![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
2
*
Agricultural Research Service, U.S. Department of Agriculture, Washington, DC 20250,
Office of Dietary Supplements and the
Office of Prevention Research and International Programs, National Institutes of Health, U.S. Department of Health and Human Services, Bethesda, MD 20892
2To whom correspondence should be addressed. E-mail: piccianm{at}od.nih.gov
 |
ABSTRACT |
|---|
 TOP ABSTRACT CURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
KEY WORDS: • ISU method • IOM method • total nutrient intake • food intake • dietary supplement intake
The primary objective of the integrated dietary component of the What We Eat in America–National Health and Nutrition Examination Survey (NHANES)4 (the "integrated survey") is to provide estimates of food and dietary supplement intakes that will guide dietary assessment, planning, research and public health policy. The success of this approach relies on the use of scientifically valid and justified techniques to develop estimates of intakes.
The first portion of this paper includes coverage of current and acceptable procedures for estimating usual daily intakes of nutrients, foods and other substances from the information provided by the integrated survey. Procedures suitable for estimating intakes of them using both traditional statistical techniques and modeling are also considered. Other uses of data from the integrated survey, such as those involving risk and safety assessments, are also briefly mentioned. Finally, implications for research, policy and practice are presented. The second section summarizes the deliberations and recommendations of discussion groups assigned to address specific questions regarding the current status of the process to develop reliable and accurate estimates of intake. Their recommendations are provided, both with respect to the content and process for updating and improving these estimates in a manner that is responsive to the public health needs of the U.S. population.
|
CURRENT AND DESIRABLE PROCEDURES FOR ESTIMATION OF USUAL DAILY INTAKE |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
) included recommendations that usual nutrient intake distributions be estimated from food sources based on at least 2 d of dietary intake data using the method developed at Iowa State University (ISU) (2
). In the integrated survey currently in the field, usual food intakes are estimated from two independent (nonconsecutive) 24-h recalls but intakes of dietary supplements are estimated from a report of dietary supplement use over the past month; actual dietary supplement intakes on specific survey days are not available. The following section covers methods that may be used for improving estimates of intake distributions from these data. Estimation of intake of foods, nutrients and other constituents
Dietary data analyses present challenges that are difficult to address with current data collection methodology. The limitations of dietary intake data, chief among them underestimation of intake, are recognized and must be dealt with in interpreting survey data. For example, energy and protein intakes reported in diet records for selected small samples of adults have been reported in numerous studies to be underestimates of 0–37% compared with energy expenditure as measured by doubly labeled water or protein intake as measured by urinary nitrogen (3
–13
). The results from a few studies have been interpreted to suggest that those reporting low energy intakes also have intakes that are lower in absolute amounts of most nutrients than those reporting higher intakes (14
), whereas they are higher in percentage of energy from protein (13
,14
) and lower in percentage of energy as carbohydrate (14
–16
). Those who report low energy intakes may also report lower intakes of desserts and sweet baked goods, butter and alcoholic beverages but higher intakes of grains, meats, salads and vegetables (14
).
Underreporting on food records probably results from the combined effects of incomplete recording and the recording process on dietary choices leading to under eating (3
,11
). Reporting bias may also be present in some individuals. The highest levels of underreporting have been found among individuals with higher body mass indexes (2
,6
,7
,14
,15
), particularly women (4
–6
,14
,17
,18
). This effect, however, may be partly because heavier individuals are more likely to be dieting on any given day (19
). Other research shows that demographic or psychological indexes such as education, employment grade, social desirability, body image and dietary restraint are also factors related to underreporting on diet records (4
,13
–15
,19
–21
).
Similar to findings based on food records, biological markers such as doubly labeled water and urinary nitrogen reveal a tendency toward underreporting of energy and protein in the range of 13–24% for 24-h dietary recalls (11
,22
–24
). However, overreporting of protein intake from 13 to 25% depending on body mass index was reported in one study (25
). In national dietary surveys, data suggest that underreporting may affect up to 15% of all 24-h recalls (26
,27
). Recent research on 
500 participants using doubly labeled water and measurements from two 24-h urine samples showed that nearly one-fourth of both men and women were defined as energy underreporters on a 24-h recall (5-step method); for protein this figure was 12% (24
).
Differential underreporting exists for different nutrients as well as by groups with different demographic and psychosocial characteristics. Underreporters tend to report fewer numbers of foods, fewer foods consumed and smaller portion sizes across a wide range of food groups and tend to report more frequent intakes of low fat and diet foods and less frequent intakes of fat added to foods than do those who do not underreport (27
). Factors such as body mass, gender, social desirability, restrained eating, education, literacy, perceived health status and race and ethnicity are also related to underreporting in recalls (11
,19
,20
,23
,27
–29
).
Numerous other limitations of dietary data have been reported and have been reviewed elsewhere (30
), including overreporting.
Estimation of intake of dietary supplements
Assessment of the use of dietary supplements is important because supplement use may independently alter risks of several chronic and other diseases. Also, supplements contribute a major and sometimes larger proportion of intakes of micronutrients or bioactive food components than diet alone. Results of nutritional epidemiologic studies are likely to be misleading if dietary supplement intake is not accurately quantified, particularly for individuals who obtain a large share of their total nutrient intakes from supplements. Dietary supplement consumption is discretionary. Usage patterns may be different from those of commonly eaten foods (e.g., periodic rather than constant and more like those of foods that are rarely consumed). In some study populations (i.e., cancer survivors) dietary supplement use may be as high as 80% (31
). Little is known about factors such as underreporting errors, reporting bias and the socioeconomic and demographic correlates of dietary supplement use. More data are needed on these topics.
A limited knowledge base exists with regard to the appropriate methodology for obtaining valid and reliable information about dietary supplement use. Determination of how best to estimate intakes of dietary supplements requires knowledge of the error structure encompassed in data collection methodology so that appropriate statistics may be applied to yield meaningful information on usage patterns among various segments of the U.S. population. Intakes of dietary supplements might be ascertained by using methods described below for the analysis of commonly and uncommonly eaten foods, but research is needed to assess this. Many other methods for improving estimates of supplement intake also exist and these also need further testing. However, such estimates may be hampered by problems in data collection and in available dietary supplement databases.
Estimation of intake of food
Estimates of food consumption in the integrated survey are useful for many assessment and planning purposes, including the description of consumption patterns. For example, Healthy People 2010 includes three objectives that focus on food intake (i.e., on variety and quantity of vegetable, fruit and grain product consumption) and five objectives that address nutrient intake (i.e., calcium, folic acid, sodium, saturated fat and total fat) (32
). For all these objectives, estimates of usual intake distributions are required along with the ability to assess changes in intake distributions over time. Infrequently consumed foods may be of interest from several standpoints. They may be associated with risks of certain diseases (e.g., consumption of alcohol, shellfish poisoning) or biohazard risks (heavy metal contamination). Data needs for identifying and assessing the significance of various foods posing food safety risks from the intentional or unintentional introduction of chemical or biological hazards also have to be considered.
The number of individuals needed for accurate estimates of group intakes of particular foods or specific dietary supplements is often much greater than the sample size needed for estimates of nutrients from conventional diets. This is especially true if the foods and dietary supplements are infrequently consumed. Therefore sample sizes may need to be larger for these purposes.
Food consumption data are also needed for risk assessment models, which require both the distributions and prevalence of intakes of specific constituents or individual foods. Modeling is a critical tool used in risk assessments to estimate the risks posed by biological or chemical hazards that may be intentionally or unintentionally introduced to the food supply. Estimates of the amounts of food ingredients or contaminants consumed may also be needed for regulatory purposes. Attention needs to be paid to what can and cannot be obtained from the data for modeling and estimating risk.
Estimates of usual intakes of commonly eaten foods. The estimation of usual single-day intakes of foods shares certain characteristics with the estimation of usual 1-d nutrient intakes. The distributions of daily intakes for many foods are similar to those of daily nutrient intake, such as extreme skew and a high degree of within-individual variation relative to between-individual variation.
The estimation of usual food intakes is complicated; however, because many foods are consumed infrequently by almost everyone in the population, leading to a large number of observed zero intakes on 24-h recall measurements with small numbers of replicates. A significant fraction of observed zero intakes comes from individuals who never consume the food in question (true nonconsumers), whereas the rest come from individuals who sometimes consume the food but nevertheless did not consume the food on the days when their intake was recorded (occasional consumers).
Estimates of consumption of specific foods. To evaluate dietary intakes relative to some recommendation or standard (e.g., Estimated Average Requirement [EAR]), it is necessary to have estimates of the distribution of usual dietary intakes rather than the distribution of 1-d or 2-d mean intakes. For the estimation of usual intakes, information on the consumption propensity (i.e., frequency of intake) of the food of interest is needed. If the food is frequently consumed, data from two 24-h recalls may be sufficient to derive this consumption propensity and estimate usual intake. If it is rarely consumed, then two 24-h recalls cannot supply sufficient information; a large number of days of data collection would be needed for very rarely consumed foods. However, capturing the rare eating occurrence is impractical.
