Tobacco Documents Online

a I I I I I I I LOW RISK EPIDEMIOLOGY AND GOOD EPIDEMIOLOGICAL PRACTICE Raanar Rylander Department of Environmental Medicine, University of Gothenburg, Gothenburg, Sweden Abstract This presentation reviews the methodological difficulties involved in low risk epidemiology. Important basic concepts relate to dose-response relationships in terms of threshold or the J-shaped curve. The possible errors in establishing exposure estimates are outlined, particularly in terms of dose descriptions, and good epidemiological practice is discussed. Finally, the responsibilities of the researcher in terms of the caution necessary in the interpretation of data, as well as the public health impacts of those interpretations, are delineated. Background I I I I I , I Investigations of risk factors are an important part of scientific efforts to assess relations between the environment and risks for disease. Risk assessment has a long tradition. As early as 3200 BC, the Sumerians had special priests - the Asipus - whose role was to evaluate risks. They aided kings, governors or individuals in evaluating risks, using a simple mathematical system based on yes and no. The sum of these directed marriage alliances, the purchase of property and other everyday events in society. Risk estimations are equally important today as during Sumerian times, and the results are still expressed in numbers. However the priests of our times are epidemiologists, toxicologists and statisticians, and the procedures followed to arrive at numbers for risk have become very complicated. Our risk estimations now deal with low numbers which adds to the complexity. Most of the large risks related to environmental agents have been defined and described - with an acknowledgment of an absence of preventive measures for some of them. In striving for good health, our attention has increasingly been directed toward low risk agents in the environment. The purpose of this presentation is to discuss some of the methodological difficulties related to studies of low risk agents and the interpretation of results, and to delineate some suggestions for good scientific practice in evaluating results. Low Risk Agents I I I Problems in low risk epidemiology have been dealt with in several publications and workshops (14,22-24,27). Wynder defined an increased low risk as up to 2 and a decreased low risk as down to 0.5 (24). The concept of the appreciation of risks at different levels is shown in Table 1. I

Table 1. Assessment of different degrees of risk Risk Discovery >10 9-2 <2 Perceived by the population itself Relation to exposure relatively easily established with epidemiological techniques Severe methodological problems The table illustrates that high risks are appreciated without the interference of scientists - they may be required to give precise figures, but will not influence the general appreciation of the risk involved. A good example is the common knowledge, as cited by the Swedish scientist Linneaus, that exposure to dust was a major cause of death among granite workers in Dalecarlia in Sweden in the 16th century. The table further illustrates that the detection of low risks requires the involvement of well trained epidemiologists. Studies of this nature have become further complicated during recent years, by developments in toxicology having to do with new principles for dose-response relationships and an increased understanding of mechanisms for the development of diseases induced by environmental agents. These problems will be treated in the following. Dose-response Relationships The traditional concept of a dose-response relationship as it relates to environmental agents was a linear curve on which even small doses were shown to cause an effect. This concept was applied particularly to radiation and carcinogens, and it allowed toxicologists to work with high doses in experimental settings. Estimations of risks from low level exposures could be made from experimental and epidemiological observations of high dose levels. It is now becoming increasingly clear that two other relationships - the threshold concept and the J-shaped dose relationship - are more appropriate for describing the human reaction to environmental agents. The three different concepts for dose-response relationships are illustrated in Figure 1. a e e 05 Figure 1. -2- I I I I I I I I I I I I I I I I I I

