| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Supplements |
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
Key Words: Nutrition • epidemiology • cancer • coronary heart disease • coronary artery disease • diet • dietary survey methods • study design
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
Much of the research that is receiving publicity comes from epidemiologic studies of the relation between diet and health. Nutritional epidemiology is the science of how nutrition affects health. It is based on the principles of nutritional science and epidemiology. Moreover, it is an essential component in the research to understand causal relations and in the development of public health policy and prevention strategies. Unfortunately, there is a great deal of confusion in both the scientific and lay press about the nature of nutritional epidemiology.
The purpose of this article is to discuss the strengths and limitations of nutritional epidemiology. The article is divided into 4 sections: an overview of nutritional epidemiology, a review of the types of studies most often encountered, a discussion of the measurement of exposure in nutritional epidemiology, and a discussion of the interpretation of epidemiologic data.
|
OVERVIEW OF NUTRITIONAL EPIDEMIOLOGY |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
The methods used in nutritional epidemiology are designed to take those features into account. They focus on measuring the exposure to nutritional factors, the frequency and distribution of disease, and the exposure to other factors that could confound the hypothesized association. Nutritional exposures are defined or measured through the intake of foods, nutrients (total from foods and supplements or separately from these 2 sources), and nonnutrients, additives, contaminants, and chemicals in foods that were incorporated into a food as part of the agricultural process or were formed during food processing or preparation. Nutritional exposures also include biochemical measures of nutritional status, biomarkers of intake, biological intermediates influenced by diet (eg, serum cholesterol concentration), anthropometric measurements, genetic factors, and clinical findings.
Estimated dietary intake, however, may not be the exposure of interest (
Table 1
). According to Margetts and Nelson (
5
), "Unlike many other exposures used by epidemiologists, diet is so complicated that special knowledge is required to enable the correct exposure measure to be used. The perspective of nutritional epidemiology thus encompasses factors that influence the availability and consumption of food, on one hand, to the whole-body expression of disease, on the other. The point(s) along this continuum at which to measure the relevant aspect(s) of diet is a key concern of nutritional epidemiology."
|
(
4
). By using that approach and depending on future scientific advancements, only certain limited aspects of this model can be studied in free-living populations (
6
). As a result, only through an iterative exchange of ideas and hypotheses among the laboratory, clinical, and epidemiologic sciences can progress ultimately be made in understanding whether the nutritional exposure is a causal factor and, given the multifactorial nature of chronic disease, what role nutrition or, more precisely, diet can play in disease prevention.
|
|
TYPES OF STUDIES |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
Observational epidemiology
Cross-sectional studies
Design.
In cross-sectional studies the question being asked is, "What is the correlation between currently having the disease and the nutritional exposure?" People are selected for inclusion in the study without regard for disease status or nutritional exposure. In these studies disease status and nutritional exposure are both measured at the same time; this produces a snapshot—a measurement of health at one specific time.
Strengths.
The chief strengths of cross-sectional studies can be a relatively large sample size and, for studies such as the National Health and Nutrition Examination Surveys (NHANES), their representativeness. These features make cross-sectional studies an ideal design for estimating the mean amounts and distributions of key risk factors (eg, dietary components, serum cholesterol concentrations, blood pressure, and weight) and the prevalence of individuals with certain diseases or conditions (eg, coronary heart disease, high blood cholesterol, and high blood pressure). Repeated cross-sectional surveys within the same population are essential for monitoring trends in reference distributions and prevalence (
9
). That information is central to nutrition monitoring and vital to the design and effectiveness of public health programs. If designed appropriately, surveys can also provide the baseline information for a prospective study.
Weaknesses.
Because both nutritional exposure and disease status are measured at the same time, cross-sectional studies are a relatively weak study design for assessing causal associations. The problem is that most of the time it is difficult if not impossible to determine whether the nutritional exposure is a cause of the disease or the nutritional exposure was affected by the disease process. For example, the disease may lead individuals to change their dietary intake patterns or the disease process may produce changes in serum concentrations of nutrients.
Cross-sectional studies are often used to assess the factors correlated with other risk factors (eg, serum cholesterol and blood pressure) in an effort to understand how their concentrations or levels are influenced. For nutritional epidemiology, the goal is to determine how usual diet influences intermediate risk factors and through them the risk of disease. However, lack of variation in usual intake by individuals in a particular population, the large within-person variation in dietary intake, and the inaccuracy of dietary survey methods all make it difficult to show correlations between diet and intermediate risk factors such as serum cholesterol concentration. As discussed below, many of these measurement issues affect the ability to study diet-disease relations, even when using more rigorous study designs. As a result, cross-sectional studies are considered to be a relatively weak method for studying diet-disease associations, and results from them should be interpreted cautiously.
