Click for next page ( 13


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 12
4 EpidemiologicalMethods The investigation of diet and cancer in human populations has been complicated by concerns about the adequacy of the various methods of assessing exposure. The following list summarizes the approaches that have been used to measure dietary exposure in epidemiological studies: Methods Based on Group Data National per-capita intakes Household food use Methods Based on Individual Data Food intake records or diaries Recent (24-hour or 7-day) diet recalls Diet histories Biochemical markers of exposure Anthropometric measure- ments These approaches and their respective strengths and limitations were described in detail in Chapter 3 of the first report by the Committee on Diet, Nutrition, and Cancer (National Research Council, 1982~. The following paragraphs summarize methodological problems that character- ize epidemiological studie s of diet and cancer and provide suggestions for research. Assessment of dietary intakes can be based either on group data or on data collected directly from individuals The most frequently used method in the first category is the estimation of per-capita intakes from national food disappearance data; however, because of the rudimen- tary nature of such data, these estimates may not be very accurate indi- cators of actual consumption levels. Assessment of household food use can be based on recall histories of consumption, collected from the chief food-preparers, or on the availability of food supplies in the home, recorded by a household member or a trained nutritionist. Since thi s me thod require s contact wi th individual household s without produc- ing individual consumption data, it is not often used. These so-called "group" methods have proved useful for aggregate correlation analyses, which are often the basis for generating new hypotheses. Such aggregate correlations, however, do not control for confounding factors, nor do they necessarily reflect associations at the individual level. The second type of method is based on data collected directly from individuals. These data are obtained by a variety of mechanisms, par- ticularly through diet recall s and the contemporaneous recording of 12

OCR for page 12
Epidemiological Methods 13 weights or volumes of foods consumed. The collection of data from food records necessarily focuses on current intake. Because this method re- quires meticulous attention to detail, it is practical only in studies involving small, selected samples. Diet recalls, on the other hand, can be collected from much larger and more representative samples. In contrast to recent recalls, which are focused on specified brief periods close to the time of interview, diet histories, which are also based on recall, assess usual intakes over a longer period in the past. Thus, diet histories can be used to study intakes prior to the onset of disease in cancer cases, even allowing for relatively long induction periods. Therefore, this method has been used most frequently in in- vestigations of the etiology of disease. The use of biochemical markers of exposure, such as components of body fluids or tissues, represents a somewhat different approach to dietary assessment. Unfortunately, few markers that reflect long- term dietary exposures have yet been identified, since the concentra- tions of substances in biological fluids such as serum and urine are usually related to recent exposures or to the functioning of ho~eo- static control mechanisms rather than to long-term intakes and body stores. In another method, certain anthropometric measures known to be correlated with obesity (including indices based on height and weight or skinfold thickness) have sometimes been used as indirect indicators of nutritional status. Since past weight, for example, can be obtained by interview, these measures are not necessarily limited to the assess- ment of current nutritional status. For studies involving large samples, the diet history is the most practical and useful method. Although many investigators have expressed reservations about approaches that rely on long-term memory, there have been few studies on the validity and reproducibility of any of the diet recall methods used in epidemiological studies. Because dietary habits do not remain constant throughout life, even diet histories must be focused on a particular period. In some instances, as in studies of putative promoters, relatively recent intakes are of interest, whereas in studies of initiators, early dietary habits are likely to be rele- vant. It is important to recognize that recall errors tend to be random and apply equally to the comparison groups (e.g., cases and controls). Thus, they will contribute to a reduction in the estimate of relative risk and tend to lead to a false negative conclusion. Therefore, in the interpretation of data from studies using the diet history method, a finding of a significant association is likely to be considered valid, whereas a negative finding (i.e., lack of an association) would not be assigned as much weight. The collection of diet histories requires consideration of a wide array of variables, since eating reflects complex social, behavioral, and cultural patterns. Thus, food preferences, the temporal pattern of