Although numerous 24-h recalls for each individual might seem like the solution to this problem, they are not practical for national nutrition monitoring because of the expense and respondent burden involved. Nonetheless, it is critical to various nutrition monitoring goals, such as being able to estimate the usual intake of rarely consumed foods, and tracking progress on a given national health objective, such as increasing the consumption of dark green or orange vegetables.
The National Cancer Institute pilot-tested the use of a nonquantified food frequency questionnaire in the 2002 integrated survey to ascertain consumption propensity for a range of foods. This instrument is based on a widely used food frequency questionnaire—the diet history questionnaire, also developed by the National Cancer Institute—with some modifications. The major difference between the propensity questionnaire and the diet history questionnaire is that the propensity questionnaire does not ask about portion size, only about frequency of consumption. The data collected using this instrument are then used to adjust the data obtained from the 24-h recalls to make estimates of usual intakes for the range of foods. Good response rates and operational feasibility for the propensity questionnaire were both demonstrated in the pilot study that was concluded late in 2002.
Estimates of consumption of food groups. Because individual food items are so numerous and varied in our current food supply, some degree of aggregation (grouping) of foods consumed is necessary for most analyses. If the grouping is broad and foods within the group are commonly consumed, even a small number of replicates of 24-h recalls may be enough for the food group to be reported as consumed by virtually all true consumers. For example, data from the 1994–1996 Continuing Survey of Food Intakes by Individuals (CSFII) indicate that over a 2-d period, 99% of individuals aged 2 y and over ate at least one serving per day of foods from the bread, cereals, rice and pasta group of the U.S. Department of Agriculture’s Food Guide Pyramid. In such cases, estimates of usual food group consumption can be accomplished with the same methods as used to estimate usual daily intake of nutrients because the propensity to consume can be treated as constant at 100%. Foods must be assigned to these various groups.
Less commonly eaten and unusual foods.
In most cases, foods or food groups of interest may not be consumed so frequently, and distinguishing true nonconsumers from occasional consumers with usual daily intakes that are not zero becomes more important. Expressing usual 1-d intake as the product of the propensity to consume on any given day times the consumption-day average intake (33
) is equivalent to defining usual intake as a long-term daily average, but it provides a foundation for statistical estimation of usual food intake. Estimation of usual intake must account for differential consumption propensity in the population as well as day-to-day variability in the amount consumed. Estimation of usual intake for groups of foods must also account for correlations among the consumption propensities for the individual foods that the group constitutes. Estimation may be further hindered by the fact that propensity to consume may be related to the amount consumed on consumption days.
Types of distributions needed.
An estimate of the average usual food intake for a subpopulation can be made from data on the average 1-d intake over all sampled individuals who are in the subpopulation of interest. Time trends in average usual food intake can also be tracked using this approach. Simple averages of 1-d food intakes are not sufficient for estimating distributions of usual food intake or related quantities, such as the fraction of the subpopulation with usual food intake above or below some cutoff value. The method of Nusser et al. (33
) may be used for estimating distributions of usual food intake for which the propensity to consume the food is independent of the amount consumed. However, this method requires a fairly large number of 24-h measurements on each sampled individual and, for rarely consumed foods, it may overestimate the proportion of the population that never consumes the food.
Recommendations for future work.
Future work on estimation methods for usual food intake will require additional sources of information on propensity to consume because the number of 24-h measurements per sampled individual that can realistically be expected from a large-scale survey (two or fewer) is not enough for the Nusser et al. (33
) method to produce reliable estimates. One such source of information could be a form of food frequency (propensity) questionnaire that allows the stratification of the population into classes of similar frequency of intake.
From the statistical standpoint, infrequently consumed dietary supplements are possibly similar to infrequently consumed foods. However, from the cognitive and usage standpoints they may not be so; for example, dietary supplements may often be used as medicines as well as for prevention. Consumer behavior and circumstances surrounding consumption may be different for those who use dietary supplements for preventive and curative purposes, and both of these uses may differ from patterns of infrequent consumption of a food. More research is needed on this topic.
Estimation of intake of other intentional and unintentional food ingredients
Risk and safety assessments. Both the Center for Food Safety and Applied Nutrition (CFSAN) of the Food and Drug Administration (FDA) through its various offices and the Office of Public Health and Science (OPHS) of the Food Safety Inspection Service (FSIS) of the U.S. Department of Agriculture are required to estimate the intake of various food components as part of safety and risk assessments. An important activity common to these agencies is the accurate assessment of exposure as an important component of risk analysis and subsequent safety decision regarding the public health impact of microbial pathogens, pesticides and other contaminants in foods as well as the introduction of new food and color additives. These food components can include food additives, generally recognized as safe (GRAS) ingredients and contaminants in food that are present either as additives or are naturally occurring. Risk assessment and subsequent safety decisions may be conducted either premarket (as for most food additives) or postmarket (as for GRAS ingredients or contaminants).
Chronic or lifetime exposure to substances is estimated because most toxic endpoints associated with consumption of substances in foods result from continued long-term bodily insults (e.g., methylmercury in fish). Usual intake is frequently assumed to be synonymous with chronic or lifetime intake. Another need is for estimates of exposure to substances that may be unlawfully present in foods that are evaluated for recall purposes. In cases when the toxic endpoint results from a single exposure (e.g., sulfites, E. coli O157:H7), information on acute or actual intake is also required.
Types of distributions needed for risk and safety assessments. Regulatory agencies need several estimates of intakes of foods that may be consumed, frequently and infrequently, as part of each person’s diet. The current survey allows reasonable estimation of single eating occasions. However, the 2-d food intake information found in the integrated survey is not adequate for estimating usual intake of additives and ingredients for infrequently consumed foods, such as shellfish and alcoholic beverages. Unlike nutrients that are commonly present in all diets, contaminants or additives are often present in foods and diets only occasionally. The true distributions of intake of these substances are expected to be skewed to a far greater extent than those for nutrients.
Regulatory decisions that result from the analyses must be scientifically defensible and, consequently, a reliable source of intakes is needed. Some combination of the currently available survey data with an indication of longer-term frequencies of consumption may allow a better estimation of usual intake for infrequently consumed foods or contaminant substances appearing in food infrequently.
Some current microbiological risk assessments use distributions of the weighted percentiles of food consumption with the intent being to more closely estimate dietary intake of the U.S. population. Data are needed for risk assessment to adequately characterize the annual number of servings of foods in food categories. Currently used estimates are backed by numerous assumptions that may not be valid.
Examples of uses for regulatory purposes.
Examples of uses of food consumption data for work at CFSAN and OPHS are found in Table 1
. In each example, the commodity and ingredient recipe file are important for estimating consumption of raw agricultural commodities and ingredients. This is relevant in all the areas of exposure assessment, whether for contaminants, pesticides or toxic elements. CFSAN and OPHS often need to develop a distribution to describe consumption rather than just estimates of the mean. Distributions are routinely used in safety assessments as well as for input into more complex probabilistic estimates of exposure and risk. CFSAN and OPHS often need to estimate consumption for specific subpopulations (especially for children, women of childbearing age and the elderly), depending on the relevant toxicity endpoint of the contaminant in question and the likely pattern of consumption.
|
Additional estimates required from dietary data for regulatory purposes
Estimating consumption of commodities. The ability to estimate consumption on a commodity basis is particularly important for evaluating dietary intake of substances such as contaminants and pesticides. For both domestic and imported food products, many contaminants and pesticides are regulated and monitored on commodities (raw or semiprocessed) rather than foods as consumed. Regulatory limits for contaminants in foods are usually set on a commodity basis, and monitoring and other concentration data are usually derived from samples of raw agricultural commodities rather than multiingredient, processed foods. Continuing support of this commodity database is essential for estimating dietary exposure to contaminants and pesticides from the current NHANES and subsequent surveys.
Support for regulatory and policy decisions.
The FDA uses dietary data for food fortification evaluation, development of food labeling regulations and dietary supplement regulations, assessment of dietary supplement use and composition and assessment of trends in dietary knowledge and behaviors (Table 2
).
|
). The interagency effort uses dietary data to track progress toward 2010 targets for several key objectives aimed at promoting health and reducing risk of chronic and other disease associated with diet. Specifically, they aim to increase the U.S. population’s consumption of fruit, total vegetables and dark green or orange vegetables, total grain products and whole-grain products and calcium and to decrease consumption of total fat and saturated fat and sodium. In addition, there is an objective to increase consumption of folic acid by women of childbearing age to reduce the risk of neural tube defects in children. Additional needs to support regulatory and policy decisions. Consumption data on foods and dietary supplements are critical to meet the needs of CFSAN‘s Office of Nutritional Products, Labeling, and Dietary Supplements. CFSAN needs to assess total usual intakes of nutrients from food and dietary supplements by various population groups in the United States for nutrition monitoring and for evaluating the appropriateness and the safety of nutrient fortification of the national food supply. The FDA Office of Science and the FSIS Office of Public Health and Science need dietary data that provide distributions of dietary intake and estimates of the annual number of servings of a food or foods for the exposure assessment component of risk assessments.
Currently, the FDA is the focal point of a variety of infant food issues. CFSAN‘s Office of Food Additive Safety needs consumption data to evaluate safety and exposure issues when new ingredients are proposed for addition to such foods. CFSAN also needs exposure estimates for new food containing bioactive constituents. The impact on national consumption patterns of an additional or substituted constituent in foods (e.g., stanols and sterols in margarine-like products) needs to be evaluated continually.
