Click for next page ( 96


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 95
F RON ECOLOGIC CORRELATIONS TO METABOLIC EPIDEMIOLOGY: THE LINK WITH NUTRITION Laurence N. Kolonel The strengths of epidemiology for the study of the etiology of human disease are increasingly being recognized in the scientific community. Epidemiology not only focuses on humans, the species of interest, but also examines the effects of human exposures in their natural context. This means that epidemiologic analyses are necessarily based on the actual routes and levels of exposure occurring among the individuals at risk, can consider the impact of other interacting variables, and usually, encompass a sufficiently wide range of exposures to permit dose-response assessments. With regard to diet and disease relationships specifically, the epidemiologic approach has certain salient advantages: . Epidemiologic studies can take into consideration some of the complexities of the human diet, including food sources, varieties, combinations, methods of preparation and preservation, temperatures, and additions (e.g., spices) that are unique to human populations and that may influence disease risk. These patterns are not easily reproduced in animal models. Dietary factors may interact among themselves or with other variables (e.g., exercise, medications, smoking, and other social habits). Epidemiologic studies have the potential to uncover such relationships, which would not likely be identified in animal studies. ~ Because epidemiologic studies deal with actual levels of exposure, they can indicate the true magnitude of the effect of the dietary behavior and, therefore, its real public health significance. Thus, they are the primary basis for meaningful risk assessments among individuals. 95

OCR for page 95
Since epidemiologic studies are carried out in free-living populations, they can identify the particular behavioral and sociocultural characteristics associated with diet-related illnesses, and thus can suggest practical and appropriate approaches to primary prevention. These approaches can be tested in well-controlled intervention trials before being implemented in the general population. Although epidemiologic studies provide the essential information on etiology, they are very limited in their ability to elucidate the mechanisms involved in pathologic changes. It is this latter aspect of the study of disease that is best carried out under controlled conditions in the laboratory with animal or human subjects. However, it is worth noting that precise knowledge about mechanisms is not essential to identify causal factors or to implement control measures that will prevent disease. IMPORTANCE OF DIET IN CHRONIC DISEASE ETIOLOGY Diet and nutrition are major factors in the risk of several chronic diseases. The two leading causes of death in the United States today, making up about 60% of all deaths, are chronic conditions: heart disease and cancer (Table 1) (Silverberg and Garfinkel, 1987~. TABLE 1 Ten Leading Causes of Death in the United States, 1984 Mortality Rate/100,000 % of Total Cause of Death Population Deaths Heart disease 269.9 37.3 Cancer 170.7 22.2 Cerebrovascular disease 52.8 7.6 Accidents 36.3 4.6 Pneumonia/influenza 19.8 2.9 Chronic obstructive 17.4 2.4 lung disease ~ Diabetes mellitus 13.0 1.8 Suicide 11.3 1.4 Hepatic cirrhosis 11.0 1.3 Arteriosclerosis 7.8 1.2 SOURCE: Data from Silverberg and Garfinkel (1987) 96

OCR for page 95
~- Through its influence on at least two primary risk factors for myocardial infarction, serum cholesterol level and obesity, diet makes a major contribution to the nigh incidence of coronary heart disease. Scientific data also suggest, although the evidence is less firm at present, that diet plays a significant role in the etiology of several leading cancers in the United States, including cancers of the large bowel, breast, and prostate, as well as several less frequent cancers, such as the stomach and pancreas (National Research Council,- Committee on Diet, Nutrition, and Cancer, 1982~. In addition, diet is a significant contributing factor to several other important chronic diseases in the United States, including hypertension, osteoporosis, diabetes mellitus, and dental caries. Thus, diet or nutrition is likely to be the major life-style factor affecting morbidity from chronic disease in the United States today, although the overall impact of cigarette smoking is better established at present. DEVELOPMENT OF NUTRITIONAL EPIDEMIOLOGY WITH RESPECT TO CHRONIC DISEASE ETIOLOGY The conclusion given above indicates that adequate assessment of dietary patterns and individual intakes is an essential part of much epidemiologic research into the causes of the major chronic diseases. Methods for conducting this research have been evolving continuously, but, as the following discussion shows, refinements are still needed. Ecologic Correlations Some of the earliest leads in the nutritional epidemiology of chronic disease came from simple correlations of food disappearance data with disease mortality rates. For example, researchers showed that mortality from heart disease in various countries was highly correlated with per capita consumption of dietary fat (Figure 1) (Jolliffe and Archer, 1959; McGandy et al., 1967~. More recently, attention has been turned to cancer, and a similar high correlation has been seen for per capita fat consumption and cancers of the breast and large bowel (Carroll, 1975; Knox, 1977~. 97