OCR for page 12
14 DIET, NUTRITION, AND CANCER: DIRECTIONS FOR RESEARCH eating, portion sizes, use of condiments, and methods of food prepara- tion (such as frying) and storage (which can affect nutrient content and provide opportunities for contamination) should all be considered, since any of these factors might have important influences on cancer risk. In few studies have investigators elected or been able to con- sider these many factors simultaneously. Moreover, current approaches to dietary data collection are generally based on the subjects' recol- lection of food items consumed, not their constituents, whereas most of the hypotheses being tested pertain to nutritive or nonnutritive food components. Since all foods are collections of such components, it has not been possible to relate the findings in epidemiological studies specifically to single constituents of foods. Data collection is only the first step in the assessment of dietary exposure in epidemiological studies of cancer. The food intake data have to be classified in a meaningful manner and, in many instances, converted into nutrient equivalents. Useful data on food composition have been published by the U.S. Department of Agriculture (1975), and many investigators have assembled additional information from a variety of specific sources. Unfortunately, the data base is not uniform or complete, and the composition of a particular food item is not fixed and consistent; it may vary by the location and the season of its growth and by the method of its processing. Thus, the average or representa- tive values in the food composition tables may be relatively inaccurate indicators of nutrient intakes for discrete population subgroups and may contribute to false negative conclusions, as explained above for diet histories. In addition, many nutrients and all nonnutritive com- ponents of foods (e.g., naturally occurring flavones or food additives) are not included in these tables, although they may be factors of interest in the testing of certain dietary hypotheses. Epidemiology provides the only direct approach to the assessment of risk for human beings. However, because dietary measurements are not precise, estimates of risk or benefit tend to be less than the true effect. Furthermore, because small increments in risk might be accounted for by chance variation or by uncontrolled confounding fac- tors, a relative risk (i.e., incidence of the disease in those exposed divided by the incidence in those not exposed) of less than 2.0 is rarely concluded to be meaningful. Other difficulties arise because of the latent period that must elapse between the initial exposure and the development of the disease--a period that could well encompass a number of decades and be affected by a multitude of modifying factors. Impre- cision in measuring dose also results in imprecision in assessing the relationship between dose level and response, an important indicator for establishing causality, especially when the dose response is linear. Nonetheless, the advantages of direct measurements of risk in epidemio- logical studies are likely to outweigh advantages inherent in laboratory studies. These are only a few of the methodological issues that the com- mittee considered in relation to the needs for future epidemiological research on diet and cancer. In making research recommendations, the committee placed its major emphasis on trying to rectify the weaknesses

OCR for page 12
Epidemiological Methods 15 in methods that precluded the studies reviewed in the first RESEARCH RECOMMENDATIONS Assessment of Dietary Exposures formulation of firm conclusions from the report. ~ New approaches to improving the quality of dietary intake data, especially in relation to long-term dietary patterns, should be ex- plored. Research in this area should include a search for new aids to facilitate individual recalls and should enlist the participation of specialists in human behavior and psychometric measurement, who might be able to identify the best ways to elicit accurate information and to provide new insight into the factors that influence the responses of people asked to describe their diets. In addition, a search should be conducted for biochemical markers that reflect relevant dietary expo- sures some time in the remote past. Where direct measurements of exposure are unsuitable (e.g., in examining serum for a nutrient under homeostatic control), indirect markers, such as serum-binding proteins or enzyme levels, may be appropriate substitutes. Because rates of absorption and metabolism of nutrients may vary among subjects, how- ever, individuals ingesting the same diet may differ in their ultimate exposure because of individual differences in metabolism and, thus, may differ in their degree of risk for cancer. These biochemical differ- ences among individuals should be studied in relation to observed differences in cancer risk. o More effort should be spent in evaluating the validity of dietary methods. One approach, which has not been widely used in epi- demiological research, would be to compare suitable biochemical mea- surements with information derived concurrently from food records or diet recalls. Although this strategy is generally useful only for assessing the validity of very recent recalls, a high degree of agreement among the different forms of measurement suggests that the recall approach applied to a more remote time period may also yield relatively accurate results. This approach could be further validated if stored biological samples from the same period were available, but the limitations for the use of stored samples described on page 17 would also apply here. ~ Food composition data bases should be improved with respect to information on both nutritive and nonnutritive constituents and on regional and seasonal variations in composition. The current data base lacks information on the components of fiber, essential trace elements (e.g., selenium), nonessential trace elements (e.g., cadmium), and food additives. o In future studies, greater attention should be given to food preparation and storage. These two processes can result in the produc- tion or modification of the nature and amount of a variety of dietary components that present a potential risk for cancer (e.g., aflatoxin