Case-control studies
Design.
Case-control studies, which are also referred to as retrospective and case-referent studies, are designed to answer the question, "Do persons with the disease (case subjects) consume diets that are different in composition from the diets consumed by persons who have not been diagnosed with the disease (control subjects)?" For example, do individuals with cancer consume less fruit and vegetables than do those without cancer? In this type of study, recently diagnosed (ie, incident) case subjects and a set of control subjects are interviewed concerning their dietary habits, or their nutritional status is assessed with a biological measure (for example, the serum concentration of a nutrient). The goal is to determine the usual nutritional exposure before the onset of the disease. In most studies, food-frequency or dietary history questionnaires are used to assess usual intake in the recent past (ie, 
1 y before onset of symptoms); such questionnaires have also been used to assess usual diet in the distant or remote past (eg, 5, 10, or 15 y earlier).
Strengths.
Case-control studies are relatively inexpensive to conduct and require smaller sample sizes than do prospective studies; this makes them especially good for the study of rare diseases. They also yield results relatively quickly.
Weaknesses.
Case-control studies have several important weaknesses. A key assumption is that the measure of nutritional exposure has not been influenced by the disease process or by a change in dietary habits. However, for a biological measure of exposure, such as serum nutrient concentration, that may be difficult to establish with certainty. Moreover, it is very difficult to estimate usual intake from recollections about the remote past.
Cohort studies
Design.
Cohort studies are also referred to as prospective, incidence, follow-up, and longitudinal studies. The question being asked in prospective studies is, "Do persons with the risk factor develop or die from the disease more frequently or sooner than those who do not have the risk factor?" For example, are persons who have high serum ferritin levels more likely to develop coronary heart disease (CHD) than persons who do not?
Cohort studies begin with a cross-sectional survey that ascertains the presence or absence of the disease of interest as well as dietary intake and other measures of exposure. After persons diagnosed with the disease being studied, eg, CHD, are excluded the population is then followed over time to see who develops the disease and when the disease develops. The persons remaining in the study are those who are at risk of developing the disease. That is, they are participants for whom it can be reasonably established that the nutritional exposure preceded the development of the disease.
A relatively new type of cohort study that is a hybrid of the cohort and case-control study designs is the case-cohort or nested case-control study ( 10 ). In such a study the researcher compares all the subjects that have developed the disease over a defined period of time with a random sample of control subjects selected from the entire cohort. For example, because of cost concerns a researcher may collect 7-d food records or serum samples from all the study subjects at the first examination but may store the data for future use instead of analyzing it immediately. A portion of the stored data may then be analyzed in the future to compare case subjects that have developed the disease since the first examination with a random sample of control subjects. Such a study is prospective in design (exposure precedes diagnosis) and is cost-effective because samples from only a subset of the entire cohort are analyzed. This design allows the researcher the flexibility to investigate hypotheses that would have been unknown at the start of the study and therefore extends the utility of prospective studies.
Strengths.
Because the time sequence in which the nutritional exposure precedes disease development can be established, prospective studies are considered to be the best of the observational study designs. Moreover, because the goal is to measure the current, usual amounts of nutritional exposures, these can be measured without the results being influenced by the presence of the disease. Prospective studies can be used to study the risk factors for additional diseases, and prospective studies yield absolute estimates of risk, whereas case-control studies yield only relative estimates.
Weaknesses.
Prospective studies are costly and time-consuming. They require that many individuals be followed for long periods of time, and attrition can be a problem. As a result, the cohort study design is used infrequently for the study of rare diseases. In most prospective studies, risk factor exposures are measured only once at the beginning of the study. The Framingham Heart Study (
11
), Nurses' Health Study (
12
), and the NHANES I Epidemiologic Follow-up Study (
13
) are examples of prospective studies in which health habits are measured regularly. One of the key assumptions in prospective studies is that diet is constant between periods of assessment (
14
). For studies in which diet is not assessed after the initial or baseline examination, dietary patterns are assumed to be relatively constant over the entire study period. For cohort studies conducted during the 1950s, 1960s, and 1970s, that may have been a reasonable assumption. However, with the greater amount and acceptance of public health nutritional advice, the successful marketing of increasing numbers of new foods and food products (eg, low-fat and reduced-cholesterol foods or foods that contain components, such as soy proteins or
trans
fatty acids, that are substantially influenced by market forces), and the increasing use of dietary supplements, it is unlikely that a nutritional exposure measured in the past accurately reflects long-term exposure.