CFSAN and OPHS need dietary data for various age and gender groups on the intake of various foods to add to the science base on issues related to substances in those foods (e.g., caffeine, soy products, nitrites). Dietary intake data for foods and dietary supplements are also needed to guide policy and regulatory decisions and to support proposed and final rules related to the nutrition labeling of products (e.g., trans fatty labeling; reference amount and serving size of breath mints and candies; health claims; nutrient content claims; daily intake reference values for foods and dietary supplements, such as vitamin K).
There is a need to explore the relationships between nutrient intakes and biochemical (e.g., serum level), clinical or anthropometric measures (e.g., bone mass). Such research generates data needed for nutrition monitoring and for developing regulatory policies on food fortification and health claims. Examples of relevant data needs are listed in Table 3
.
|
|
STATISTICAL ANALYSES OF INTAKE DATA |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
Other data sets
Other data sets commonly or potentially used in conjunction with the integrated survey data set are health inventory information, disappearance and availability data sets, market share by brand information, Economic Research Service marketing data and the FDA’s Total Diet Study.
Data needed
Some data users need more information on subgroups than is likely to be available from 1 or 2 years’ worth of data from the integrated survey. For example, sampling is such that reliable estimates on subgroups such as pregnant and lactating women will take many years to collect unless these groups are oversampled. If geographical subgroups are needed, current sampling methods will need to be modified.
|
SUMMARY OF DISCUSSION GROUP DELIBERATIONS AND RECOMMENDATIONS |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
), included members from the range of stakeholder groups who attended the workshop, who represented the spectrum of federal agencies, academic programs and organizations involved in the national nutrition monitoring. Estimating usual intakes
Nutrients.
The discussion groups were in agreement that the methods described in the IOM report Dietary Reference Intakes: Applications in Dietary Assessment (1
) are an improvement over what has been done in the past and are reasonable to follow in planning future analyses of the integrated survey. However, the discussion groups cautioned that greater awareness is needed about when it is appropriate to apply the Dietary Reference Intake (DRI) methodology and how to apply it correctly. Some techniques require more evaluation. For example, the multivariate approach for comparing the prevalence of inadequacy in groups that differ on several characteristics needs to be evaluated to assess how it performs in practice.
The discussion groups endorsed the ISU method (2
) as the best method available for estimating usual nutrient intake from NHANES 1999–2001 and from the integrated survey in 2002–2004. For nutrients consumed daily by most people, the method requires a second 24-h food recall for at least a portion of the sample. For foods or other dietary constituents that are consumed daily by most people, the ISU method can be applied using two 24-h recalls for both the nutrients in foods and also for other commonly consumed constituents (such as pesticide residues present in foods consumed frequently).
Foods. The discussion groups suggested that one possible approach for obtaining estimates of infrequently consumed foods or dietary supplements is to collect information both on amounts of food consumed and frequency or propensity to consume various foods over a given period. However, the items or food groups of interest must be preselected to implement the collection of such information. Foods selected for attention vary depending on the purpose and would differ if the focus was, for example, on nutrients in foods or on toxins in foods. The foods that need to be preselected for further detail are often not known far enough in advance to be included in interview schedules. Estimation of usual intake of foods not consumed daily by most people requires some method for distinguishing individuals who never consume the food from those who do but did not consume it during the 2 recall days. A food propensity questionnaire that was pilot-tested by the National Cancer Institute for inclusion in the integrated survey may help to solve this problem.
The discussion groups noted that the approaches that are proposed above are not likely to produce reliable estimates of usual intakes for infrequently consumed food items such as shellfish or good estimates for very disaggregated food groups such as kale, liver or fresh tuna fish unless these foods are included. Additional questions and probes on these or other surveys will be needed to obtain information at this level of detail.
Dietary supplements. The discussion groups viewed the ISU method as potentially applicable to the estimation of usual intake not only of nutrients from foods but intake of dietary supplements and constituents consumed less frequently than daily. In the case of dietary supplements, valid application of the ISU method would require two 24-h recalls and a propensity questionnaire to distinguish between individuals who never consume the dietary supplement and those who do consume the dietary supplement but did not consume it during the recall days.
The discussion groups noted that development of a measure of intraindividual variability for dietary supplements was necessary for the valid application of the ISU method. Without measuring intraindividual variability directly, additional assumptions that individuals can accurately report their usual intake of the item during the appropriate time period must be made. That is, it must be assumed that the current dietary supplement frequency questionnaire accurately captures amounts of intake and that day-to-day variability is zero, which is unlikely because it is known that some people take dietary supplements irregularly.
User needs
The discussion groups were aware that many needs are not being met but also recognized that all user needs cannot ever be met through a single survey. Rather, a system of information sources about dietary intake needs to be developed. The discussion groups identified several pressing user needs:
Risk assessment. Such assessments include estimates of contaminants in certain foods, which require estimates of the annual number of servings of specific foods or food groups. These estimates may be unreliable because they are derived from only 1 or 2 d of intake data and an inordinate number of assumptions. For risk assessments, better estimates of the distribution of the number of servings of infrequently consumed foods or food groups are often needed, and such data are not currently available in the survey. The discussion groups acknowledged the need for a commodity database when the goal is estimating intakes of contaminants and strongly urged that one be created; this database must be maintained, updated and have a stable source of funding.
Intakes of specific dietary supplements. Obtaining information on use of some dietary supplements of high public health priority and in specific subgroups—such as women in the immediate postpartum period, lactating women and participants in various food programs—would be particularly important.
New approaches
Propensity questionnaires.
The use of 24-h recalls to estimate usual intakes of foods that are consumed less frequently than every other day is problematic because their use will miss consumption of a given food item for a large portion of the population. That is, two 24-h recalls, if not supplemented, overestimate the number of nonconsumers. The discussion groups concluded that remedying this situation with additional 24-h recalls would be impractical because too many would be required (35
). An alternative is to include a questionnaire to estimate the proportion of users of infrequently consumed foods.
Propensity questionnaires were deemed to have the best potential for augmenting information from 24-h recalls regarding the probability of consuming certain foods on a given day. Their advantage is that they provide information about propensity with much less respondent burden than numerous 24-h recalls. Unlike traditional food frequency questionnaires, propensity questionnaires are not used by themselves to provide quantitative estimates of the absolute intake of nutrients or foods, but they complement two 24-h recalls. Results of the Observing Protein and Energy Nutrition (OPEN) Study show that the 24-h recalls provided better estimates of amounts of energy and protein consumed than did a semiquantitative food frequency questionnaire (24
). Usual intake estimates for nutrients or other constituents that may not be consumed daily require both recall on 2 d and data on frequency of consumption (propensity), which is obtained from a nonquantified questionnaire. Both use of 24-h recalls of food and dietary supplements and this type of food and dietary supplements propensity questionnaire may provide improved estimates of intakes, although research is needed to validate this assumption.
Statistical analysis issues
Total intake of nutrients. The discussion groups agreed that combining nutrient intakes from foods and dietary supplements is straightforward as long as data are collected using similar survey instruments. Therefore, it is important to assess whether the collection of dietary supplement intake information using 24-h recalls, augmented perhaps by a propensity questionnaire, improves estimates of the probability of consumption of dietary supplements by individuals. However, the discussion groups cautioned that these methods have not yet been implemented and research is needed to assess validity of the existing and proposed alternative methods. Pilot-testing and simulation studies might be helpful in this regard.
With a frequency questionnaire as the sole methodology for capturing intake of dietary supplements, the only option for measuring total nutrient intake is to add an unadjusted estimate of intake of a single recall of dietary supplements to an adjusted intake of nutrients from foods. This estimate might be improved by adding 2 d of dietary supplement recall and a propensity questionnaire for dietary supplements, but this assumption needs to be validated. Intake distributions of dietary supplements are steeper than for foods because they are often measured in tablets or doses.
The problem of foods fortified with various nutrients also needs to be dealt with; moving the dietary intake questionnaire into the home would increase ability to ascertain intakes of fortified foods. Additional research is also needed to refine the propensity questionnaire (36
).
The discussion groups recognized an important difference between nutrient intake from food and from dietary supplements: everyone consumes foods containing nutrients, but a significant portion of individuals in the population consumes no dietary supplements, whereas some consume dietary supplements occasionally and others are habitual consumers of dietary supplements. Thus, measuring intake of nutrients from dietary supplements solely with two independent recalls may not provide the information needed to obtain a reliable estimate of the distributions of total usual nutrient intake. Dietary supplements may also have patterns of use that differ from those of foods. For example, some individuals may take certain dietary supplements such as vitamin C or zinc when they have a cold and others take a dietary supplement daily until they run out and then stop use for several days. Propensity of dietary supplement use is not as straightforward as it is for foods.
Use of a propensity questionnaire might reveal reduced precision in measuring intakes for specific dietary supplements that is not present with a single recall covering a long period. Although the more elaborate measure might be a more accurate estimate, it would also raise troublesome problems of interpretation that are presently not apparent. For example, some individuals may report not taking a supplement on the 2 recall days but report taking a supplement such as vitamin C on the propensity questionnaire. For this case, there would be no information on what strength of vitamin C is taken, and it has not been demonstrated that strength can be reliably reported. Also, intakes of certain nutrients can vary more in supplements than foods (e.g., from 30 to 1000+ mg of vitamin C). These issues highlight the need for more research on methodology for the collection of dietary supplement intake.