OCR for page 12
16 DIET, NUTRITION, AND CANCER: DIRECTIONS FOR RESEARCH contamination during storage of grain or production of mutagens during cooking). Better data are needed on long-tern trends in the composition of average national diets. On a regular basis, cross-sectional infonma- tion should be collected from representative samples of the U.S. popu- lation through surveys such as the annual Health and Nutrition Examina- tion Surveys conducted by the National Center for Health Statistics, Department of Health and Human Services. The accumulated data can then be correlated with trends in cancer incidence. Information on such trends can also be obtained by repeated measurements of intake at well- spaced intervals among the subjects in prospective cohort studies. The resultant data will provide information not only on the changing pat- terns of eating during a lifetime but also on the reproducibility of the dietary assessment method used. Types of Epidemiological Study Each of the several epidemiological study designs offers particular advantages in certain circumstances (e.g., rare cancers can be examined in case-control studies but seldom in cohort studies). On the other hand, each of these designs has inherent limitations, as noted in the first report (National Research Council, 1982~. For this reason, stud- ies of all types, if appropriately applied, will be of value in further investigations of diet and cancer in human populations. Correlation studies, if well designed, can continue to provide useful information on the relationship between diet and cancer. When- ever possible, these studies should be based on exposure data collected directly from individuals rather than on per-capita consumption esti- mates. When sufficient information is available, they should also be based on morbidity rather than on mortality data, since the latter are influenced by survival patterns and can be misleading (National Research Council, 1982~. O Consideration should be given to carrying out case-control studies in populations such as those with unusual dietary habits or levels of exposure as well as those at unusual risk for specific cancers. International, collaborative case-control studies should be encouraged, but the sample size in each geographical area should be sufficiently large to permit separate analysis of the data. This would enable investigators to replicate the findings in diverse settings using a common methodology. To the greatest extent possible, the determination of sample sizes in all case-control studies should allow for meaningful examination of particular subgroups, statistical control of all important confounding factors, and the examination of interac- tions between separate dietary components or between dietary and nondietary factors. In estimating sample size requirements, investi- gators need to recognize that errors in measurement of exposure

OCR for page 12
Epidemiological Methods 17 necessitate an increase in the sample size to demonstrate statistically significant associations. Case-control studies have limited value for investigations based on biochemical indicators of exposure, since these markers may be altered by the disease in the cases. However, clues to the possible biochemical markers of exposure may be obtained from such studies. ~ More cohort studies of diet and cancer are needed. This is the most suitable approach for investigations based on biochemical indi- cators of dietary exposures. Cohort studies are best carried out in populations at high risk for diet-associated cancers, where representa- tive dietary data can be readily obtained and where long-term follow- up can be aided by population-based cancer registries and good vital statistics data for end points. Whenever possible in such studies, biological samples (such as serum, urine, and feces) should be col- lected and stored for later biochemical analysis of specimens from controls and from subjects who subsequently develop cancer. Long-term storage also enables the investigators to incorporate newer hypotheses or analytical procedures into the study at a later time. However, the utility of stored biological samples is limited because the samples reflect only one or possibly a few finite periods in the individual's life and because the samples may deteriorate during storage and han- dling. Because cohort studies are expensive and of very long dura- tion, a search should continue to be made for existing cohorts for which there is relevant information on dietary exposures. In designing cohort studies, the sample size should be sufficiently large to ensure that an adequate number of cases will be identified, and their selec- tion should be based on considerations discussed above for case-control studies. Intervention studies (trials) have many advantages, but they should be considered only when supporting evidence from other types of studies is strong and only after the risks involved have been carefully weighed. For ethical reasons, these studies will have to be limited to examinations of putative protective factors or of the effects from reduced exposure to risk factors. Several trials of specific microcon- stituents (e.g., 6-carotene) have recently been initiated. There is also a great need to study the effects of specific foods and food groups (e.g., dark green and deep yellow vegetables, and those of the genus Brass~ca--members of the cabbage family), since the biological effectiveness of a food component is probably affected by the presence of other constituents in the diet, and since the effects observed in epidemiological and experimental studies may be due to a mixture of different inhibitors of carcinogenesis. To obtain more definitive data, the randomization procedure in these studies should 1oe based on individuals rather than on groups, whenever possible. Analysis of Data Analysis of data from epidemiological studies should include examinations of specific foods and food classes, even when the hypothe-

OCR for page 12
18 DIET, NUTRITION, AND CANCER: DIRECTIONS FOR RESEARCH sis pertains to nutrients. Relationships involving nutrients from selected groups of foods or involving nonnutritive components of foods might be uncovered by thi s approach. Studies should be conducted to determine the limitations of the logistic regression model for the analysis of epideraiological data on diet and cancer. The extent to which the actual data in a study can deviate from the theoretical distribution and still yield meaningful results should be defined. Moreover, since many of the dietary and nondietary factors in studies on cancer are highly intercorrelated, efforts should be made to explore statistical methods that are less sensitive to collinearity than is multiple logistic regression analysis. Fi nally, statistical techniques need to be developed to describe snore accurately the various forms of interaction among dietary variables.