Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
9 Conclusions and Recommendations In keeping with its mandate, the subcoNmittee's main suggestion is the adoption of a new approach to the inter pretation of dietary intake data that are collected in large surveys such as the Nationwide Food Consumption Sur- vey (NFCS). The subcommittee has also recommended changes in the design of these surveys to facilitate or improve the reliability of interpretations of nutritional adequacy and to improve the data bases, thereby facilitating application of the recommended approach. Some of these recommendations are applicable immediately, even retrospectively, to exist- ing data from surveys, whereas others could be implemented in the next NFCS survey. Still others relate to longer range activities and research programs sponsored by U.S. Department of Agriculture (USDA) and other agencies. Although the subcommittee recognizes that recommendations related to longer term activities will require more time for implementation, it urges that immediate steps be taken toward their adoption in view of their importance. ANALYSIS OF DIETARY ADEQUACY - - The subcommittee concluded that many disadvantages and erroneous interpretations are associated with the appli- cation of fixed cutoff points, i.e., the Recommended Dietary Allowances (RDAs) or percentages of them, as criteria for the interpretation of observed nutrient intake and that the use of fixed cutoff points may lead to erroneous estimates of the prevalence of inadequate intake. ~ The subcommittee recommends that a probability approach to the interpretation of computed nutrient intake be devel- oped and adopted, where feasible. This approach would lead 79
80 to estimates of the prevalence of inadequate intake along individuals in the total population and in population sub- groups. This approach is not necessary for nutrients con- sumed in excess of the RDA by almost all members of the population, for whom adequacy of nutrient intake can be assumed. The subcommittee recognizes that nutritional assessments based on multiple levels of adequacy are important to govern- ment agencies and others for use in planning various nutri- tion programs and in reaching a better understanding of the nutritional needs of the U.S. population. With a multilevel approach, for example, one might look at the intake levels deemed inadequate to prevent clinical manifestations of deficiency, to maintain the functional integrity of meta- bolic systems, or to maintain high tissue concentrations of nutrients if that level is continued for a long period. Each of these levels is different. Although the subcom- mittee did not propose a particular set of criteria for this multilevel approach, it concluded that a multilevel assess- ment of expected dietary adequacy in the U.S. population will provide a more useful picture of nutritional adequacy for planners. · The subcommittee further recommends the development of multiple criteria for nutritional adequacy and estimates of intake required to maintain the various levels of adequacy. To implement the probability approach, it is necessary to bring together information about the distribution of nutri ent requirements among similar persons (e.g., young adult - males, young adult females, children of specified ages). This information about the distribution of nutrient require- ments is not always presented in directly applicable form in the Recommended Dietary Allowances (NRC, 1980) or in other reports' although it might be derived from the studies that are reviewed in the text of such reports. Sensitivity anal- yses have shown that estimates of the average requirement and some idea about the symmetry of the distribution are more important than a precise estimate of the variability. To implement the first recommendation given above, such infonma- tion must be reviewed and judgments must be made about the distributions. · The subcommittee recommends that working descriptions of the distributions of nutrient requirements be developed
81 and improved. Such descriptions should place major emphasis on deriving estimates of, or judgments about, both the central tendency of the distribution and the symmetry of the distribution. Although precise knowledge of the characteristics of the distribution is desirable, it is not essential; estimates of the range can be used. The probability approach proposed in this report pre- cludes the classification of individuals as having ade- quate or inadequate diets. In addition, it would be very difficult, if not impossible, to estimate the proportion of populations with multiple dietary inadequacies. For sub- groups of the population, however, it is possible to esti mate the prevalence of deficiency for separate nutrients. Thus the existence of multiple inadequacies in a population group and possibly in different people can be estimated, even though this is not possible for individuals. The subcommittee recommends that multiple dietary inadequacies be assessed from NFCS data only for popula- tions and subpopulations--not for individuals. The probability approach cannot be applied to the inter- pretation of observed energy intake, even if perfectly mea- sured, because the intake and requirements are highly cor- related in well-fed populations such as that of the United States. · The subcommittee recommends that the probability approach not be applied to the interpretation of observed energy intake. In well-fed populations, energy status as judged by stores must be assessed by anthropometry. Energy intake might be viewed as a potential measure of the implied distribution of physical activity in the total population and in population subgroups. The subcommittee concluded that for same nutrients, it is inappropriate to base inadequate intake estimates on dietary information. Major environmental variables that influence requirement (e.g., the importance of sunlight exposure in the determination of vitamin D requirements) have not yet been assessed for all nutrients. Moreoever, information about requirements is often fragmentary (e.g., for calcium) and data on current food composition may be inadequate (e.g., for folate). For these nutrients, it is nevertheless appropriate to derive descriptive information about observed intake and to make comparisons among subpopu-