OCR for page 95
Such correlations suffer from the imprecision of food disappearance data as measures of food intake by individuals, not only because these data fail to account for such variables as storage and household waste, but also because they do not usually distinguish among even such heterogeneous groups as men and women or adults and children. Some investigators have attempted to overcome certain of these limitations in their ecologic studies. For example, Kato et al. (1987) used data from actual household surveys in Japan to correlate food and nutrient consumption with mortality from stomach and large bowel cancers in 12 geographic areas. Among their findings were positive associations in both sexes of stomach cancer with vitamin A intake, colon cancer with cheese intake, and rectal cancer with protein intake. In Hawaii, Kolonel and colleagues (1981) obtained diet histories on representative samples of the population and overcame several limitations of ecologic analyses by correlating nutrient intakes by sex, age, and ethnicity with corresponding cancer incidence rates. They found strong associations between dietary fat, especially saturated fat, and cancers of the breast, prostate, and endometrium (Table 2 and Figure 1) (Kolonel et al., TABLE 2 Significant Correlations of Mean Daily Lipid Intake and Cancer Incidence among Age-, Sex-, and Ethnic-Specific Groups in Hawaii Cancer Total Site Fat Correlation Coefficients Saturated Unsaturated Fat Fat Cholesterol Breast 0.94 0.98 Corpus uteri Prostate Lung Larynx 0.95 0.90 1.00 0.95 0.87 0.94 0.76 aPartial correlation coefficent, adjusted for sex where appropriate. 98

OCR for page 95
c - ~ on :' - ~u c :~ I 1 1 1 U' 0~) C ~ .,' C . ~ o C .,, CL a. . - 0 ,., , , , , , , , , 2 - - Y 0 4 C _. _ a' . C 0 o o o o o {U ~ o o o ID ~ uo;;el ndod 000 'oO; Had aaUepT aUI ' (ASH age) pa~eaunu1 99 0. of Go Id I O 0 L ,' ~ _ ~ d 0 ~ h

OCR for page 95
1981~. Because of the availability of a population-based tumor registry in Hawaii, they had the added advantage of working with morbidity rather than mortality data, which eliminated any concern that differences in survival, influenced by treatment and other factors, may have accounted for the observed correlations. While ecologic studies based on individually collected intake data, wish'' attention given"to "sex, age, a'''nd other' possible confounding variables, represent an improvement of the per capita estimates from food disappearance data, the problem remains that the exposure and outcome groups are separately identified and may not overlap (ecologic fallacy). Thus, these studies have generally been thought of as hypothesis-generating only, needing confirmation with more analytic study designs. However, because of methodology limitations in the assessment of individual diets (discussed below), there may be good reason to place greater confidence in the results of ecologic than of more analytic types of studies. _ Case-Control and Cohort Studies Studies based on diet histories. To explore particular dietary hypotheses, many epidemiologists have carried out case-control studies in which subjects were asked for their past intake of selected food items or of the total diet. These studies often relied on frequency data only (how often an item was eaten), and if nutrients needed to be estimated, standard portion sizes were assumed. For example, Lubin~et al. (1981) compared the dietary habits of breast cancer cases and controls in western Canada based on a brief food-frequency questionnaire. They reported higher intakes of beef and pork and, using fixed portion sizes, of animal fat by the cases, in support of the hypothesis that fat is a risk factor for breast cancer. In order to refine this method, other investigators have used certain props, such as geometric or plastic food models and colored photographs (Hankin, 1986; Morgan et al. and Jain, 1978'), to estimate portion sizes. We '' prefer this approach, because it allows for greater variability in estimates and because usual portion sizes can differ substantially in heterogeneous groups such as the multiethnic population of Hawaii. For example, 100

