Preventive Health Strategies and the Policy Makers5 Paradox Eric Y. Brown, MD; Catherine M. Viscoli, PhD; and Ralph I. Horwitz, MD

• The likelihood of developing many diseases is predicted by levels of risk factors. Many public health strategies have been created to apply interventions (for example, drugs, diets, exercise) intended to lower levels of these factors and thereby prevent disease. Often, these strategies are based on the interpretation of incomplete evidence for the effectiveness of the interventions. The reason this evidence is and will likely remain incomplete is explained by the policy makers' paradox. The paradox occurs when evidence for an intervention's effectiveness is obtained in persons with the highest levels of a risk factor, but the application of the intervention may have the greatest potential for reducing disease burden in persons with lower levels. Resolution of the paradox requires consideration of the type and quality of evidence, the society's time preference for risk, and the society's choice about allocation of scarce resources. Annals of Internal Medicine. 1992;116:593-597. From the Yale University School of Medicine, New Haven, Connecticut. For current author addresses, see end of text.

Xlealth policy is generally based on expert analyses of available data. Frequently these data stratify populations according to risk factors and find that although those with the highest levels of a factor have the highest relative probability of developing a disease, it is the numerically larger portion of the population with low to moderate levels of the factor that accounts for the largest absolute number of cases. Nevertheless, measuring the effectiveness of interventions is often most feasible for persons with the highest levels of the factor. These two considerations can result in a paradoxical relationship between the population in which an intervention is tested and the population in which it may have its greatest effect, a circumstance we refer to as the policy makers' paradox. Different resolutions of this paradox may lead to dissimilar health policies for a society. One of the more striking examples of this phenomenon has been the alternate strategies chosen by the National Cholesterol Education Program (NCEP) in the United States, and the Toronto Working Group in Canada, to target interventions intended to prevent coronary artery disease. Both groups base their recommendations on interpretations of the same data, but they differ in defining the level of risk at which intervention is recommended (1). The NCEP recommends measurement of serum cholesterol in all Americans over 20 years of age. If serum cholesterol is above 6.21 mmol/L

(240 mg/dL) and the individual has at least one other cardiac risk factor, the NCEP recommends measurement of lipoproteins. If the individual is a man without coronary disease and has a low density lipoprotein (LDL) level above 4.2 mmol/L (160 mg/dL), the NCEP recommends a cholesterol-lowering diet and, if that is unsuccessful, cholesterol-lowering drugs (2). In contrast, the Toronto Working Group recommends measurement of cholesterol only in selected groups: men, aged 35 to 59, with one or more other cardiac risk factors; men, aged 20 to 34 and 60 to 69, with two or more risk factors; and, optionally, women who are at high risk for other reasons. Furthermore, they recommend measurement of lipoproteins only if cholesterol is greater than 6.85 mmol/L (265 mg/dL), and intervention only if the LDL level is greater than 4.9 mmol/L (190 mg/dL). Finally, compared with the NCEP, the Toronto group suggests that a higher level of LDL be required to consider that dietary therapy has failed (1, 2). Although the differences in these two policies may appear small, their social and economic effects are enormous. It has been estimated that if the NCEP's strategy were replaced in the United States by that of the Toronto Working Group, more than 95 million Americans would be excluded from routine screening, serum lipoproteins would be measured in one third fewer persons, and cholesterol-lowering interventions would be imposed on one fifth as many persons (1). When experimental evidence on the effects of an intervention is incomplete, as it is for cholesterol screening, the selection of a health strategy is actually the choice of an approach to the available evidence, a choice that requires resolution of the policy makers' paradox. In the example discussed above, the Canadian group has used a stricter application of this evidence whereas the NCEP has relied on a more expansive interpretation. Although we recognize that this difference is partially explained by differences between the two health care systems, our intent is to give an alternative description: the policy makers' paradox. The following discussion attempts to make explicit the problem of the policy makers' paradox and some common assumptions that allow it to be resolved. The Policy Makers' Paradox A discussion of the paradox must begin with an explanation of the terms risk and risk factor. The classical epidemiologic definition of risk is " . . . the probability of developing a disease over a stated period of time" (3). A risk factor, on the other hand is: . . . identified on the basis of observed variations in disease frequency. If a particular factor is associated © 1992 American College of Physicians