People tend to focus on recent intakes regardless of how the question is phrased on frequency questionnaires. The optimal reference period for determining dietary supplement use is not yet known (37
). Dietary supplement use may change seasonally and the issue of changes in intakes over seasons needs to be considered. For some dietary supplements, such as iron, the relevant reference period for assessing health outcomes may be even longer because several years of intake may produce health outcomes very different from those for a smaller number of years. The most appropriate reference period also merits further research.
Adequacy of nutrient intakes.
The discussion groups indicated that the IOM method is satisfactory for assessing the prevalence of inadequate intakes of nutrients. Comparisons with EAR levels provide an estimate of the prevalence of inadequacy, but the discussion groups agreed that biomarkers are also needed to assess overall nutritional status (1
,38
). For assessing the adequacy of nutrient intake, it is important to have a reliable estimate of the usual total nutrient intake distribution and, in particular, of its low and high percentiles that include all sources of nutrients. For total nutrient intake estimation, better data on dietary supplements and fortified foods are needed to improve estimates of high intakes. Either the full probability approach (38
) or the EAR cut-point approach (1
) can then be used to estimate the proportion of individuals with intakes below requirements. The EAR cut-point method described in the IOM report (1
) can be used to assess those at risk for undernutrition. The EAR is set at a level intended to promote health and wellness and decrease the risk of chronic disease rather than simply avoid acute nutrient deficiencies. However, the interpretation of what it means when an individual does not meet the recommended intake is less clear. Evaluation of whether DRI cut-points are reasonable—based on the notion of wellness rather than deficiency—is needed.
Adequacy of patterns of intakes.
Discussion groups noted that biomarkers might be desirable in addition to dietary data to assess adequacy of intakes compared with the food groups in the Food Guide Pyramid (39
). However, it is unlikely that accurate biomarkers of intake of food groups will become available in the near future. There is no distribution of requirements to use for food groups, so applying the IOM method to intake of food groups would not work (1
).
Adequacy for individuals.
The IOM report is focused mainly on population groups (1
), not on assessment at the individual level. The discussion groups cautioned that, although the IOM report proposes an approach that is methodologically sound, it might result in inaccurate estimates for individuals because data on only a few days of intake are available. Usual intake cannot be adequately estimated with two 24-h recalls for individuals, although it may be for groups.
Excessive nutrient intakes.
The discussion groups recognized that there is also an interest in assessing excess intakes. The problem of estimating the proportion of individuals with usual intakes below the UL is similar to that described for estimating the proportion of individuals with intakes below requirements. The IOM method can also be used to estimate the proportion of the population above the UL for some nutrients. However, this percentage indicates only that this proportion of the population may be at risk of excess. This fact is an artifact of the way the UL is defined and, therefore, it makes more sense to assess the proportion of individuals with usual intakes below the UL. For those individuals, it is possible to identify the risk of excess. For further discussion of the approach, see Carriquiry (36
).
Planning nutrient intakes. Discussion groups suggested that data from the integrated survey also should be used to plan so that most people obtain adequate intakes.
Estimating sample sizes. The sample size needed depends on the specific intended purpose. The discussion groups noted that the total sample in the integrated survey is only 5000 persons per year, so if the interest is in subgroups defined by age, sex and race/ethnicity, sample sizes can easily become limiting. It may be possible to include additional samples in the integrated survey for special purposes, such as for evaluating food programs nationwide or obtaining more information on specific age, sex or other life-stage groups, but only if resources are available, such additions are planned and approved in advance and respondent burden remains acceptable. However, there are other alternatives that do not involve the integrated survey. For example, the discussion groups suggested that it may be possible to train people at the state or local level to use the standardized protocols and software developed for the integrated survey for their own surveys of additional populations. The use of common elements in these special surveys will facilitate comparisons between them and the integrated survey and also capitalize on existing survey development costs. However, public-use software must be developed for such purposes.
Combining dietary data from What We Eat in America–NHANES with other surveys. The small sample sizes for some population subgroups in the integrated survey limit the accuracy of estimates of day-to-day variability in intakes. The discussion groups recognized that some users wish to combine data from the integrated survey with other data sets having larger sample sizes for certain subpopulations of interest. A recommended approach was to use estimates from larger similar groups to analyze other small groups of interest. More research using food consumption surveys is needed to estimate the likely degree of error introduced by such an approach.
Bridging studies for time series analyses.
Trend analysis is important to national health monitoring and dietary assessment because methods have changed considerably from those used in earlier surveys (e.g., NHANES I, II and III and the several CSFII surveys). The discussion groups noted that for time series analyses, bridging techniques are necessary to minimize artifacts due to changes in dietary collection methods so that long-term changes in dietary intakes can be examined. When survey methodology changes, survey administrators are urged to include a subsample in the design that is to be measured using the old methodology to adjust for methodology effects. The bridging study of Guenther and Perloff (40
) in the 1987 Food Consumption Survey is a useful approach. Additionally, new analysis methods should be applied to both old data and new data to illustrate the effects of changes in those methods.
Selecting appropriate defaults for use in estimates of intakes. The discussion groups suggested that in some instances information from previous surveys and from Nutrition Facts on food and dietary supplement labels may be useful when defaults are needed. Food and dietary supplement sales data from Information Resources, or AC Nielsen may provide market share information that can be used to weight estimates. However, what is sold may not be synonymous with what is consumed. Market analysis may provide some insights when other information is not available.
Research priorities: methods
The discussion groups identified the following three research priorities:
Methods. 1) Under- and overreporting and reporting bias. Priority areas include determining ways to identify and remove bias associated with under- and overreporting, additional pilot-testing of the propensity questionnaire and development of improved methods for assessing dietary supplement use. Many practical questions must be answered if the propensity questionnaire is to be used, such as whether it is necessary for it to be used only on a small sample of the population or on everyone, how to best link it to the 24-h recall data and whether it should be done on different samples from the same population. Better intake collection methods that will reduce under- and overreporting; modeling approaches, when feasible, should also be considered. 2) Total dietary intake. The issue of how best to assess and estimate total nutrient intake from foods, dietary supplements and other sources is of great importance but requires additional discussion and research. Many different suggestions were made on various methods for collecting dietary supplement information and making estimates, but there was no consensus on a single method and little research is available that compares the results of potential methods. There was consensus that priorities included commonly used dietary supplements, especially those contributing substantially to intakes of key nutrients and botanicals of particular public health significance. More research and discussion are needed to refine priorities for data needs and estimation methods that will provide the best information for each intended use. The best way to assess severity of inadequacy and excess was also identified as an area that needs to be researched.
Database needs. The application of the ISU method relies on sound data on food intake and nutrient databases that are valid, accurate and complete. More detail needs to be provided in the databases and in related food codes to capture intakes of fortified foods, bioactive food constituents, specific ethnic foods and foods prepared in different locations (e.g., restaurants, home). The dietary supplement database could be improved by analytical substantiation. Because the number of dietary supplements on the market is large, priorities must be set for selecting categories of supplements for analysis in the database that is currently maintained by the National Center for Health Statistics. High priority dietary supplements should include those that present a risk of excess intake or those that contribute greatly to intakes of nutrients and especially to intakes of nutrients for which a high prevalence of inadequate intakes exists. For dietary monitoring, the priority dietary supplements should be those containing nutrients. Later, the survey could be extended to other dietary supplements of interest because of their possible health benefits or potential harm.
Policy. The need to inform policy makers about new methods for assessing dietary intakes and how they differ from previous methods was acknowledged by discussion groups. The multivariate model described in the IOM report has not yet been applied to a policy problem and, therefore, it is not clear that the transition to policymaking using all of these tools has been made.
|
RECOMMENDATIONS |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
summarizes recommendations that emerged from the discussions of workshop participants.
|
An interdepartmental memorandum of understanding (IMU) between the National Center for Health Statistics of the U.S. Department of Health and Human Services and the Agricultural Research Service of the U.S. Department of Agriculture should be put in place early in 2003 to enable work to begin and to stimulate progress on the methodological research and other issues that have emerged from this workshop. The IMU should include a plan for prioritized methodological research with assigned responsibilities for each task.
A first step is to ascertain and describe current and ongoing activities. Key players in various agencies need to meet regularly. The effort should begin immediately with small, leveraging interagency agreements. Over the long term, a coordinating body to help oversee data collection efforts across agencies and to help with nutrition monitoring priorities and their implementation could also be useful. However, the present focus should be on the immediate needs for the 2005 survey.
It is important to have an IMU in place that continues interdepartmental collaboration. To achieve the objectives described below by 2005 and to have improved methods for total nutrient intake estimation in place by 2007, appropriate data must be collected in the current survey. Statistical methodologies that pull together all the dietary data for making estimates of total nutrient intake must also be developed and tested. Because the collaborative efforts having to do with dietary data involve several departments, the organizational complexities are considerable and time will be needed to put agreements into place. However, the departments involved have collaborated successfully for many years already on developing and applying food security measurements.