OCR for page 95
typical portion sizes of meat are much smaller for Filipinos than for Caucasians living in Hawaii, leading to greater differences in actual meat intake between these groups than would be estimated from frequency data alone. Nonetheless, at least in some other settings, there appears to be little difference in the findings regarding diet-disease associations whether frequency-or frequency plus portion size is used. For example Humble et al. (1987) found similar odds ratios for the association of vitamin A intake with lung cancer among whites in New Mexico, a relatively homogeneous group, whether they used frequency data alone or frequency modified by portion size. There are a number of additional limitations to the diet history method as applied in epidemiologic research. A major concern is the ability of individuals to recall accurately their food intakes in epidemiologic inquiries. Although several attempts have been made to assess the validity of such recall, with generally encouraging results (Block, 1982; Byers et al., 1987a,b; McKeown-Eyssen et al, 1986; Rohan and Potter, 1984), many investigators nevertheless remain skeptical about the reliability of such information. At best, one assumes that such recall data must reflect substantial measurement error (systematic as well as random), such that the diet-disease relationships are attenuated, in some instances, to the extent that true associations of modest magnitude may be missed altogether (Prentice, in press). Errors of recall are made more likely by the fact that the diet in such developed countries as the United States is extremely varied in the number and nature of the foods eaten. Thus, the within-subject variability can actually exceed that between individuals. It has been reported, for example, that because of such variation in intake, one would need more than 6 weeks of daily food consumption records in order to classify 80% of individuals into only their appropriate fertile of intake of vitamin A (James et al., 1982~. This observation points out one of the potential strengths of the diet history method, however. When subjects recall their usual intake of particular foods (for diet history, in contrast to a 24-hour recall or a 7-day food record, for example), they are actually accounting for this - day-to-day variability in consumption by providing an 101

OCR for page 95
average or representative intake. Such a diet history also permits the investigator to cover a wider spectrum of foods, thus providing a better reflection of intake than can be achieved from only a few days' records. Although the more quantitative approach to the dietary or ~ history, that is, using variable portion size estimates, would seem to be desirable, this additional recall information may also add to the error in measurement, thereby lessening somewhat the gain in accuracy. This could be a reason for the failure of the New Mexico study (Humble et al., 1987) to find any significant difference in the nature and magnitude of the effect of vitamin A on lung cancer risk, whether variable portion sizes were included or excluded from the analysis. While this is only speculation at present, if confirmed, it would provide additional evidence that improved methods of quantification need to be developed. In most instances, cohort studies are considered to be preferable to case-control studies, because they are less subject to bias (Lilienfeld and Lilienfeld, 1980~. In dietary research, however, they often lack the representativeness of the case-control study, in that the intake information is usually based on consumption over a brief period surrounding the time of initial data collection (in some studies only 24 hours), which may not be even the most relevant period of life in terms of subsequent disease incidence. Furthermore, secular changes in the diet are usually not obtained in such groups, because of the considerable costs entailed in efforts to make repeated contact with the subjects to obtain additional information. Despite the hazards of the diet history method, it has proven to be a successful tool in epidemiologic research. For example, as estimated from a diet history, saturated fat intake showed a significant association with serum cholesterol level in middle-aged American men, and unsaturated fat and dietary cholesterol showed significant associations (the former inverse) with coronary heart disease mortality (Shekelle et al., 1981b). Similarly, several studies using the diet history method have shown that the consumption of dietary sources of carotenes is inversely associated with the risk of lung cancer, especially among smokers (Table 3) (Bond et al., 1987; Byers et al., 1987; Hinds et al., 102