I I I I I I I I I I I I I I I The J-shaped curve is of particular interest for environmental exposures. It implies that a small dose of an agent decreases the risk, as compared to no dose at all. At higher dose levels, risk increases appear. The concept is illustrated by the following examples. Vitamin A in foods is a necessary nutritional item, whereas the intake of high doses of vitamin A is toxic and involves a risk for liver cancer. Alcohol in moderate doses decreases the adherence of platelets and reduces the risk for cardiovascular disease. At high levels, alcohol is toxic and increases the risk for cardiovascular death. The J-shaped dose-response curve probably also relates to air pollution. Low levels of irritation, such as are caused by respiratory infection or air pollution, seem to decrease the risk for IgE-related sensitization to inhaled allergens. There are also data suggesting that the risk for lung cancer is reduced by exposure to inflammatory agents. The Balance Concept The balance concept is an important consideration concerning the relationship between the environment and disease development. It is now understood that the development of disease caused by environmental agents seldom follows a direct cause-effect relationship. The body has a series of defense systems which deactivate many environmental agents or their metabolites. Particularly important is the P 450 enzyme system. Paradoxically, it may indeed be suggested, that of all the defense against the alien agents in our environment, the best ones are against carcinogenic substances, the reason being the large number of natural carcinogens that is present in the normal environment or produced in the body itself by bacterial metabolism in the gastrointestinal tract. An important part of the balance system is the defense brought about by foods. It is generally agreed that fruit and vegetables are important protective factors against cancer (1,3,4,8,10) and probably other diseases, such as atherosclerosis. A trace element such as selenium is also important, mainly in its capacity as an antioxidant. Implications for Low Risk Epidemiology A consequence of the dose-response and balance concepts referred to above is that low risk epidemiology must take into consideration the potential beneficial effects of a particular exposure as well as the presence or absence of protective factors. Particularly dangerous is the situation in which the factors influencing the risk covary with the agent studied. An example is smokers, from whom the exposure to tobacco smoke according to cigarettes per day is inversely related to the consumption of such protective factors as vegetables and fruit (12,13,19). Dose Errors The potential errors in low risk epidemiology are not different from those in epidemiology in , general, but there is a need for high precision in view of the normal random variation in a studied material. Extensive reviews of possible errors have been presented previously, in particular during a workshop reported by Wynder (23). Here remarks will be limited to the exposure description. N 00 i V O W , 3 V O I

Exposure determinations can be made using questionnaires or biological markers. Biological markers describe susceptibility, internal dose or the biological effect. A major advantage of biological markers is that they reduce the risk for misclassification, which is a particularly important source of error in low risk epidemiology. However, biological markers are not available, for most substances particularly in relation to long term exposure, and questionnaires offer the only possible method for dose determination. As for study design, case-control studies are often the sole alternative, as exposure descriptions are poor or nonexistent in most health registers. Criteria for exposure assessment in case-control studies have recently been reviewed (6). When dealing with high risk factors, the dose description is less critical. While an exposure estimation error may cause the risk to vary between e.g. 7.8 and 8.9, the conclusion will be that an exposure is related to a substantial risk. For low risk agents, this error becomes crucial. This is illustrated in Figure 2. 20 dsk 15 10 0 ; i I I _-._ I . 0 3 6 9 dose Figure 2. An observed low risk on the borderline of statistical significance may entirely be the result of a poor exposure description (overestimation of dose) and the correct conclusion is that no risk exists. Alternatively, if the dose is underestimated, the risk may be larger and statistically significant. The problem increases in complexity when several risk factors are involved. Tobacco smoke, coke oven emissions, radon, asbestos, keeping pet birds and ETS have been identified as risk factors for lung cancer. They cause a risk at different levels, but nonetheless cause risk for the same disease. As several of these factors are interrelated, the dose for all of them must be described with equal precision to arrive at a conclusion concerning the risk for an individual factor. -4- I 1 I I I I ' I I I ' I I I

a I I I I I I I I I I I I I I I I I It is particularly important to describe the exposure to a significant risk factor before analytical work on small risk factors is undertaken. An example of the need to control for the major risk factor can be taken from studies on ETS exposure and cardiovascular disease. Support for the hypothesis of covariation has been found in some epidemiological studies. Only one has controlled for dietary fat intake however, which is a major risk factor for cardiovascular disease (9). When fat intake was corrected for, the statistical significance disappeared in all but one study. Another illustration is the study of lung cancer and exposure to diesel exhaust among truck drivers as an occupational group. A slightly increased risk has been demonstrated, for both lung cancer and cardiovascular disease. The common interpretation is that these effects are caused by exposure to diesel exhaust. Recently, Alavanja et .(1) showed that fat is an important risk factor for lung cancer, with odds ratios as high as 11 in nonsmoking females. Fat is also a well-known risk factor for cardiovascular disease. In view of this, it can be hypothesized that the increased risk for lung cancer and cardiovascular disease among truck drivers is not due to the relatively low exposure to diesel exhaust but rather to dietary habits - eating high-fat foods at cafeterias during night shifts. Not until this factor has been controlled for can a final evaluation of the previous hypothesis be made. Paradigm Bias In the interpretation of low risk relationships, particularly those "on the borderline" of statistical significance, paradigm bias is important. An overview of the scientific literature suggests that a ruling paradigm, itself often based on only a small material, is difficult to overcome. The resistance is found among reviewers of journal, editors, colleagues and the public. This paradigm defense is particularly important for studies which report a negative finding in the perspective of a previous positive finding. Paradigm bias is also reflected in the number of articles that appear confirming a new result. When the data on cold fusion were published, there was initially quite a number of studies confirming the original observations. Not until several months later did the negative studies begin to dominate. Paradigm bias is also present in the judgment of data that do not support the paradigm. Such data are referred to as "strangely different," "based on different populations," resulting from poor techniques or simply ignored. A good example of paradigm bias is studies which cannot demonstrate an association between exposure to environmental tobacco smoke (ETS) and lung cancer. One of the first negative studies on ETS stems from Hong Kong, where a thorough investigation of the subjects was made that incorporated lifestyle factors such as consumption of vegetables (11). Comments on this study are that conditions in Hong Kong are not relevant for other societies when, in fact, a deviating set of data from a different society may have a much larger potential to assess the general validity of a fii nding for the very reason that conditions are different (29, 30). Good Epidemiological Practice In toxicology, the concept of good laboratory practice has existed for many years. These are a set of rules for the structure of the investigation, responsibility and quality control. While these rules do N 0 they represent an important basis for quality not exclude the possibility of errors or even frauds 00 , .a 5 I