82 rations, even though the assessment cannot be made as accu- rately as for the majority of nutrients. · The subcommittee recommends that prevalence estimates of inadequate intakes be attempted for nutrients only when acceptable information about requirements and adequate food composition data are available. When no probability assessment can be made for a nutrient, the subcommittee recommends the descriptive presentation of the mean, variance, and percentile dis- tributions. With the probability approach described in this report, one can estimate the absolute prevalence of inadequate dietary intakes, defined as intakes that will not main- tain appropriate biochemical stores or functions. Many uses of the NFCS do not require such absolute prevalence estimates. In particular, differences between population groups can be determined with other statistically more powerful methods. The subcommittee recommends that the probability approach not be used to generate prevalence estimates for statistical testing of comparisons between or among subpop- ulations for which statistically more powerful methods exist. In the judgment of the subcommittee, all the above recommendations can be applied immediately. The proba- bility approach can be investigated, developed, and imple- mented with several existing survey data bases. STUDY DESIGN The interpretation of the probability approach and its application to the assessment of observed nutrient intake involves several important statistical considerations. These considerations should help to determine the design of data collection, the methods of analysis, and the inter- pretation. A major requirement for the application of the proba- bility approach, or any other approach used to interpret
83 the distribution of observed intakes, is the use of statis- tica1 procedures that take~into account the effect of day-to-day variation in nutrient intake within the indi- vidual. The subc`-~.,~ittee addressed this matter and outlined procedures for implementation. Absolute prerequisites for these procedures are repeated observations on individual daily intake. There must be sufficient numbers of replica- tions for population subgroups as well as for the popula- tion as a whole. The precise neat - ' of replications needed should be considered when designing the survey. Moreover, there appears to be an increase in statistical power if the replicated 1-day intake estimates are obtained by the same method and on independent rather than on adjacent days. It is important also to sample days of the week in the design and to include questions on dietary supplements as well as on food intake. Considering these survey design requirements and the pre- v~ously identified need for statistical services in connec- tion with analysis and interpretation, the subcommittee makes the following recommendations: m e scope of the statistical services that are inte- gral to the design, analysis, and interpretation of the NFCS should be reinforced and expanded. During the planning of future surveys, the following aspects of the design should be considered: tl) the number and distribution of replicate intakes required for statisti- cally reliable adjustments of the distribution of observed intakes to estimate the distribution of usual intakes for the population and subgroups, (2) the wisdom and feasibility of modifying data collection to include a single method for use over all days of observation rather than the two systems presently used to collect dietary information, and (3) the need and feasibility of sampling on independent rather than on consecutive days. Investigations of dietary methodology should con- t~nue to be presented and published in order to obtain additional information on such matters as (1) the associa- tion between intake estimates made on the first days of observation and those made on subsequent days (correlation of intake across days) tuna hence the relative importance of sampling on independent days and (2) sampling, respondent, or interviewer biases that may affect the reliability of
84 population intake estimates and hence methodological approaches that might avoid or minimize any biases that do exist. The characteristics of respondents likely to have biases in reported intake data should be identified, and the direction, and where possible the magnitude, of the biases should be estimated. Priorities for methodological research of this kind for the NFCS should include statis- tical considerations pertinent to the planning of survey design, data analysis, and data interpretation. Because the assessment of inadequate (or excessive) nutrient intake is not the only important use for NFCS data, the subcommittee concluded that data requirements for other applications may differ from those identified in the pres- ent report or implied by the recommendations contained here in. Thus, it will be necessary to consider all intended purposes of the survey during the design phase. · The subcommittee recommends that the USDA review the important uses of the data to ensure that the survey design is adequate for an appropriate balance among these uses. SENSITIVITY ANALYSIS Although the subcommittee did not arrange the recommen- dations in any order of priority, the report does show some of the approaches that might be used to derive a priority ranking. For example, it is possible to test the effect of - incluaing food composition errors in est~mates of the preva- lence of inadequate intake. (See examples presented in this report for testing the sensitivity of the prediction to var- iability of requirement, to errors in estimates of nutrient intake, and to variability of food composition. ) The rel- ative effect of various sources of error on estimations of the prevalence of inadequate or excessive intakes can be identified. To the extent that these effects of error dif- f er between nutrients, they can be used to establish pri- ority ranking for nutrients. In addition to these consid- erations, the relate ve cost of reducing an identified source of error should be considered in judgments about priorities for data improvements. The following recommendations, then, are not presented in any order of priority, although these actions should be implemented to the extent that resources permit.