OCR for page 95
TABLE 3 Epidemiologic Studies in the United States Showing an Inverse Association between Lung Cancer and Dietary or Serum Carotene Year of No. of Study Subjects (Principal Location Type of (Cases/ Sex of Author) of Study Study Controls) Subjects 1981(a,b) Illinois Cohort 33/1,9548 Male (Shekelle) 1984 Hawaii Case- 364/627 Male and (Hinds) control female 1985 New Mexico Case- 447/759 Male and (Samet) control female 1985 California Case- 220/220 Female (flu) control 1985 Hawaii Cohortb 74/302C Male (Nomura) 1986 New~Jersey Case- 763/900 Male (Ziegler) control 1986 Maryland Cohortb 99/196C Male and (Menkes) female 1987(a,b) New York Case- 450/902 Male and (Byers) control female 1987 Texas Case- 308/308 Male (Bond) control aNumber of cases/n~mber of subjects in cohort. bBased on serum measurements. CNumber of cases/number of controls analyzed from the cohort. 1984; Menkes et al., 1986; Nomura et al., 1985; Samet et al., 1985; Shekelle et al., 1981a; Wu et al., 1985; Ziegler et al., 1986~. On the other hand, certain diet-disease associations have been more difficult to establish with any consistency. For example, despite impressive ecologic correlations (shown earlier) and strong animal evidence, 103

OCR for page 95
an association between dietary fiat and breast cancer has been reproduced inconsistently and only weakly in case-control and cohort studies by diet history methods (National Research Council, Committee on Diet, Nutrition, and Cancer, 1982; Rohan and Bain, 1987~. For many investigators, this has raised questions about the value of these methods, particularly in situations in which the true relative risks may be modest in magnitude. This concern has served as a major stimulus in recent years for efforts to find meaningful biochemical measures of dietary intake for use in epidemiologic research. Studies based on biochemical measures. Suitable biochemical markers of dietary intake have not been easy to identify. To be useful, a marker (1) should be sensitive to (and relatively specific for) the particular nutrient (or other dietary constituent) of interest, (2) should vary in a definable manner in response to changes in dietary intake, (3) should achieve an integration of highly variable intakes, (4) should reflect the relevant period of exposure, and (S) should be easily and reliably measured. Such requirements are not readily met. For example, it has not yet been possible to find a marker of total fat intake that can satisfy all of these criteria. One can use biopsies of adipose tissue to provide information on fatty acid distributions, although not total fat intake. But even this procedure cannot be contemplated for the large numbers of subjects involved in most epidemiologic studies, especially healthy volunteers. Furthermore, in case-control studies, the marker levels may be influenced by the disease process itself, so that one is measuring a consequence, not a precursor, of the disease. _ ~ ~ ~ e _ ~ e Nevertheless, some epidemiologic studies have used biochemical measurements. For example, at a group level, serum cholesterol has been positively associated with coronary heart disease mortality (Keys, 1980~. However. since the serum cholesterol level is not greatly influenced by dietary cholesterol (Keys, 1984), this biochemical measure is not really a good marker for dietary cholesterol intake. Similarly, serum retinal has been examined relative to cancer risk. Since this parameter is under homeostatic control and only varies at the extreme ranges of intake, it does not satisfy any of the criteria for a marker noted above. Thus, as one 104