assurance, data interpretation and verification. Among the different conditions, the request for data storage in a form that would make a later reanalysis possible is particularly important. Similar rules have been discussed in epidemiology and proposals have been made by different bodies including WHO (5, 16, 17, 18, 20), but no generally accepted standards have yet been presented. There is a need for this in low risk epidemiology. Even if such rules may initially hurt the pride of or seem self-evident to the knowledgeable scientist, experience gained in toxicology has demonstrated their usefulness. Table 2 illustrates some of the basic concepts to be defined in rules for good epidemiological practice. Table 2. Basic principles for good epidemiological practice Defined organizational structure Defined principal investigator Personnel qualified or trained for study Establishment of study plan Documentation of collected data Appropriate storage of data for later reanalysis Responsibility of the Researcher Put in one perspective, the sound use of epidemiological techniques remains the responsibility of the researcher. Initially, it is paramount to realize that analytical epidemiology is actually the weakest link in a chain of evidence relating an exposure to a disease (26). It is essential to consider evidence from other studies, including toxicology, exposure assessment and molecular biology. Associations found in epidemiological studies are not a proof of causality and the researcher should be aware of the many pitfalls involved in his own interpretation of his data as concerns causality. Wish bias is an important error in the interpretation of results of epidemiological studies (25), particularly in studies of low risk agents. The researcher is also responsible for the use of his data in public health practice. An overuse of some preliminary results or data supporting a paradigm hypothesis is not only unethical but also approaches a scientific fraud. Increasing the importance of small findings by multiplying a low risk with the number of persons in a population must only be done when the evidence is good and with careful caveat as to the uncertainties involved. Another example of erroneous reporting of results is a recent study on respiratory infection in children and different risk factors (2). The authors found an increased risk for exposure to ETS but when -6- I I I I I I I I I I I I I I I

 ' I I the data was corrected for day-care attendance, an important risk factor for respiratory disease in children (7), the risk disappeared. In spite of this, the summary contains a statement regarding the risk for infection resulting from ETS. Another responsibility of the researcher is to acknowledge his own limited knowledge. Understanding of the complex relation between environmental agents and disease develops continuously and what are today accepted ideas, such as the importance of diet for the risk of disease and the concept of special risk individuals, were not known some decades ago. Conclusion I I I I I I , I I I With prudence, caution and good epidemiological practice, epidemiology can, in spite of its inherent methodological problems, bring new knowledge to the understanding of disease and environment, with corresponding gains in preventive power and population health. Without these precautions, epidemiology can bring chaos and in the end, a mistrust of public health and environmental medicine. This would obviously bring about a severe delay in the progress in the field of prevention, a critical field in increasing the health of the population. -7- I