Cross-department, cross-agency dialogue should be continued on a regular basis in working meetings on these methodological issues at the level of the scientists who are actually doing the work. The scientists who are engaged in the analysis of data from the integrated survey have much to learn from each other and from colleagues elsewhere in government who have interests in the same problems. To best use existing resources, it was recommended that even closer collaborations be fostered across federal agencies and departments. A plan is needed for prioritizing the methodological research, assigning responsibilities and determining possible sources of funds. Funds from many sources and the efforts of many scientists in many agencies will be needed to accomplish these tasks. There is no place for duplication of effort.
Currently, resources for methodological research are limited. Therefore it is critical for federal agencies to prioritize key research areas and to avoid duplication.
Immediate priorities
DRI standards. The new DRI are generally accepted standards for assessing intake of nutrients. Continuing analysis for updating databases to facilitate the process of using the DRI standards for dietary assessment is of utmost importance.
Propensity questionnaire and other dietary data. Two 24-h recalls and a propensity questionnaire for infrequently consumed foods were recommended to be included in the integrated survey. The addition of a propensity questionnaire should enhance estimation of usual intake for foods consumed moderately infrequently. This is especially important for foods targeted in Healthy People 2010, such as dark green vegetables. The instruments were tested and will be implemented. An important caveat is that this method will not solve the problems of very rarely consumed foods, fine gradations of foods (e.g., variety of apples or dietary supplements), very small subgroups and underreporting and other biases. Continued research to improve and update food intake collection methodology is also needed.
Pyramid servings standard.
Estimation of intakes often involves comparison of intakes of foods or food groups against a food standard, such as the Food Guide Pyramid. The Pyramid Servings Database has been developed for this purpose (41
). Maintenance of the database and the necessary supporting data should be a priority and be continued. The Food Guide Pyramid Database is essential for assessing progress toward meeting the Dietary Guidelines for Americans and achieving various Healthy People 2010 objectives (32
).
Assessment of dietary supplement intakes. Development of the methodology for assessment of dietary supplement intakes is a priority. Developmental research needs to be done on methods to assess daily and usual intake of dietary supplements. The goal is to collect accurate and reliable data on usual dietary supplement intake. One way to accomplish that goal is to implement two 24-h recalls that include dietary supplements, with an emphasis on nutrient-containing dietary supplements. Later, a dietary supplement propensity questionnaire to deal with the entire range of dietary supplements (those containing nutrients and those containing other bioactive components) could be developed.
A measure of intraindividual variability for dietary supplements is necessary for estimating usual total nutrient intake with the ISU method. If intraindividual variability is not measured directly, it must be assumed to be zero, which is not a plausible assumption. With the present 30-d dietary supplement recall, no estimate of intraindividual variability is available. It is also necessary to assume that reported dietary supplement use is valid and accurately measured. This assumption has also not been validated.
If it is not feasible to obtain two 24-h recalls of foods plus nutrient-containing dietary supplements and a more general dietary supplement propensity questionnaire for all individuals, an alternative is to obtain two recalls of food plus dietary supplements and a dietary supplement propensity questionnaire on a subsample; one recall of food plus dietary supplements and a dietary supplement propensity questionnaire for another subsample; and one recall of food plus dietary supplements for the remaining sample. The alternative approach would provide better estimates than obtaining only recalls or only dietary supplement propensity questionnaires for the whole sample. However, this alternative is not optimal.
Parallel work must begin on developing an analytically based dietary supplement database and a list of priorities for dietary supplements of greatest interest. The categories of dietary supplements in the current NHANES dietary supplement recall and public health goals are useful starting points for establishing these priorities.
Diet and health knowledge questionnaire. Discussion groups also urged that the diet and health knowledge questionnaire included in prior CSFII surveys be considered for inclusion in the integrated survey in the near future and that analyses include associations with intakes of food groups and nutrients.
Other critical methodological research. A model for collecting information on food and dietary supplements that involves not only recalls but also a propensity questionnaire should be tested and validated. Simulated data should be used when needed for testing the concept and analytical approaches to determine implications and ramifications.
Research on over- and underreporting should be continued and expanded. Research is needed to improve methods to correct for over- and underreporting of intake. Dietary data collection methods (i.e., data capture) and the use of biomarkers such as doubly labeled water (which are independent of dietary reporting errors) as well as statistical modeling methods need improvement. Less expensive biomarkers would be helpful because doubly labeled water is too costly for use in a large national survey.
Strategies for improving estimates for target groups of special interest need to be developed. Certain target groups (e.g., lactating women, toddlers) are not represented in large numbers in the national survey, but detailed data on their dietary intakes are needed. For such target groups with small sample sizes in the national survey, combining community level estimates with national estimates may help to improve estimates of usual intake. This should be possible if the same methods for data collection and analysis are used. The plans for a Community Health and Nutrition Examination Survey—and also for other regional or local surveys—provide a possible model for such work. Dietary data collection using data-capture methods from the national survey on a state or local sample would provide better estimates for state or local decision makers. Variability estimates from the national survey could be used to improve usual intake estimation from the smaller sample.
Training and technical assistance. Training and technical assistance on estimation techniques should be fostered. The appropriate and scientifically sound use of dietary data from the survey will require a collaborative educational effort among multiple agencies and nonfederal partners. The software for collecting information on food and dietary supplement intakes should be made available for public use. The ISU method for usual intake estimation is a useful advance that needs to be disseminated more widely. User-friendly enhancements should be considered. These include providing the software along with the public data release.
|
APPENDIX: ESTIMATING USUAL DAILY INTAKES: IMPORTANT CONCEPTS AND STATISTICAL AND MODELING APPROACHES |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
Although dietary intake surveys that use 24-h recalls as the survey instrument collect data on 1 or 2 d of food intake, interest typically is on the usual daily intake of a nutrient or some other food component. This is also true for nutrient intake from dietary supplement sources.
Intakes (measured by 24-h recalls) of a food or dietary supplement usually vary from individual to individual and from day to day. That is, the variability has two components: the between-individual variability (which is of interest) and the within-individual variability (or between days for an individual, which is considered to be a nuisance from the statistical standpoint). The relative sizes of the variability depend on the nutrient in question, the food pattern and whether the individual consumes dietary supplements regularly. For energy, present in nearly every food item, the ratio of the within- to the between-individual variances is much smaller than for vitamin A, which is present in only selected fruits and vegetables. However, if a group of individuals consume a daily dietary supplement containing vitamin A, in that group the ratio of the within- to the between-individual variance of vitamin A consumption might tend to resemble the ratio for energy.
Usual daily intake
is defined as the habitual intake of the component by the individual. In principle this could be estimated as the mean of a very large number of 1-d intakes if such data were available. The number of days required to obtain a reliable estimate of the usual daily intake by averaging observed daily intakes depends on the nutrient, as shown by Basiotis et al. (1
). For vitamin A, for example, the number of days needed to estimate a usual daily intake with an acceptably low standard error is well over 100 when only food sources are considered but is likely to be smaller if intakes from dietary supplement sources are considered. In general, this approach is impractical for all nutrients because of cost and respondent burden, yet it is the only approach available to estimate usual nutrient intake for individuals when assessment is based on dietary intake data alone.
For a population or group, however, statistical approaches can be used to partially remove the "nuisance" variance (day-to-day variance) from the 1-d intake data and estimate a distribution of usual nutrient intakes for the group. The estimated usual intake distribution should have a variance that reflects only the individual-to-individual variability in intakes. Although these statistical approaches facilitate obtaining reliable estimates of the distribution of usual nutrient intakes in a group, equal confidence about assessing individual nutrient intake requires a very large number of daily intake values for the individual. For individual assessment, it is important to complement the information provided by dietary intake with biochemical, clinical, anthropometric and lifestyle information.
The statistical method that adjusts daily intakes (i.e., partially removes the within-individual variance from the daily intake data) is in essence based on a simple premise set forth by the National Research Council (NRC) in 1986 (2
): the daily intake observed for an individual in 1 d is simply the usual daily nutrient intake plus a deviation (or measurement error) from that habitual intake. In this simple model, neither usual intake nor the measurement error is observable, yet interest is in estimating the distribution of usual intakes for the group under study. The 1-d intakes are recorded and for the purposes of this discussion will be assumed to measure the real nutrient consumption (from either foods or dietary supplements) by the individual on that day without mistake. This assumption is rarely satisfied because all dietary survey instruments, to some degree, do not capture daily nutrient intake accurately. Because these inaccuracies cannot be assumed to be random or independent of individual characteristics, the statistical approaches described here are not designed to account for them or to correct for these systematic biases in the data. In this light, the term "measurement error," as it is used in statistics, is often misunderstood as referring to the errors incurred when attempting to capture true daily nutrient intake using the conventional survey instruments (24-h recalls, food frequency questionnaires, etc.). Although it is arguable that these ought to be referred to as the measurement errors (because they are indeed errors in the measurements), the error term in the model above accounts only for the random deviation of the observed intake on any single day from the habitual or usual intake and for other sources of random error. The systematic errors in the measurements typical in data collected via the usual survey instruments introduce a bias in the analyses that is not accounted for in the simple model (3
).