OCR for page 95
would have expected, most studies have found no association between serum retinal levels and cancer risk (Bertr~m et al., 1987~. Willett et al. (1983) examined selenium levels in blood as a marker for dietary intake of this trace element. They found that low levels in serum, especially when combined with low levels of vitamins A and E, were associated with a higher risk for cancer. Since all three of these nutrients have antioxidant properties, this observation suggests that greater power in defining risk groups for a particular disease might be achieved if a single index for several factors that share a common mechanism of action could be developed. There is yet another complexity to the study of dietary or nutritional factors in relation to human disease. This is the question of individual susceptibility. Even a biological marker that can be measured with considerable precision and does reflect usual intake might still not serve to distinguish high- from low-risk groups if individuals vary substantially in their susceptibilities to the effects of exposure. Such differences in susceptibility could be genetic in origin or could result from other environmental exposures. Thus, two individuals with similar exposures to a dietary risk factor of interest could have different rates of metabolism of the agent, which, in the case of a carcinogen precursor (procarcinogen), could result in the greater or lesser formation of an active carcinogen. This could be entirely hereditary, or it could result from another exogenous exposure that induces enzymes in the same metabolic pathway. Intervention trials. Recently, diet has been the focus of intervention trials involving cancer and coronary heart disease (Greenwald et al., 1987; Multiple Risk Factor Intervention Trial Research Group, 1982~. Such investigations are costly, require considerable nutritional expertise, and for ethical reasons, can only assess the effects of presumed beneficial dietary modifications. Unfortunately, incomplete knowledge or control of such factors as optimal levels of intake to achieve the desired effect, individual variation in susceptibility, compliance in the intervention groups, and dietary drift in the nonintervention groups all can easily undermine these research efforts. 105

OCR for page 95
FUTURE NEEDS AND DIRECTIONS From this brief review of the interest in nutrition and the evolving use of nutritional methods in epidemiologic research, it is obvious that further methodologic developments must occur. No single investigative approach seems to be without its limitations with regard to nutritional assessment. For this reason, efforts in all areas of epidemiologic research, including ecologic analyses as well as case-control, cohort, and intervention studies, should be encouraged. A convergence of findings from many sources by a variety of approaches continues to offer the best hope for identifying meaningful etiologic relationships. Beyond this general statement, however, certain specific needs related to nutrition can be highlighted: Improvements in our ability to obtain accurate diet history information are very much needed. Although further validation of current methods can be one part of this effort, there is a need to develop techniques that can better distinguish among individuals with meaningful differences in their intake levels. Since the true relative risks for many diet-disease relationships that have not yet been established may be small, even moderate degrees of misclassification of consumption may be sufficient to result in false-negative results. The fact that such relative risks may not be impressive in magnitude should not lead to the conclusion that these relationships are of no public health importance, however. Since the combined incidence of diet-related illnesses is extremely high and the consumption of suboptimal diets is widespread, the number of individuals at risk is substantial; thus, even small relative risks can have very significant population impacts. . As part of a total effort to advance this area of research, we should continue to seek biological markers of dietary intake. Such markers do not necessarily have to be direct tissue or serum analogs of the food components of interest. For example, a tissue enzyme that is influenced by changes in intake of a particular food constituent may serve as a better marker than the serum or target tissue level of the constituent itself (or even one of its metabolites). Similarly, the dietary factor that best reflects a biochemical measure of interest may not be the most obvious one. This is well 106

OCR for page 95
illustrated by cholesterol, the serum level of which is more clearly influenced by saturated fat than by cholesterol in the diet. Furthermore, other indices, not necessarily biochemical, such as central to peripheral ratios to adipose tissue or other anthropometric measurements, can be useful adjuncts in the evaluation of dietary exposures. Greater attention should be given to assessing individual differences in susceptibility to the effects of dietary exposures. This emphasis should include both genetic and nongenetic factors. Such a refinement in classification could offset a significant portion of the dietary measurement error, which probably will never be completely eliminated. Since different food components or nutrients may act in the same way to promote or prevent illness, consideration should be given to the development of indices that combine the separate effects of several agents. For example, vitamins A and E, selenium, and certain nonnutritive constituents in foods all have antioxidant properties. A single measure of the combined effect of all antioxidants in reducing cancer risk (using either a biochemical measurement or a computed value from dietary histories) might better account for differences among individuals than could any component alone. Other needs related to data analysis should also be addressed, such as better means for assessing interactions among nutrients, more attention to eating patterns and food temperature, and greater accuracy of data in food composition tables. More nutritionists are needed to support this area of research. This is a special category of nutritionists who should have a basic understanding of the principles of epidemiology and biostatistics, since the focus in such studies is quite different from that in dietary counseling or in laboratory-based nutritional research. For example, such nutritionists must recognize certain limitations inherent in the conduct of field research on relatively large numbers of subjects or in designing practical methods of dietary assessment when individuals need to be classified precisely in a relative but not necessarily absolute sense. 107