Ecologic studies
Design.
In ecologic studies, aggregate data representing entire populations are compared. The most common example of this type of study is one in which mortality rates for different countries are correlated with a nutrient measurement based on food disappearance data. The design can be much more complex, however, as in the Seven Countries Study (
15
), the Ni-Hon-San Study (
16
), and the Intersalt Study (
17
). Such studies typically include countries that differ substantially in nutrient exposure and disease outcome to enhance the potential to detect differences and thus test the studies' hypotheses (
18
). In these more complex studies, dietary intake, serum cholesterol concentration, and blood pressure might be measured for all subjects. Mortality rates for persons enrolled in the study might also be measured. In the analysis of the study data, the mean values of the exposure and outcome variables for each locale are correlated to assess the extent of the associations between them.
Strengths.
Populations that are relatively homogeneous tend to exhibit little variation in dietary patterns. Therefore it may be difficult if not impossible to find an association between diet and disease risk within a single population. Multipopulation studies are often necessary to show a direct link between a nutritional exposure and risk of disease (
19
). Measuring dietary intake in groups of individuals greatly minimizes many of the problems associated with the measurement error affecting both dietary data and the outcome variable, especially if the latter is a biological variable such as urinary sodium excretion or serum cholesterol concentration.
Weakness.
Ecologic fallacy is always possible (
1
); ie, the aggregate relations may not accurately reflect the associations that would be seen in individuals.
Ethical considerations in observational studies
Most epidemiologic studies are observational rather than experimental for operational and ethical reasons; the operational difficulties with randomized clinical trials are discussed below. The ethical constraints are related to the hypotheses themselves. Typically, the aspects of the proposed mechanism that are studied in humans are based on information gained from animal models. Portions of the mechanism that require the use of invasive measurement methods or that use nutrient intake to initiate or promote the disease or condition are studied in animal models. Experiments in humans must, by their very nature, be restricted to those that study the effects of removing, reducing, or ameliorating the risk factor.
For example, a diet high in saturated fatty acids and cholesterol increases a person's risk of developing CHD, which may lead to a heart attack and possibly even death. It is not ethical to randomly assign some people to a diet high in saturated fatty acids and cholesterol and see if they die sooner and in greater numbers than control subjects assigned to a diet low in saturated fatty acids and cholesterol.
Experimental epidemiology
Feeding or metabolic-ward studies
Design.
Feeding studies are experiments in which groups of individuals are fed precisely measured diets with one or more components varied, and the effect on a biologic variable is then measured. The design used most often is a crossover design in which the individuals act as their own controls. Such studies are usually performed in a research setting such as a clinical research center.
Strengths.
Most nutrients are highly correlated with each other. In a feeding study it is possible to test for the independent effects of one nutrient while holding all other factors constant. For example, the precise effects of saturated fatty acids on serum total cholesterol concentrations were determined by using feeding studies (
20
). This is rarely, if ever, possible in observational studies. Admittedly, the study of nutrients even in an experimental design is more complex than the previous statements imply because of the interrelations of micronutrients and the need to adjust macronutrients to maintain constant body weight. However, compared with observational studies, control of nutrient intake is possible in experimental studies.
Weaknesses.
Experimental studies are based on few subjects, are generally of short duration, and may involve designed foods not consumed by or available to free-living individuals. Even in feeding studies there are limits to the ability to change one facet of the diet while holding the rest constant. For example, if you are going to reduce the total fat content of the diet, then the protein or carbohydrate content will have to be increased if the total energy in the diet and body weight are to remain constant. Moreover, experimental studies rarely look at the effects of diet on a clinical endpoint (eg, heart attack risk).
Clinical trials
Design.
Clinical trials are prospective experiments. They include all the aspects of prospective studies and, in addition, they are usually randomized and double-blind (
21
). Clinical trials are considered by many biomedical scientists and policy makers to represent the ultimate test of causality. In a clinical trial, one group of individuals is randomly assigned to be the treatment group and the other group is randomly assigned to be the control group. Both groups are followed over time and the safety and efficacy of the treatment are evaluated. Examples of clinical trials include testing whether medications that lower blood cholesterol concentrations lead to a reduction in CHD and testing whether supplementing the diet with ß-carotene decreases the risk of lung cancer.
