang C. L. (1965) An index of health: mathematical models, Washington, D.C., US Government Printing Office (National Center for Health Statistics, Series 2, No. 5) Dreze, J. H. (1962) Rev. franc. Recherche operationnelle, 23, 3-28 Fein, R. (1958) Economics of mental illness, New York, Basic Books Fishburn, P. C. (1964) Decision and value theory, New York, Wiley, chap. 4 Schelling, T. C. (1968) The life you save may be your own. In: Chase, S. B., ed., Problems in public expen- diture analysis, Washington, D.C., The Brookings Institution Weisbrod, B. A. (1961) Economics of public health, Philadelphia, University of Pennsylvania Press; Lon- don, Oxford University Press