OCR for page 95
CONCLUSION Nutrition is receiving increasing attention by epidemiologic researchers, since many of the major diseases of public health importance today are ones that are likely to have a strong basis in dietary behavior. Research efforts to date have identified certain significant associations, such as that between saturated fat intake and coronary heart disease, between simple sugars and dental caries, or somewhat less definitively, between dietary fat intake and some cancers. Further advances are being constrained by methodologic limitations, however, including substantial errors of measurement in the diet history as these measurements are currently implemented, the lack of functional biological markers of specific dietary intakes, an inability to classify subjects into risk groups with sufficient refinement, and unmet needs related to data analysis. Nutritionists who are well trained for this area of research are rare and are very much needed. They could have a significant impact on the elucidation of important diet-disease relationships, leading to meaningful public health measures that could greatly reduce chronic disease morbidity in the future. REFERENCES Bertrand, J.S., L.N. Kolonel, and F.L. Meyskens. 1987. Rationale and strategies for chemoprevention of cancer in humans, Cancer Res. 47:3012-3031. Block, G. 1982. A review of validations of dietary assessment methods. Am. J. Epidemiol. 115:492-505. Bond, G.C., F.E. Thompson, and R.R. Cook. 1987. Dietary vitamin A and lung cancer: Results of a case-control study among chemical workers. Nutr. Cancer 9:109-121. Byers, T., J. Marshall, E. Anthony, et al. 1987a. The reliability of dietary history from the distant past. Am. J. Epidemiol. 125:999-1011. Byers, T.E., S. Graham, B.P. Haughey, et al. 1987b. Diet and lung cancer risk: Findings from the Western New York Diet Study. Am. J. Epidemiol. 125:351-363. Carroll, K.K. 1975. Experimental evidence of dietary factors and hormone-dependent cancers. Cancer Res. 35:3374-3383. 108

OCR for page 95
Greenwald, P., C. Clifford, R.R. Butrum, et al. 1987. Feasibility studies of a low-fat diet to prevent or retard breast cancer. Am. J. Clin. Nutr. 45:347-353. Hankin, J.H. 1986. 23rd Lenna Frances Cooper Memorial Lecture: A diet history method for research, clinical, and community use. J. Am. Dietet. Assoc. 86:868-875. Hinds, M.W., L.N. Kolonel, J.H. Hankin, et al. 1984. Dietary vitamin A, carotene, vitamin C and risk of lung cancer in Hawaii. Am. J. Epidemiol. 119:227-237. Humble, C.G., J.M. Samet, and B.E. Skipper. 1987. Use of quantified and frequency indices of vitamin A intake in a case-control study of lung cancer. Int. J. Epidemiol. 16:341-346. James, W.P.T., S.A. gingham, and T.J. Cole. 1982. Epidemiologic assessment of dietary intake. Nutr. Cancer 2:203-212. ~ Jolliffe, N., and M. Archer. 1959. Statistical associations between international coronary heart disease death rates and certain environmental factors. J. Chronic Dis. 9:636-652. Kato, I., S. Tominaga, and T. Kuroishi. 1987. Per capita foods/nutrients intake and mortality from gastrointestinal cancers in Japan. (Gann) 78:453-459. ~_ , _ _ Jpn. J. Cancer Res. Keys, A. 1980. Seven countries: A multivariate analysis of death and coronary heart disease. Pp. 121-135. Harvard University Press, Cambridge. Keys, A. 1984. Serum cholesterol response to dietary cholesterol. Am. J. Clin. Nutr. 40:351-359. Knox, E.G. 1977. Foods and diseases. Br. J. Prevent. Sac. Med. 31:71-80. Kolonel, L.N., J.H. Hankin, J. Lee, et al. 1981. Nutrient intakes in relation to cancer incidence in Hawaii. Br. J. Cancer 44:332-339. Lilienfeld, A.M., and D.E. Lilienfeld. 1980. Foundations of Epidemiology, 2d ed. Oxford University Press, New York, 375 pp. Lubin, J.H., P.E. Burns, W.J. Blot, et al. 1981. Dietary factors and breast cancer risk. Int. J. Cancer 28:685-689. McGandy, R.B., D.M. Hegsted, and F.J. Stare. 1967. Dietary fats, carbohydrates and atherosclerotic vascular disease. N. Engl. J. Med. 277:186-192. McKeown-Eyssen, G.E., K.S. Yeung, and E. Bright-See. 1986. Assessment of past diet in epidemiologic studies. Am. J. Epidemiol. 124:94-103. 109