Downloaded from by Karolinska Institute user on 01/18/2019


with an elevated frequency of occurrence of the disease, and if the association cannot be explained on the basis of confounding or methodological biases, then that factor is regarded as a risk factor (3). The contribution of risk factors to disease occurrence may be measured in two ways. The first is absolute risk: the rate at which the disease occurs at differing levels of the risk factor. The second is relative risk: the ratio of the rate of disease at one level of the factor compared to another level. Although cases of the disease would be expected to occur at all levels of a dimensionally measured factor, such as cholesterol; intuitively, it also would be expected that the segment of the population with the highest levels of the factor would account for the largest number of cases. In reality, the relationship between the level of a factor and disease occurrence is more complex because it depends both on the pattern of relative risk at different levels of the factor and on the distribution of that factor within the population. If the segment of the population with the highest levels is small in size, it may account for only a small proportion of cases of the disease despite a high relative risk. For example, women may be stratified by family history into groups with a risk for breast cancer up to nine times that of the general population. Despite this high relative risk, the small number of women with such genetic risk factors is estimated to account for not more than 15% of new cases (4). The paradox for the policy planner results directly from a tradeoff between the feasibility of performing an experiment (the objective of the investigator) and the generalizability of its results (the concern of the policy maker). Groups of individuals with lower levels of risk factors have lower rates of the outcome than those with higher levels and studying interventions in such groups requires larger numbers of subjects to detect clinically meaningful differences in event rates. Studying the same interventions in groups with higher rates of outcome requires fewer subjects and is logistically more feasible. Unfortunately and paradoxically, it is the former group in which interventions are most often recommended (because they may have the greatest disease burden) and the latter group in which experiments are most often performed.

Table 1. The Effects of Lowering Cholesterol on the Relative and Absolute Risk for Coronary Artery Disease Trial


Sex Cholesterol*

y LRCC-PPT§ 35 to 59 Male 40 to 55 Male HHSU

Relative Absolute Risk Risk Reduction! Reduction^


mmol/L 7.21 7.47

19 34

1.7 1.4

* Mean total cholesterol, in mmol/L, at baseline. t Relative risk reduction is the percent reduction in the rate of cardiac death or nonfatal myocardial infarction in the treated as compared with the untreated groups. $ Absolute risk reduction is the difference between the rate of cardiac death or nonfatal myocardial infarction at the end of the trial period in the treated and control groups. § Lipid Research Clinics Coronary Primary Prevention Trial (6). H Helsinki Heart Study (7).


This is also illustrated by cholesterol and coronary artery disease. The Lipid Research Clinics Coronary Primary Prevention Trial and the Helsinki Heart studies are two examples of carefully done, randomized, controlled trials of the effects of cholesterol-lowering drugs on the subsequent occurrence of coronary artery disease (5-7). The results of these trials, in terms of both relative and absolute risk reduction, are outlined in Table 1. These data are often cited as evidence that lowering an individual's serum cholesterol with cholestyramine or gemfibrozil will reduce their risk of coronary artery disease. These trials, however, offer direct support for the application of such interventions only in the group directly studied (middle-aged men with very high serum cholesterol values). Paradoxically, the bulk of coronary mortality occurs in individuals with lower levels of cholesterol. In the Framingham Study, for example, only 20% of the coronary mortality was in individuals whose serum cholesterol measurements were in the highest 10% of the population (8-10). More recently, it has been estimated that " . . . about 70% of new coronary heart disease cases in men and about 50% of new cases in women may be in persons with cholesterol levels of less than 6.47 mmol/L (250 mg/dL)" (11). Yet it is in just this segment of the population, with low to moderate cholesterol levels, that the effectiveness of interventions to reduce the incidence of coronary disease has not been evaluated. Similarly, although the effectiveness of screening mammography has only been demonstrated in women over the age of 50, several groups, including the American Cancer society and the National Cancer Institute, have recommended that yearly mammograms begin between the ages of 35 and 40 (12). Resolving the Policy Paradox: Assumptions The dilemma caused by the policy paradox is whether the scientific evidence should be interpreted to support interventions in broadly defined populations or to support them only in those segments of the population that have been directly studied. One authority has referred to this dilemma as an ideologic dispute between "evangelists" who support a broad application of available evidence and "snails" who favor a strict interpretation (13). The NCEP is an example of a policy created by evangelists. These policy makers have extrapolated the evidence by assuming that a benefit for reducing cholesterol exists at lower levels than those directly studied and by recommending interventions for younger and older men and for women. If these assumptions about extrapolation are correct, a strategy like this can be shown to be extremely effective in reducing the incidence of coronary disease (11, 14). Despite these calculations of effectiveness, other policy makers, like the Toronto Working Group, may assume that available experimental data should be interpreted more strictly. Resolving the paradox involves an appreciation of this elasticity of the evidence. The degree to which data are interpreted broadly or narrowly reflects a set of assumptions that are often not made explicit. These assumptions consider the type and quality of available evidence, including its biologic plausibility, the soci-