The distinction we have drawn between feeding and metabolic-ward studies and clinical trials is somewhat artificial; both types of studies are experiments, and although we usually think of clinical trials as being large studies with long periods of follow-up, there are many types, both large and small, some of which include feeding special diets. But because feeding and metabolic-ward studies are such an important and distinct part of nutritional epidemiology we thought it necessary to discuss them separately.
Strengths.
Because clinical trials are randomized experiments, the results are less likely to be due to chance or confounding. Thus, they can be used to establish the safety and efficacy of a treatment and to suggest prevention strategies. By showing the effectiveness of a treatment in curing a disease, modifying its course, or preventing a disease by altering a risk factor we can bolster the conclusion that the risk factor is indeed a cause of the disease.
Weaknesses.
Clinical trials are expensive and require lengthy follow-up. Because they are usually conducted in people at highest risk of the disease in question, to reduce the costs and follow-up time, they are subject to certain constraints that limit the extent to which they assess causality. A trial may fail for any number of reasons including
1
) the study's prescribed treatment dose was too high or too low,
2
) the side effects of the treatment outweighed its positive effects,
3
) the disease had progressed too far,
4
) the wrong delivery system was used,
5
) the participants would not use the treatment or use it properly or the dropout rate was too high,
6
) the follow-up time was too short,
7
) the treatment group became less compliant while the control group adopted the treatment, and
8
) the initial hypothesis was flawed. If a trial is successfully designed, implemented, and completed, its results provide evidence regarding the efficacy of a clinical intervention. A dilemma in interpreting the findings of clinical trials is evaluating the generalizability of the results. For example, are the beneficial effects of high calcium intake on bone density in middle-aged women applicable to women or men over age 65 y?
MEASUREMENT OF DIETARY EXPOSURE
Food consumption and how consumption affects nutritional status and disease risk are highly complex. If the goal is to study the relation of dietary exposure to disease risk, dietary intake cannot be assumed to be the same as dietary exposure. Deciding how dietary exposure should be measured in any one study requires careful consideration of the key hypotheses. Depending on the study design and the hypotheses, any number of methods can be used. Because of space limitations, the discussion here will focus on dietary survey methods designed to assess food and nutrient intake. Biomarkers of intake or nutritional status, anthropometric measures, and clinical measures are discussed elsewhere (
22
,
23
).
Selection of the dietary survey methodology
Studies in nutritional epidemiology focus primarily on the relation between an individual's usual intake and the risk of disease (
24
). The most controversial aspect of these studies is often the selection of the dietary survey methodology (
25
). For large studies, the only practical methods are the 24-h recall, hand-written food records, food records using precoded or list-based forms, and food-frequency questionnaires, which are also precoded or list-based (
Table 2
). The choice must be based on the design and objectives of the study and on the type of information required (
27
).
|
In principle, any of several dietary assessment methods may be used. Because of the large amount of within-person or day-to-day variation in dietary intake, one 24-h recall or food record is not considered to be a valid estimate of an individual's usual relative or absolute intake ( 25 ). Because food-frequency questionnaires are designed, in theory, to estimate an individual's usual intake over some period of time (eg, 1 y) and because they have low interviewer burden and can be processed rapidly, they are currently the method preferred by most epidemiologists. However, the measurement errors characteristic of food-frequency questionnaires make it difficult to assess the distribution of intakes or to rank an individual's intake correctly ( 28 – 30 ). Furthermore, 2 or 3 replicate 24-h recalls or food records are just as useful in estimating the association between diet and disease as a food-frequency questionnaire for most nutrients ( 29 , 31 ). Notable exceptions in which food-frequency questionnaires may be the preferred method are the measurement of intakes of alcohol, vitamin A, nutrient-dense food sources of vitamin A and carotenes, and foods that tend to be consumed rarely.
Currently, no dietary assessment method is likely to measure the true intake for an individual ( 32 – 34 ). All of the methods involve measurement error; thus, the choice of a method must be made on the basis of assessed relative validity. For example, regardless of the method, energy intake is underreported by as much as 20–30% ( 33 ). It is unclear, though, whether all macronutrients are underreported to that extent. For example, in a recent study men overreported protein intake and women reported it accurately as compared with protein intake estimated from urinary nitrogen excretion ( 35 ). Without information about the direction of measurement bias, studies cannot adjust observed nutrient levels and estimate the study or population intake as can be done by calibrating a biological measurement to a standard value. It is also likely that the extent to which intakes of energy and its macronutrient components are underreported will vary from study to study. That is, it is unlikely that there is a universal fixed rate at which macronutrient intakes are underreported. Because memory is less of a problem with one or several days of food records and 24-h recalls, they provide relatively more accurate intake assessments than do food-frequency questionnaires, which attempt to assess usual intake over a longer period of time ( 16 ). Food-frequency questionnaires have been used in case-control studies to estimate usual intake in the distant past, but such assessments of diet appear to be influenced by current intake and hypothesized assumptions ( 36 ).