STATISTICAL APPROACHES
The challenges in estimating the distribution of usual daily intake of a nutrient in a group using a single day’s intake data were recognized by NRC in its 1986 report (2
). NRC proposed the simple model presented above and suggested a statistical method for obtaining an estimate of the distribution of usual nutrient intakes in a group of interest. The ISU approach (4
), recommended by the IOM in its 2001 report (5
), is an extension of the method proposed by the NRC in 1986 (2
).
The IOM and ISU approaches to estimating usual nutrient intake distributions both adjust the 1-d intake data to partially remove the day-to-day variability in intakes and produce an adjusted intake distribution with variance that reflects only the individual-to-individual variability in nutrient intake. The actual procedure to obtain the adjusted distributions is different in both methods and so are the statistical properties of the resulting adjusted intake distributions. The discussion here is on data needs, assumptions and uses of the results of analyses rather than the methodologies themselves.
Nutrients consumed regularly from foods Many of the nutrients of interest to nutritionists are present in various food items. When individuals consume a varied diet, most nutrients are consumed almost daily. Some constituents of interest, such as lycopene, are exceptions to this rule because they are present in only very few food items. The largest contributor of lycopene to the diet is tomato, so unless an individual consumes tomatoes or tomato-based foods during the interview days, lycopene consumption may be recorded as zero.
The current design of the integrated survey is appropriate for reliably estimating adjusted nutrient intake distributions. Under the design, a replicate 24-h recall will be obtained from each individual in the sample. As long as the group sample size is sufficiently large, one replicate 24-h recall suffices to estimate the day-to-day variability in intakes needed to adjust daily intakes and obtain an estimated (or adjusted) usual nutrient intake distribution.
Nutrients consumed irregularly from foods In addition to the replicate observation, the methodology proposed by IOM and ISU relies on a less obvious assumption: that the observed 1-d intake distribution (i.e., the raw distribution of the 24-h data) is smooth (i.e., no pronounced spikes or other irregularities in the distribution used to obtain the estimated usual intake distribution of interest). This smoothness justifies a separate discussion of data needs and statistical approaches for constituents such as lycopene, which may exhibit a spike at zero in the 1-d consumption distribution. This spike occurs because many respondents will not have consumed tomato products on either of the 2 interview days. It is important to distinguish the real nonconsumers of tomatoes (who would presumably report no tomato consumption on either survey day) from the occasional or regular tomato consumers, who by chance may not have consumed the product during the survey days and would therefore also show up as zeros in the data set. The observation for lycopene is very similar to that for consumption data for a food item (e.g., green leafy vegetables). In this case, a typical observed intake distribution looks like a mixture of two distributions: a spike at zero and a distribution of positive consumptions.
Because these types of data violate the smoothness assumption required by the IOM and ISU methods for adjusting intake distributions, the ISU group proposed an alternative approach for estimating usual intake distributions of items that are not consumed daily (6
). The method assumes that in addition to the amount consumed, which varies among individuals in the group, the propensity to consume the food item (or the infrequently consumed nutrient) also varies among individuals in the group. By combining the distributions of propensity to consume with the distribution of amounts consumed, a usual intake distribution can be obtained for the rarely consumed food item or nutrient.
The method just described requires a minimum of three replicate observations for each individual in the sample, a requirement that is inconsistent with the current design of the integrated survey. Promising work that would permit implementing a similar approach is under way at the National Cancer Institute. That approach combines information obtained from a frequency-type question with information obtained from the usual 24-h recalls for the same individual. This permits separation of the true zeros and accidental zeros in the observed consumption data and it then allows calibration of the 24-h recall from information provided by the frequency-type question.
Nutrients consumed from dietary supplements
Data that will be available from the integrated survey to estimate usual nutrient intake from dietary supplement sources are limited because an estimate from a single recall does not provide a measure of variability, and it is difficult to combine supplement information with nutrient intake data from food sources. Interviewers ask respondents whether they consumed any dietary supplement over the previous 30 d and, if so, in what doses and frequency. A replicate observation for dietary supplement consumption will not be obtained. With this limited information on nutrient intake from dietary supplement sources, all that is possible is to estimate a distribution of usual dietary supplement intake in the group by the empirical distribution of self-reported usual dietary supplement intake. The limitations of usual self-reported nutrient intakes are discussed elsewhere (7
).
Data needed to estimate usual intake of nutrients from rarely consumed dietary supplement sources are similar to those for rarely consumed nutrients from food sources or even for food items: a minimum of two 24-h recalls supplemented by a food frequency type of questionnaire that permits evaluation of the observed (and inevitably abundant) zeros in the daily consumption data. However, many nutrient supplements are reported to be consumed daily or nearly daily. The validity of reported information needs to be evaluated.
Total nutrient intake: combining information from various sources
For dietary assessment and for planning intakes for a group or for individuals, it is important to sometimes consider total nutrient intake, that is, nutrients consumed from both foods and dietary supplement sources. This is particularly important for nutrients such as calcium and folate because some individuals obtain a significant portion of their intake of these nutrients from dietary supplements. In NHANES III, 42% of women ages 20–29 y in the sample reported at least some consumption of dietary supplements. An estimated 39.5% of the population took a dietary supplement in the previous 30 d (8
).
Including the intake of nutrients from dietary supplement sources in assessment is particularly relevant because some individuals in the population are taking significant amounts of dietary supplements daily or almost daily. Individuals with adequate nutrient intake are known to also consume dietary supplements (9
). This finding is supported by the information provided by NHANES III showing that for some nutrients the prevalence of nutrient inadequacy in several subgroups does not decrease noticeably when total nutrient intake rather than nutrients from foods is used for estimation. However, the upper tail of the usual intake distribution becomes longer, thereby increasing the prevalence of usual nutrient intakes above the Tolerable Upper Intake Level (UL; the highest level of intake that is likely to pose no risk to healthy individuals). Exceeding the UL is not recommended. Therefore, for assessment and monitoring purposes, nutrient intakes from dietary supplement sources must be added to nutrient intakes from food sources (including fortified food sources) and they serve as an important indicator of the nutritional status of a group.
The statistical approaches to estimating usual nutrient intake distributions from 1-d nutrient intake data require that at least one replicate measurement be available for each individual in the sample. To fit this statistical approach, dietary supplement intake would need to be measured, as food intake is, by using a 24-h recall instrument on 2 nonconsecutive days. If this occurred, combining nutrient intake information from foods and dietary supplements would be trivial and would consist of adding the intake of nutrient X from all sources before analysis. The total 1-d intake could then be adjusted using the statistical approaches described earlier. Neither NHANES III nor the integrated survey provides information on nutrient intake from dietary supplements that are amenable to statistical analysis using either the NRC (2
) or the Nusser et al. (4
) approach.
A rough approach to combining nutrient intake data from dietary supplements and foods has been proposed (5
,10
). First, an estimate of the individual’s usual intake is obtained by using, for example, the method proposed by NRC (2
) or Nusser et al. (4
). Next, an estimate of each individual’s total usual nutrient intake is obtained by adding to this estimate the self-reported usual nutrient intake from dietary supplement sources. The empirical distribution of the sum of these two usual nutrient intakes can then be used as a rough estimate of the usual nutrient total intake distribution. Numerical estimates of those variances can be computed to obtain estimates of standard errors of percentiles of the distribution or even of the prevalence of usual intakes below the EAR or above the UL for the nutrient.
MODELING APPROACHES
Two basic methods can be used to estimate the usual intake of a group: a purely statistical method or a modeling method. With the purely statistical methods (classical or Bayesian) very little is assumed about the process that generated the data other than that the intakes come from a particular statistical distribution (classical statistics) or even less (Bayesian). By contrast, for statistical modeling at least partial knowledge of the data generating process is assumed to exist. For example, there may be reason to believe that the more educated people are, the more their intakes will comply with current dietary recommendations. This means that for an estimate of a group’s average usual intake of, for example, saturated fat, a better estimate will be achieved by analyzing the effect of education (and other factors) on the average usual intake of saturated fat. This is done by the use of modeling. In a more mathematical depiction,
 |
where F means that usual intake of saturated fat (UISF) depends on education and other factors. Some of these other factors may be known or may be available in the survey data.
Techniques
Estimation of average usual intakes by modeling 1-d intakes typically uses a regression model, the most common one of which is the multiple linear regression model. For example, if in addition to education, the group’s average usual intake of saturated fat is affected by income, age, gender and race coefficients. Depending on those coefficients, the average usual intake of each subgroup would be estimated by
 |
where the b is the coefficient to be estimated (the independent effects of each factor on usual intake) and e is an error term—the net effect of all the other factors that were left out of the model, either because they were not known or because no data existed for them. After these coefficients are estimated, the average usual intake of any group according to education level, income, age, gender and race can be calculated.
This calculated or predicted average usual intake always follows a normal distribution. That is because each estimated b is a weighted average of many random variables (the observed 1-d intakes) and thus, according to the central limit theorem, each estimated b itself has a normal distribution. Therefore, the predicted intake has a normal distribution, being the weighted average of several normal random variables.