1 April 1992 • Annals of Internal Medicine • Volume 116 • Number 7

Downloaded from by Karolinska Institute user on 01/18/2019

ety's perception of and time preference for risk, and the society's choices about resource allocation. The Type and Quality of Available Evidence For a preventive intervention to be recommended to any population, the first assumption made is that the evidence for its effectiveness is both internally and externally valid. Internal validity refers to the inherent truth of the evaluation and is assessed through a twostep process. First, the experimental method is evaluated with respect to a hierarchy of type of evidence in which randomized controlled trials are considered to be the most valid and expert opinion the least (15, 16). Second, the method is analyzed with respect to commonly held scientific standards of epidemiologic research; this is a determination of validity by quality (16). No matter how valid internally, experimental evidence can be directly used only to support the application of interventions to groups with a risk profile similar to the study population. As mentioned previously, because of issues in the feasibility of performing an experiment, these directly studied populations will tend to be narrowly defined. The extent to which extrapolation to other groups is appropriate reflects the external validity of the evidence. This, in turn, is a function of the biologic plausibility and efficiency of the intervention. Biologic plausibility relates the effect of the intervention to the biologic theory of the disease. With coronary artery disease, for example, the pathogenic role of cholesterol has been confirmed by decades of experimental work (18). Lowering serum cholesterol is, therefore, an intervention with high biologic plausibility and, because of this, arguments have been advanced that evidence for effectiveness of cholesterol-lowering obtained in men with high levels may be reasonably extrapolated to men at lower levels. Similarly, biologic plausibility has been used as an argument to support extrapolations to men of other ages and races and to women. Another component of external validity is the practical efficiency of an intervention (17). To be efficient, an intervention must not only be effective in the studied populations, it must also be feasible to apply it to other groups in a proficient manner and for an adequate duration. Finally, it must be acceptable and affordable to individuals. Other decisions about generalization require assumptions about societal perceptions and preferences and will be the topic of the remainder of this discussion. Risk Perception and the Use of Evidence For an individual or a society to adopt preventive interventions, it must want to avoid the disease. If the society is assumed to be very averse to the risk for a particular disease, policy makers may be more willing to extrapolate the evidence. Assumptions about societal and individual risk aversion are, therefore, very much a part of the process involved in resolving the paradox. The epidemiologic definition of risk used thus far is the probability that an individual will develop a disease. From the perspective of individual or societal perceptions of risk, a more appropriate definition is that found

in a standard dictionary, " . . . the possibility of suffering harm or loss; danger" (19). Individual behavior is based considerably on intuitive judgments about the perceived likelihood and significance of harm which may have little relation to epidemiologic risk (20). The perception of risk is an instinctive reaction that is affected by several features of the adverse outcome including knowledge about and dread for it, controllability of the circumstances surrounding it, and whether or not it results from a voluntary activity. For the responsible health policy maker, however, the choice of targets for preventive interventions will be based, in large part, on measures of epidemiologic risk. Perceived risk, on the other hand, may be useful as a framework on which decision makers base their assumptions about evidence. One of the inherent characteristics of preventive interventions is that more people will undergo the intervention than would have developed the disease (21). If a disease has a high level of perceived risk, no matter what the epidemiologic risk for the disease or the evidence for specific effectiveness of the intervention, individuals who adopt the intervention can gain a psychologic benefit by lowering their perceived risk. Under this condition of high-risk perception, policy makers may be more willing to extrapolate the available evidence. This may have been the case with the NCEP because concern about cholesterol is a well-recognized national characteristic in the United States. If a disease has a low perceived risk, on the other hand, there are few psychologic benefits from adopting the intervention. In this case, policy makers may be inclined to reserve interventions for populations at highest epidemiologic risk or those for whom there is specific evidence of effectiveness. Alternatively, they may attempt to increase the awareness or perception of risk for the disease in the general population; this seems to have been the strategy adopted by the National Cancer Institute in its efforts to increase self-examination of the breast. The Time Preference Assumption Policy makers must also temper their assumptions about risk aversion and the use of evidence with a consideration of time preference. Interventions, after all, are made to forestall or mitigate a future outcome, and individuals and groups may prefer their outcomes to occur at different times. In other words, in order to adopt a preventive intervention, individuals or groups must prefer avoiding or delaying the outcome to the discomfort, inconvenience, and expense of the intervention itself. Each individual and each society will have a different time preference; some will be willing to undergo substantial early discomfort to avoid or postpone disease. Others will not be willing to undergo any present discomfort for future benefit, and many will fall in between (22). The NCEP guidelines, for example, assume that individual members of the society are willing to sacrifice some comfort in the present time (be it the discomfort and time involved in having cholesterol measured or the expense and nausea involved in taking cholestyramine) in order to delay coronary artery disease (23). If this