OCR for page 95
Menkes, M.S., G.W. Comstock, J.P. Vuilleumier, et al. 1986. Serum beta-carotene, vitamins A and E, selenium, and the risk of lung cancer. N. Engl. J. Med. 315:1250-1254. Morgan, R.W., and M.G. Jain. 1978. A comparison of dietary methods in epidemiologic studies. Am. J. Epidemiol. 107:488-498. Multiple Risk Factor Intervention Trial Research Group. 1982. Multiple risk factor intervention trial: Risk factor changes and mortality results. J. Am. Med. Assoc. 248:1465-1477. National Research Council, Committee on Diet, Nutrition, and Cancer. 1982. Diet, Nutrition, and Cancer. National Academy Press, Washington, D.C. Nomura, A.M.Y., G.N. Stemmermann, L.K. Heilbrun, et al. 1985. Serum vitamin levels and the risk of cancer of specific risks in men of Japanese ancestry in Hawaii. Cancer Res. 45:2369-2372. Prentice, R.L., M. O'Sullivan, and S.G. Self. In press. Dietary fat and breast cancer: An assessment of the epidemiologic literature in relation to the design of the women's health trial. Cancer Res. Rohan, T.E., and C.J. Bain. 1987. Diet in the etiology of breast cancer. Epidemiol. Rev. 9:120-145. Rohan, T.E., and J.D. Potter. 1984. Retrospective assessment of dietary intake. Am. J. Epidemiol. 120:876-887. Samet, J.M., B.J. Skipper, C.G. Humble, et al. 1985. Lung cancer risk and vitamin A consumption in New Mexico. Am. Rev. Respir. Dis. 131:198-202. Shekelle, R.B., S. Liu, W.J. Raynor, et al. l9Bla. Dietary vitamin A and risk of cancer in the Western Electric Study. Lancet ii: 1185-1190. Shekelle, R.B., A. MacMillan Shryock, O. Paul, et al. 1981b. Diet, serum cholesterol, and death from coronary heart disease: The Western Electric Study. N. Engl. J. Med. 304:65-70. Silverberg, E., and L. Garfinkel. 1987. Cancer statistics. Ca--A Cancer J. 37:2-19. Willett, W.C., S. Pressel, et al. 1983. Prediagnostic serum selenium and risk of cancer. Lance t ii:130-134. Wu, A.H., B.E. Henderson, M.C. Pike, et al. 1985. Smoking and other risk factors for lung cancer in women. J. Natl. Cancer Inst. 74:747-751. Ziegler, R.G., T.J. Mason, A. Stemhagen, et al. 1986. Carotenoid intake, vegetables, and the risk of lung cancer among white men in New Jersey. Am. J. Epidemiol. 123:1080-1093. 110