The practical importance of this observation is that it allows testing for differences in average usual intake between population subgroups with differing education, income, age, gender or race. Note that the usual intake of one individual can be estimated only with the assumption that the individual behaves exactly as the average or typical individual in this particular education, income, age, gender and race group. This approach is suitable only for modeling average usual intake, which can be well approximated by the average of many 1-d intakes as previously stated. In general, when there is extensive day-to-day variation in consumption, this simple model is not suitable for estimating other characteristics of the usual intake distribution, such as the percentage of individuals whose intake is below some cutoff point. However, extensions to this type of modeling allow such estimation.
Although this method works well for intakes observed for each individual more or less daily, it becomes more challenging, albeit feasible, when intakes are infrequent or just not in the individual’s consumption set. To estimate usual intakes correctly in such circumstances, whether the intake data are missing because the individual never consumes the item or did not consume it during the survey observation period, needs to be known. Unlike the situation where intake data are positive for almost all individuals, the major consequence of not modeling intakes with zero values correctly is that the resulting estimates will be biased in repeated trials.
Advantages and disadvantages of modeling for usual diets and intakes of unusual foods
The basic advantage of statistical modeling to estimate usual intakes lies in the willingness to assume that there is at least a partial knowledge of the process that generates the intake data. If assumptions about the process are correct, then more information is incorporated in the estimation method and better estimates (i.e., less variable estimates) should be achieved than would be achieved without this knowledge. If the assumptions are incorrect, then statistical modeling will generate less precise estimates in general. For practical purposes, estimating individuals’ usual nutrient intakes using modeling and food consumption survey data results in R2 estimates of 0.05–0.20, or only 5–20% of the variation in intakes in the data being explained by knowledge of the data-generating process and the variables available in the data set. Evaluating modeling of food item or food group intakes is less straightforward because of the zero consumption problem mentioned above.
APPENDIX LITERATURE CITED
1. Basiotis, P. P., Welsh, S. O., Cronin, F. J., Kelsay, J. L. & Mertz, W. (1987) Number of days of food intake records required to estimate individual and group nutrient intakes with defined confidence. J. Nutr. 117:1638-1641.
2. National Research Council (1986) Nutrient Adequacy: Assessment Using Food Consumption Surveys 1986 National Academy Press Washington, DC.
3. Kipnis, V., Carroll, R. J., Freedman, L. S. & Li, L. (1999) A new dietary measurement error model and its application to estimates of relative risk: application and four validation studies. Am. J. Epidemiol. 150:642-651.
4. Nusser, S. M., Carriquiry, A. L., Dodd, K. W. & Fuller, W. A. (1996) A semiparametric transformation approach to estimating usual daily intake distributions. J. Am. Stat. Assoc. 91:1440-1449.
5. Institute of Medicine (2001) Dietary Reference Intakes: Applications in Dietary Assessment 2001 National Academy Press Washington, DC.
6. Nusser, S. M., Fuller, W. A. & Guenther, P. M. (1996) Estimation of usual dietary intake distributions: adjusting for measurement error and nonnormality in 24-hour food intake data. Lyberg, L. Biemer, P. Collins, M. Deleeuw, E. Dippo, C. Schwartz, N. Trewin, D. eds. Survey Measurement and Process Quality 1996:689-709 Wiley New York, NY. .
7. Kipnis, V., Carroll, R. J., Freedman, L. S. & Li, L. (1998) The impact of realistic measurement error modeling on the analysis of dietary studies. Eur. J. Clin. Nutr. 52(suppl. 2):S79.
8. Ervin, R. B., Wright, J. D. & Kennedy-Stephenson, J. (1999) Use of dietary supplements in the United States 1984–94. Vital Health Stat. 11:1-14.
9. Hartz, S. C., Otradovee, C. L., McGandy, R. B., Russell, R. M., Jacob, R. A., Sahyoun, N., Peters, H., Abrams, D., Scura, L. & Whinston-Perry, R. A. (1988) Nutrient supplement use by the healthy elderly. J. Am. Coll. Nutr. 7:119-127.[Abstract]
10. Arab, L., Scott, S., Gaudett, M. & Carriquiry, A. L. (2003) Ethnic differences in the nutrient intake adequacy of pre-menopausal American women results from the National Health and Nutrition Examination Survey III. J. Am. Diet. Assoc. (in press).
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
3 P. Peter Basiotis, Mary M. Bender, Bernadette K. Bindewald, Alicia L. Carriquiry, Anne K. Courtney, Nancy T. Crane, Kevin W. Dodd, Katie Egan, Kathleen C. Ellwood, Susan E. Gebhardt, Joanne F. Guthrie, James M. Harnly, Joanne M. Holden, Clifford Johnson, Susan M. Krebs-Smith, Paul M. Kuznesof, Carol E. Lang, Margaret McDowell, Alanna Moshfegh, Pamela R. Pehrsson, Kathy Radimer, Amy F. Subar, Christine A. Swanson and Wayne R. Wolf.
4 Abbreviations used: CFSAN, Center for Food Safety and Applied Nutrition; CSFII, Continuing Survey of Food Intakes by Individuals; DRI, Dietary Reference Intake; EAR, Estimated Average Requirement; FDA, Food and Drug Administration; FSIS, Food Safety Inspection Service; GRAS, generally recognized as safe; IMU, interdepartmental memorandum of understanding; IOM, Institute of Medicine; ISU, Iowa State University; NHANES, National Health and Nutrition Examination Survey; NRC, National Research Council; OPHS, Office of Public Health and Science; UL, Tolerable Upper Intake Level.
|
LITERATURE CITED |
|---|
|
TOPABSTRACTCURRENT AND DESIRABLE PROCEDURES...STATISTICAL ANALYSES OF INTAKE...SUMMARY OF DISCUSSION GROUP...RECOMMENDATIONSAPPENDIX: ESTIMATING USUAL DAILY...LITERATURE CITED |
|---|
1. Institute of Medicine (2001) Dietary Reference Intakes: Applications in Dietary Assessment 2001 National Academy Press Washington, DC.
2. Nusser, S. M., Carriquiry, A. L., Dodd, K. W. & Fuller, W. A. (1996) A semiparametric transformation approach to estimating usual daily intake distributions. J. Am. Stat. Assoc. 91:1440-1449.
3. Goris, A. H., Westerterp-Plantenga, M. S. & Westerterp, K. R. (2000) Undereating and underrecording of habitual food intake in obese men: selective underreporting of fat intake. Am. J. Clin. Nutr. 71:130-134.
4. Taren, D. L., Tobar, M., Hill, A., Howell, W., Shisslak, C., Bell, I. & Ritenbaugh, C. (1999) The association of energy intake bias with psychological scores of women. Eur. J. Clin. Nutr. 53:570-578.[Medline]
5. Sawaya, A. L., Tucker, K., Tsay, R., Willett, W., Saltzman, E., Dallal, G. E. & Roberts, S. B. (1996) Evaluation of four methods for determining energy intake in young and older women: comparison with doubly labeled water measurements of total energy expenditure [see comments]. Am. J. Clin. Nutr. 63:491-499.
6. Black, A. E., Prentice, A. M., Goldberg, G. R., Jebb, S. A., Bingham, S. A., Livingstone, M. B. & Coward, W. A. (1993) Measurements of total energy expenditure provide insights into the validity of dietary measurements of energy intake. J. Am. Diet. Assoc. 93:572-579.[Medline]
7. Black, A. E., Bingham, S. A., Johansson, G. & Coward, W. A. (1997) Validation of dietary intakes of protein and energy against 24-hour urinary N and DLW energy expenditure in middle-aged women, retired men and post-obese subjects: comparisons with validation against presumed energy requirements. Eur. J. Clin. Nutr. 51:405-413.[Medline]
8. Martin, L. J., Su, W., Jones, P. J., Lockwood, G. A., Tritchler, D. L. & Boyd, N. F. (1996) Comparison of energy intakes determined by food records and doubly labeled water in women participating in a dietary-intervention trial. Am. J. Clin. Nutr. 63:483-490.
9. Rothenberg, E. (1994) Validation of the food frequency questionnaire with the 4-day record method and analysis of 24-h urinary nitrogen. Eur. J. Clin. Nutr. 48:725-735.[Medline]
10. Bingham, S. A., Gill, C., Welch, A., Cassidy, A., Runswick, S. A., Oakes, S., Lubin, R., Thurnham, D. I., Key, T. J., Roe, L., Khaw, K. T. & Day, N. E. (1997) Validation of dietary assessment methods in the UK arm of EPIC using weighed records, and 24-hour urinary nitrogen and potassium and serum vitamin C and carotenoids as biomarkers. Int. J. Epidemiol. 26(suppl. 1):S137-S151.
11. Bathalon, G. P., Tucker, K. L., Hays, N. P., Vinken, A. G., Greenberg, A. S., McCrory, M. A. & Roberts, S. B. (2000) Psychological measures of eating behavior and the accuracy of 3 common dietary assessment methods in healthy postmenopausal women. Am. J. Clin. Nutr. 71:739-745.
12. Day, N. E., McKeown, N., Wong, M. Y., Welch, A. & Bingham, S. (2001) Epidemiological assessment of diet: a comparison of a 7-day diary with a food frequency questionnaire using urinary markers of nitrogen, potassium and sodium. Int. J. Epidemiol. 30:309-317.