1 April 1992 • Annals of Internal Medicine • Volume 116 • Number 7 Downloaded from by Karolinska Institute user on 01/18/2019


time preference could not be assumed, either the evidence would have to be interpreted more strictly or the level of perceived risk would have to be higher in order for the paradox to be resolved in favor of such a broadly applied intervention. Social Considerations The previous discussion has outlined a process whereby policy makers make assumptions that allow them to interpret available evidence on preventive interventions and apply them to their society. Two additional assumptions are relevant to this process. The first is that the disease is considered important enough that the preventive efforts are preferred to the available alternatives, such as treating future cases of acute disease, devoting resources to finding better interventions for the same disease, diverting resources to another disease, or diverting resources to another sector of the society altogether (20). The potential benefits of an intervention in the few destined to develop the disease (as well as psychologic benefits for those with high levels of perceived risk) must be balanced against the costs and harms of the strategy to all those exposed to it (21). The second social consideration is also related to resource allocation. In many societies health policy is recommended by organizations that are not responsible for allocating resources. In this circumstance, any single group is best served by assuming its problem to have an exaggerated social importance. In doing so, it is able to justify broader and more elaborate intervention policies and, thereby, to divert more attention and resources to itself and its strategies. A similar problem has been described in the Tragedy of the Commons, a paradigm used to explain individual decision making when social resources are limited (24, 25). Although this paradigm has been generalized to health policy, it was originally used to describe how individual herdsmen decide how many cattle to keep on a village commons of limited size: . . . the rational herdsman concludes that the only sensible course for him to pursue is to add another animal to his herd. And another; and another. . . . But this is the conclusion reached by each and every rational herdsman sharing a commons. Therein is the tragedy. Each man is locked into a system that compels him to increase his herd without limit—in a world which is limited. Ruin is the destination toward which all men rush, each pursuing his own best interest in a society that believes in the freedom of the commons. Freedom in a commons brings ruin to all (24). Similarly, every special interest group will attempt to divert as many resources to itself as possible. Faced with this Medical Commons, policy makers must be aware that blind acceptance of strategies advocated by special interest groups may result in irrational policy for the society as a whole.

terventions. In attempting to understand one way this may occur we have described the policy makers' paradox, a circumstance in which the evidence for the effectiveness of an intervention is obtained in one segment of a population although the greatest disease burden occurs in another segment. On the basis of this paradox, we have described some of the common assumptions behind the interpretation and application of available data: first, a fundamental assessment of the type and quality of the scientific evidence; second, appreciation of the societal and individual risk perception and time preference; and third, understanding of the relevance of the disease to the society, including potential alternative uses of resources and the tendency for special interest to be confused with social interest. Although our approach has emerged from an analysis of health prevention strategies, it has broad relevance to other social problems. Regulations concerning air pollution and toxic waste disposal, for example, impose a high cost on business but are not believed to be particularly cost-effective and are often based on incomplete evidence for effectiveness (26). For a society to adopt such interventions in a rational fashion, it will implicitly follow the structure we have outlined for health policy decisions. In this discussion, we have not concentrated on the role of better evidence in refining estimates of risk and determining which groups are best served by preventive interventions. To wait for conclusive evidence on the effects of interventions in those with low to moderate levels of known risk factors will be to wait forever. Instead, we have endeavored to describe an implicit model behind the application of presently available evidence. In the setting of the policy makers' paradox, this evidence may be interpreted elastically. When perceived risk for a disease or other adverse outcome is low and the costs of an intervention high, policy makers may prefer to interpret the available evidence narrowly. Evidence may be viewed differently, however, as when the social and clinical contexts combine to favor a broad interpretation. We suspect that the different policy recommendations in the United States and Canada for cholesterol screening and treatment reflect this elasticity in the interpretation of evidence. Acknowledgments: The authors thank Dr. Alvin Tarlov and two anonymous reviewers for their advice. Grant Support: By the Henry J. Kaiser Family Foundation grant 873295 and by the John D. and Catherine T. MacArthur Foundation Research Network on the Determinants and Consequences of Health Promoting and Health Damaging Behavior. Requests for Reprints: Eric Brown, MD, Section of Nephrology, Department of Internal Medicine, Yale University School of Medicine, 333 Cedar Street, 2073 LMP, New Haven, CT 06510-8056. Current Author Addresses: Dr. Brown: Section of Nephrology, Department of Internal Medicine, Yale University School of Medicine, 333 Cedar Street, 2073 LMP, New Haven, CT 06510-8056. Drs. Viscoli and Horwitz: Department of Internal Medicine, Yale University School of Medicine, IE-61, P.O. Box 3333, New Haven, CT 06510-8025.