If the hypotheses being studied involve only a few select nutrients, precoded food records will suffice or, depending on the food or nutrient, food-frequency questionnaires may be appropriate. For example, much of our knowledge about the protective effects of fruit and vegetables with regard to cancer risk comes from case-control studies in which dietary intake was based on foodfrequency or dietary history questionnaires. Because of the complex ways in which diet and nutrition may affect risk of disease, it is important to have as much dietary intake information as possible. Precoded instruments, regardless of whether they are used as food records or food-frequency questionnaires, generally do not allow for recording of information on food preparation methods, foods consumed together as meals and as mixed dishes, type of food, brand name, packaging, ingredients, and the size of individual portions in the detail that can be obtained from a 24-h recall or a hand-written food record ( 25 ). Another limitation of precoded instruments is that the number of foods on the form is limited. Therefore, 24-h recalls and hand-written food records give the researcher much more flexibility to address dietary issues and questions that were not anticipated during the design of the study. This is not meant to imply that 24-h recalls or food records can be used to study all of the complex issues related to endogenous and exogenous influences on nutrients and other dietary components, but these methods do provide information that is not usually provided by food-frequency questionnaires. If and when biomarkers of absolute nutrient intake become more available, detailed study of the effects of nutrients or particular metabolites of nutrients and their relation to disease may become possible ( 23 , 24 ).
Data collection procedures and nutrient databases
The selection of an appropriate dietary data collection method does not eliminate all issues related to long-term evaluation of dietary intake for a study or a population. Procedures for collecting and coding dietary intake data and selecting an appropriate food composition database with valid nutrient data (including specific brand names and time-sensitive data) are required to convert dietary intake data into nutrient intake data (
37
). Nutrient databases are extremely complex and maintaining one can be very expensive and time-consuming. Users of databases need to understand their limitations, including:
1
) databases need to be maintained to keep up with the rapid introduction of new foods and the continuous updating of nutrient composition data for foods already in the database,
2
) nutrient values are often assigned rather than determined in the laboratory, and
3
) missing data is a problem in nutrient databases (
37
). Moreover, in studies of multicultural, multilingual populations, it is also important to recognize that cultural factors and language influence the ways in which foods are reported by members of different ethnic groups.
|
INTERPRETATION OF EPIDEMIOLOGIC DATA AND ASSESSMENT OF CAUSALITY |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
The effects of measurement error can be enormous; an example is the study of the relation of diet to risk of CHD ( 39 ). The proposed mechanism of the diet-heart hypothesis is that the proportions of total energy from saturated fatty acids and to a lesser extent dietary cholesterol have a direct effect on serum total and LDL cholesterol concentrations, that atherosclerosis in the coronary arteries is directly related to those concentrations and the length of time the arteries are exposed to those concentrations, and that CHD is primarily a result of atherosclerosis.
One of the primary criticisms of the hypothesis was the failure to find a correlation between the percentage of energy from saturated fatty acids and serum total cholesterol in cross-sectional studies ( 18 , 19 , 39 ). Although Keys ( 40 ) and others pointed out that this problem was largely due to within-person or day-to-day variation in individual intakes, the criticism persisted. As late as 1980, 2 reports, one from the Food and Nutrition Board of the National Academy of Sciences ( 41 ) and the other from a task force sponsored by the American Society for Clinical Nutrition ( 42 ), cited this problem as a part of their rationale for withholding their support of the hypothesis.
A second challenge to the hypothesis was the need to show in prospective studies that dietary intakes of saturated fatty acids and cholesterol are directly related to CHD risk. The first study to show an effect was the Western Electric Study in 1981 ( 43 ). This was followed by data from the Honolulu Heart Program in 1984 ( 44 ), the Zutphen Study in 1984 ( 45 ), and the Ireland-Boston Diet-Heart Study in 1985 ( 46 ). It was difficult to show an association in free-living individuals due to the lack of variability in fat consumption within single populations, within-person variation and other measurement error associated with the dietary survey methods, and the fact that serum cholesterol concentrations are affected by a number of factors including diet. However, as important as these results were, they came late in the debate. The National Institutes of Health–sponsored Consensus Development Conference on Lowering Blood Cholesterol to Prevent Heart Disease, published in 1985, set to rest the arguments opposing the diet-heart relation ( 47 ). The primary stimulus for the conference, the Lipid Research Clinics Coronary Primary Prevention Trial, showed that lowering serum cholesterol using bile acid sequestrants led to a significant reduction in CHD events ( 48 ).