13. Kaczkowski, C. H., Jones, P.J.H., Feng, J. & Bayley, H. S. (2000) Four-day multimedia diet records underestimate energy needs in middle-aged and elderly women as determined by doubly-labeled water. J. Nutr. 130:802-805.
14. Pryer, J. A., Vrijheid, M., Nichols, R., Kiggins, M. & Elliott, P. (1997) Who are the "low energy reporters" in the dietary and nutritional survey of British adults?. Int. J. Epidemiol. 26:146-154.
15. Lafay, L., Basdevant, A., Charles, M. A., Vray, M., Balkau, B., Borys, J. M., Eschwege, E. & Romon, M. (1997) Determinants and nature of dietary underreporting in a free-living population: the Fleurbaix Laventie Ville Sante (FLVS) Study. Int. J. Obes. Relat. Metab. Disord. 21:567-573.[Medline]
16. Kortzinger, I., Bierwag, A., Mast, M. & Muller, M. J. (1997) Dietary underreporting: validity of dietary measurements of energy intake using a 7-day dietary record and a diet history in non-obese subjects. Ann. Nutr. Metab. 41:37-44.[Medline]
17. Johnson, R. K., Goran, M. I. & Poehlman, E. T. (1994) Correlates of over- and underreporting of energy intake in healthy older men and women. Am. J. Clin. Nutr. 59:1286-1290.
18. Hirvonen, T., Mannisto, S., Roos, E. & Pietinen, P. (1997) Increasing prevalence of underreporting does not necessarily distort dietary surveys. Eur. J. Clin. Nutr. 51:297-301.[Medline]
19. Ballard-Barbash, R., Graubard, I., Krebs-Smith, S. M., Schatzkin, A. & Thompson, F. E. (1996) Contribution of dieting to the inverse association between energy intake and body mass index. Eur. J. Clin. Nutr. 50:98-106.[Medline]
20. Hebert, J. R., Clemow, L., Pbert, L., Ockene, I. S. & Ockene, J. K. (1995) Social desirability bias in dietary self-report may compromise the validity of dietary intake measures [see comments]. Int. J. Epidemiol. 24:389-398.
21. Stallone, D. D., Brunner, E. J., Bingham, S. A. & Marmot, M. G. (1997) Dietary assessment in Whitehall II: the influence of reporting bias on apparent socioeconomic variation in nutrient intakes. Eur. J. Clin. Nutr. 51:815-825.[Medline]
22. Kroke, A., Klipstein-Grobusch, K., Voss, S., Moseneder, J., Thielecke, F., Noack, R. & Boeing, H. (1999) Validation of a self-administered food-frequency questionnaire administered in the European Prospective Investigation into Cancer and Nutrition (EPIC) Study: comparison of energy, protein, and macronutrient intakes estimated with the doubly labeled water, urinary nitrogen, and repeated 24-h dietary recall methods. Am. J. Clin. Nutr. 70:439-447.
23. Johnson, R. K., Soultanakis, R. P. & Matthews, D. E. (1998) Literacy and body fatness are associated with underreporting of energy intake in US low-income women using the multiple-pass 24-hour recall: a doubly labeled water study. J. Am. Diet. Assoc. 98:1136-1140.[Medline]
24. Subar, A. F., Kipnis, V., Troiano, R. P., Midthune, D., Schoeller, D. A., Bingham, S., Sharbaugh, C. O., Trabulsi, J., Runswick, S., Ballard-Barbash, R., Sunshine, J. & Schatzkin, A. (2003) Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: the Observing Protein and Energy Nutrition (OPEN) study. Am. J. Epidemiol. (in press).
25. Heerstrass, D. W., Ocke, M. C., Bueno-de-Mesquita, H. B., Peeters, P. H. & Seidell, J. C. (1998) Underreporting of energy, protein and potassium intake in relation to body mass index. Int. J. Epidemiol. 27:186-193.
26. Briefel, R. R., McDowell, M. A., Alaimo, K., Caughman, C. R., Bischof, A. L., Carroll, M. D. & Johnson, C. L. (1995) Total energy intake of the US population: the third National Health and Nutrition Examination Survey, 1988–1991. Am. J. Clin. Nutr. 62:1072S-1080S.
27. Krebs-Smith, S. M., Graubard, B. I., Kahle, L. L., Subar, A. F., Cleveland, L. E. & Ballard-Barbash, R. (2000) Low energy reporters vs others: a comparison of reported food intakes. Eur. J. Clin. Nutr. 54:281-287.[Medline]
28. Klesges, R. C., Eck, L. H. & Ray, J. W. (1995) Who underreports dietary intake in a dietary recall? Evidence from the Second National Health and Nutrition Examination Survey. J. Consult. Clin. Psychol. 63:438-444.[Medline]
29. Poppitt, S. D., Swann, D., Black, A. E. & Prentice, A. M. (1998) Assessment of selective under-reporting of food intake by both obese and non-obese women in a metabolic facility. Int. J. Obes. Relat. Metab. Disord. 22:303-311.[Medline]
30. Bingham, S. A. (1987) The dietary assessment of individuals: methods accuracy, new techniques and recommendations. Nutr. Abstr. Rev. 57:705-742.
31. Newman, V., Rock, C. L., Flatt, S. W., Wright, F. A. & Pierce, J. P. (1998) Dietary supplement use by women at risk for breast cancer recurrence. The Women’s Healthy Eating and Living Study Group. J. Am. Diet. Assoc. 98:285-292.[Medline]
32. U.S. Department of Health and Human Services (2000) Healthy People 2010 2000 U.S. Government Printing Office Washington, DC. Available at http://www.health.gov/healthypeople/document/(accessed August 30, 2002).
33. Nusser, S. M., Fuller, W. A. & Guenther, P. M. (1996) Estimation of usual dietary intake distributions: adjusting for measurement error and nonnormality in 24-hour food intake data. Lyberg, L. Biemer, P. Collins, M. Deleeuw, E. Dippo, C. Schwartz, N. Trewin, D. eds. Survey Measurement and Process Quality 1996:689-709 Wiley New York, NY. .
34. Dwyer, D., Picciano, M. F. & Raiten, D. J. (2003) Introduction. Future directions for the integrated CSFII-NHANES: What We Eat in America–NHANES. J. Nutr. 133:576S-581S.
35. Biro, G., Hulshof, K. F., Ovesen, L. & Amorim Cruz, J. A., for the EFCOSUM Group (2002) Selection of methodology to assess food intake. Eur. J. Clin. Nutr. 56:S25-S32.
36. Carriquiry, A. L. (2003) Estimation of usual intake distributions of nutrients and foods. J. Nutr. 133:601S-608S.
37. Subar, A. F., Thompson, F. E., Smith, A. F., Jobe, J. B., Ziegler, R. G., Potischman, N., Schatzkin, A., Hartman, A., Swanson, C., Kruse, L., Hayes, R. B., Lewis, D. R. & Harlan, L. (1995) Improving food frequency questionnaires: a qualitative approach using cognitive interviewing. J. Am. Diet. Assoc. 95:781-788.[Medline]
38. National Research Council (1986) Nutrient Adequacy: Assessment Using Food Consumption Surveys 1986 National Academy Press Washington, DC.
39. U.S. Department of Agriculture () Food Guide Pyramid Available at http://www.nal.usda.gov:8001/py/pmap.htm (accessed August 12, 2002).
40. Guenther, P. M. & Perloff, B. P. (1990) Effects of Procedural Differences Between 1977 and 1987 in the Nationwide Food Consumption Survey on Estimates of Food and Nutrient Intakes: Results of the USDA 1988 Bridging Study 1990 U.S. Department of Agriculture, Human Nutrition Information Service Washington, DC. Nationwide Food Consumption Survey 1987–88 NFCS Rep. No. 87-M-1, 48 pp.
41. U.S. Department of Agriculture () Pyramid Servings Database Available at http://www.barc.usda.gov/bhnrc/cnrg/intro.html (accessed August 12, 2002).
This article has been cited by other articles:
|
|
 |
|
  B. P. Marriott, N. Cole, and E. Lee National Estimates of Dietary Fructose Intake Increased from 1977 to 2004 in the United States J. Nutr., June 1, 2009; 139(6): 1228S - 1235S. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. A. Nicklas, C. E. O'Neil, and V. L. Fulgoni III The Role of Dairy in Meeting the Recommendations for Shortfall Nutrients in the American Diet J. Am. Coll. Nutr., February 1, 2009; 28(Supplement_1): 73S - 81S. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. F. Picciano, J. T. Dwyer, K. L. Radimer, D. H. Wilson, K. D. Fisher, P. R. Thomas, E. A. Yetley, A. J. Moshfegh, P. S. Levy, S. J. Nielsen, et al. Dietary Supplement Use Among Infants, Children, and Adolescents in the United States, 1999-2002 Arch Pediatr Adolesc Med, October 1, 2007; 161(10): 978 - 985. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Dwyer, M. F. Picciano, and D. J. Raiten Collection of Food and Dietary Supplement Intake Data: What We Eat in America-NHANES J. Nutr., February 1, 2003; 133(2): 590S - 600. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Dwyer, M. F. Picciano, and D. J. Raiten Food and Dietary Supplement Databases for What We Eat in America-NHANES J. Nutr., February 1, 2003; 133(2): 624S - 634. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