I References 1. Alavanja, MCR; Brown, CC; Swanson, C and Brownson, RC. Saturated fat intake and lung cancer risk among nonsmoking women in Missouri. J. Natl. Cancer Inst. 1993; 85:1906-1916. 2. Berg, AT; Shapiro, ED; and Capobianco, LA. Group day care and the risk of serious infectious illness. Am. J. Enidemiol. 1991; 133:154-163. 3. Bjelke, E. Dietary vitamin A and human lung cancer. Int. J. Cancer 1975; 15:561-565. 4. Block, G; Patterson, B and Subar, A. Fruit, vegetables, and cancer prevention: A review of the epidemiological evidence. Nutr. Cancer 1992; 18:1-29. 5. Chemical Manufacturers' Association Epidemiology Task Group, Guidelines for good epidemiology practices for occupational and environmental epidemiology research. J. Occun. Med. 1991; 33:1221-1229. 6. Correa, A; Stewart, WF; Yah, H and Santos-Burgoa, C. Exposure measurement in case-control studies: Reported methods and recommendations. Enidemiol. Rev. 1994; 16:18-31. 7. Dahl, IL; Grufman, M; Hellberg, C and Krabbe, M. Absenteeism because of illness at day care centers and three-family systems. Acta Pediatr. Scand. 1991; 33:1221-1229. 8. Fontham, E. Protective factors and lung cancer. Int. J. Epidemiol. 1990; 19:24-31. 9. He, Y; Lam TH; Li LS; Du RY; Jia GL; Huang JY and Zheng JS. Passive smoking at work as a risk factor for coronary heart disease in Chinese women who have never smoked. Br. Med. J. 1994; 308:380-384. 10. Kant, AK; Block, G; Schatzkin, A and Nestle, M. Association of fruit and vegetable intake with dietary fat intake. Nutr. Res. 1992; 12:1441-1454. 11. Koo, LC; Ho JH; Saw D and Ho CY. Measurements of passive smoking and estimates of lung cancer risk among nonsmoking Chinese females. 1nL J. Cancer 1987; 39:162-169. 12. Margetts, BM and Jackson, AA. Interactions between people's diet and their smoking habits: the dietary and nutritional survey of British adults. Br. Med. J. 1993; 307:1381-1384. 13. Morabia, A and Wynder, EL. Dietary habits of smokers, people who never smoked and ex- smokers. Am. J. Clin. Nutr. 1990; 52:933-937. I I I I I 1 I I I I I , I I ~ 14. Rylander, R; Lebowitz, M and Peterson, Y. Assessing low risk agents for lung cancer: ~ methodological aspects. Int. J. Epidemiol. 1990; 19:S2-S87. ~. iV ~ W K) 8 v ' U I

I I I 15. I 16. 17. I 18. I 19. I 20. I 21. I 22. I 23. I 24. I 25. I 26. , 27. I 28.  29. ' , ' Rylander, R. Environmental exposures with decreased risk for lung cancer? Int. J. Epidemiol. 1990; 19:S67-S72. Szklo, M. Design and conduct of epidemiologic studies. Prev. Med. 1987; 16:142-149. Stellman, SD. Confounding. Prev. Med. 1987; 16:165-182. Schlesslman, JJ. "Proof' of cause and effect in epidemiologic studies: criteria for judgment. Prev. Med. 1987; 16:195-210. Thompson, DH and Warburton, DM. Lifestyle differences between smokers, ex-smokers and nonsmokers and implications for their health. Psychol. Health 1992, 7:311-321. WHO. Guidelines on studies in environmental epidemiology. Environmental Health Criteria 1983; nr 27:1-351. Von Mutius, E; Martinez, FD; Fritsch, C; Nicolai, T; Roell, G and Thiemann, H-H. Prevalence of asthma and atopy in two areas of west and east Germany. Am. J. Resn. Crit. Care Med. 1994; 149:358-364. Wynder, EL. Weak associations in epidemiology and their interpretation. Prev. Med. 1982; 11:464-476. Wynder, EL. Workshop on guidelines to the epidemiology of weak associations. Pre. Med. 1987; 16:139-141. Wynder, EL. Guidelines to the epidemiology of weak associations. Pre. Med. 1987; 16:211- 212. Wynder, EL.; Higgins, ITT and Harris RE. The wish bias. J. Clin. Epidemiol. 1990; 43:619- 621. Wynder, EL.; Cohen, LA; Rose, DP and Stellman, SD. Dietary fat and breast cancer: where do we stand on the evidence? J. Clin. Epidemiol. 1994; 47:217-230. Wynder, EL. Epidemiological issues in weak associations. Int. J. Enidemiol. 1990; 23:1491- 1496. Wynder, EL; Taioli E and Fujita Y. Ecologic study of lung cancer risk factors in the U.S. and Japan, with special reference to smoking and diet. Japan J. Cancer Res. 1992; 83:418-423. Wynder, EL and Stellman, SD. The "overexposed" control group. Am. J. Epidemiology 1992; 135:459-461. N C 00 ~ ti 0* -9- N 4 6>