Conclusion Different policy-making groups may interpret the same evidence and define markedly different yet equally rational strategies for the application of preventive in596

1 April 1992 • Annals

of Internal


References 1. Garber AM. Where to draw the line against cholesterol. Ann Intern Med. 1989;111:625-7. 2. Report of the National Cholesterol Education Program Expert Panel

• V o l u m e 116 • N u m b e r 7

Downloaded from by Karolinska Institute user on 01/18/2019

3. 4. 5.




9. 10. 11.


on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Arch Intern Med. 1988;148:36-69. Kelsey JL, Thompson WD, Evans AS. Methods in Observational Epidemiology. New York: Oxford University Press; 1986:33-4. Petrakis NL. Genetic factors in the etiology of breast cancer. Cancer. 1977;39:2709-15. Garber AM, Sox HC Jr, Littenberg B. Screening asymptomatic adults for cardiac risk factors: the serum cholesterol level. Ann Intern Med. 1989;110:622-39. The Lipid Research Clinics Coronary Primary Prevention Trial results. II. The relation of reduction in incidence of coronary heart disease to cholesterol lowering. JAMA. 1984;251:365-74. Frick MH, Elo O, Haapa K, Heinonen OP, Heinsalmi P, Helo P, et al. Helsinki Heart Study: primary-prevention trial with gemfibrozil in middle-aged men with dyslipidemia. N Engl J Med. 1987;317: 1237-45. Kannel WB, Castelli WP, Gordon T, McNamara PM. Serum cholesterol, lipoproteins, and the risk of coronary heart disease. Ann Intern Med. 1971;74:1-12. Kannel WB, McGee D, Gordon T. A general cardiovascular risk profile: the Framingham Study. Am J Cardiol. 1976;38:46-51. Kannel WB, Castelli WP, Gordon T. Cholesterol in the prediction of atherosclerotic disease. Ann Intern Med. 1979;90:85-91. Goldman L, Weinstein MC, Williams LW. Relative impact of targeted versus populationwide cholesterol interventions on the incidence of coronary heart disease. Circulation. 1989;80:254-60. United States Preventive Services Task Force. Guide to Clinical Preventive Services. Washington: U.S. Department of Health and Human Services; 1989.

13. Sackett DL, Holland WW. Controversy in the detection of disease. Lancet. 1975;2:357-9. 14. Kottke TE, Gatewood LC, Wu SC, Park HA. Preventing heart disease: is treating the high risk sufficient? J Clin Epidemiol. 1988; 41:1083-93. 15. Canadian Task Force on the Periodic Health Examination. The periodic health examination. Can Med Assoc J. 1979;121:1193-54. 16. Feinstein AR. Scientific standards in epidemiologic studies of the menace of daily life. Science. 1988;242:1257-63. 17. Cochran AL. Effectiveness and Efficiency. Abingdon: Burgess and Son; 1972. 18. Ross R. Factors influencing atherogenesis. In: Hurst JW; ed. The Heart 6th ed. New York: McGraw-Hill; 1986:801-16. 19. Morris W; ed. The American Heritage Dictionary of the English Language. Boston: Houghton Mifflin; 1969. 20. Slovic P. Perception of risk. Science. 1987;236:280-5. 21. Russell LB. Is Prevention Better Than Cure? Washington, DC: The Brookings Institution; 1986. 22. Lipscomb J. Time preference for health in cost-effectiveness analysis. Med Care. 1989;27:S233-53. 23. Brett AS. Treating hypercholesterolemia. How should practicing physicians interpret the published data for patients? N Engl J Med. 1989;321:676-80. 24. Hardin G. The Tragedy of the commons. Science. 1968;162:1243-8. 25. Hiatt HH. Protecting the medical commons: who is responsible? N Engl J Med. 1975;293:235-41. 26. Rhodes SE. The Economist's View of the World Cambridge, England: Cambridge University Press; 1985.

1 April 1992 • Annals of Internal Medicine • Volume 116 • Number 7 Downloaded from by Karolinska Institute user on 01/18/2019


Preventive health strategies and the policy makers' paradox.

The likelihood of developing many diseases is predicted by levels of risk factors. Many public health strategies have been created to apply interventi...
1002KB Sizes 0 Downloads 0 Views