In 1967 Ancel Keys ( 40 ) predicted the problems in trying to relate diet to CHD risk in observational studies. He wrote: "Interview and recall methods for dietary surveys can produce valuable data but at best they are only semiquantitative . . . All experience leads to the same conclusion. Within a culturally homogeneous population is it fruitless to attempt to characterize with any reliability the individuals in respect to the nutrient variables relevant to serum lipids, atherosclerosis, and coronary heart disease. This conclusion refers to the methods used so far but the methods are not responsible; the inescapable limitation is the spontaneous variability of the individuals themselves. Recognition of this basic fact will prevent much wasted effort and pointless argument."
To show that diet is indeed related to CHD risk in humans required a 2-stage process (
49
) in which the black box was conceptually split into 2 parts
(Figure 2)
. In one stage, feeding studies in humans in addition to studies in nonhuman species were used to document the relation between diet and serum cholesterol concentrations. In the second stage, epidemiologic studies consistently showed a direct, graded relation between serum cholesterol concentrations and risk of CHD (
50
). These results were augmented by findings from autopsy studies, laboratory research, clinical studies, and additional animal studies, especially in nonhuman primates, that showed a direct relation between diet, atherosclerosis, and CHD (
51
). However, the point is that a direct relation between diet and CHD risk in humans was difficult to show in the observational epidemiologic studies of the time.
|
Because of multicollinearity, it is usually not possible to determine the effects of a particular nutrient independent of the other nutrients with which it is correlated ( 55 , 56 ). For example, McGee et al ( 57 ) found that because the components of energy intake were highly correlated, the effect of fat as distinct from that of protein cannot be shown in observational studies: "If the question is simply whether the dietary variables as a group predict CHD and we are not interested in the exact relation of a particular diet variable to CHD, solutions clearly exist. If, on the other hand, we are interested in interpreting how a particular diet variable relates to outcome controlling for other diet variables, the collinearity of the data appears to be a structural rather than a mathematical problem with no apparent solution." This is the fundamental limitation in observational nutritional epidemiology.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
We hope that the guidelines presented in this article will serve as a starting point for helping clinical nutritionists, primary care physicians, the lay public, and nutrition researchers to understand the strengths and limitations of nutritional epidemiology.
|
FOOTNOTES |
|---|
2 Reprints not available. Address correspondence to CT Sempos, Office of Research on Minority Health, National Institutes of Health, Building 1, Room 255, MSC 0164, Bethesda, MD 20817–0164. E-mail: semposc{at}od.nih.gov.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONOVERVIEW OF NUTRITIONAL...TYPES OF STUDIESINTERPRETATION OF EPIDEMIOLOGIC...CONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C. J. Kavanaugh, P. R. Trumbo, and K. C. Ellwood The U.S. Food and Drug Administration's Evidence-Based Review for Qualified Health Claims: Tomatoes, Lycopene, and Cancer J Natl Cancer Inst, July 18, 2007; 99(14): 1074 - 1085. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W. Yang, T. Kelly, and J. He Genetic Epidemiology of Obesity Epidemiol. Rev., June 12, 2007; (2007) mxm004v1. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. R Trumbo and K. C Ellwood Lutein and zeaxanthin intakes and risk of age-related macular degeneration and cataracts: an evaluation using the Food and Drug Administration's evidence-based review system for health claims. Am. J. Clinical Nutrition, November 1, 2006; 84(5): 971 - 974. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S L Jowett, C J Seal, M S Pearce, E Phillips, W Gregory, J R Barton, and M R Welfare Influence of dietary factors on the clinical course of ulcerative colitis: a prospective cohort study Gut, October 1, 2004; 53(10): 1479 - 1484. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 U. Palaniappan, R. I. Cue, H. Payette, and K. Gray-Donald Implications of Day-to-Day Variability on Measurements of Usual Food and Nutrient Intakes J. Nutr., January 1, 2003; 133(1): 232 - 235. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. Byers, B. Lyle, and W. Participants Summary statement Am. J. Clinical Nutrition, June 1, 1999; 69 (6): 1365S - 1367S. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