85 IMPROVEMENT OF THE FOOD COMPOSITION DATA BASE The discussions in Chapters 7 and 8 and in Appendices C and E have illustrated that bias and variation in food com- position data can have a significant impact on prevalence estimates. The subcommittee recommends that the USDA recognize the importance of the food composition data to the accuracy of the such estimates and that it implement the recommendations in this chapter. . The subcommittee recognizes the need to improve the data base on almost all nutrients in some foods and on some nutrients in almost all foods. It also recognizes that such improvements may have limited importance for public policy purposes either because there is little or no public health concern about the nutrient and, hence, limited reason to improve precision of estimates or because the missing or unreliable data refer to foods that make a minor contri- bution to intakes and, hence, whose errors have minimal impact on the estimate of prevalence of inadequate intakes. The subcommittee has not attempted to judge or assess the relative priorities for improving the data base. Rather, it has chosen to recommend areas where improvements could be made. Before these recommendations are assigned relative priorities, sensitivity analyses of the type presented in this report should be conducted to establish whether orient improvement of the data base would have practical slgn~rlcance. The subcommittee was asked to discuss four nutrients for which there are inadequate methodologies. Of these, fola- cin, the carotenoids, and vitamin C may be of present or future public health significance.in the United States. Existing methodologies are also inadequate for Vitamin B12. · Thus the subcommittee recommends that the USDA pro- vide the necessary resources for developing an adequate methodology for assaying.folacin, the carotenoids, vitamin B12, and vitamin C in foods if justified by their public health significance and by sensitivity analysis. Fox the other nutrients listed in the charge to the subcommittee for which there are adequate analytical method- ologies, there are a number of foods for which there either are no data or only imputed data. For some foods, nutrient values are determined by adding the individual ingredients listed in the formulation rather than by direct analysis.
86 · Given the potential impact of imputed data on the prevalence estimates, the subcommittee recommends that the USDA impute values in the food composition data bases so that no zero values are substituted by default for missing values. · The USDA should differentiate between imputed values and values that are calculated or determined by laboratory analyses in food composition data bases and clearly iden- tify calculated and imputed values so that users can readily differentiate among these values when interpreting the prev- alence estimates. The subcommittee believes that priorities for the acqui- sition of new analytical data should be based on the contri- bution of the food to the nutrient intake of the population. Therefore the subcommittee recommends that the USDA perform sensitivity analyses on the imputed values of all foods for which data on key nutrients are missing to deter- mine the probable impact of these foods on the total nutri- ent intake in the populations of interest. When the results of the sensitivity analyses are known, the subcommittee recommends that the USDA perform the fol- lowing functions: . It should collect analytical data for foods contain- ing nutrients shown by sensitivity analyses to have the greatest impact on prevalence estimates. · It should ensure that the foods sampled represent the total foods consumed in the United States. Where the range of nutrient composition for a currently classified food is wide, it should consider subdividing the existing classi- ficat~on into smaller groups with narrower ranges of com- position. Such restructuring may be most appropriate for processed foods as was the case for breakfast cereals in the current data bases. Again, this decision should be preceded by sensitivity analysis. The USDA should also increase the number of assays for selected nutrients to decrease standard errors of the mean, where sensitivity analyses indicate that the preci- sion of the prevalence estimates will be significantly improved .
87 The subcommittee believes that the usefulness of the dietary survey data bases Can be improved. Toward this end, it recommends the following: . The USDA should publish and document the algorithms and procedures used in the computation of the nutrient intakes and their variance, the means and the standard error of the means in food composition data bases, and the prevalence estimates and their variances. · The USDA should publish the concentrations of the individual isomers of the nutrients in food composition data bases, tables, and dietary intake data tapes. The subcommittee believes that the bioavailability of some nutrients may affect prevalence estimates. Thus the subcommittee recommends the following: The USDA should promote and support research that leads to an understanding of the major factors that influ- ence the bioavailability of key nutrients and the develop- ment of algorithms for predicting the bioavailability of these nutrients in meals. When the understanding and algorithms have been devel- oped, the agency should use sensitivity analysis to assess the impact of including bioavailability in the calculation of prevalence estimates. Further actions should be based on the outcome of these sensitivity analyses. The subcommittee believes that public interest in the maintenance of optimal health will continue and that the food consumption surveys will be expected to provide more information on this aspect of diet and public health. Many of the dietary components believed to enhance or retard the development of diseases such as cancer and cardiovascular disease have not traditionally been listed in food composi- tion tables. Thus it has not been possible to assess dietary intake of such compounds in the population through the food consumption surveys. The subcommittee believes that listing these compounds in the food composition tables would significantly enhance the ability of the USDA to respond to anticipated questions on the intake of food com- ponents by the U.S. public and will provide the U.S. popula- tion with much better information on the adequacy and safety of its diet. However, the analysis necessary to acquire
88 this information could also be very costly. Thus the sub- c~ittee recommends the following: · The USDA should investigate the possibility of ex- panding the food composition data bases to include listings for all compounds in foods that are believed to affect human health so that the intake of these components can be assessed in future surveys. · Where potential health benefit is likely, and the quality of the chemical analysis warrants, the USDA should analyze these components, many of which are not nutrients, and include them in the food composition data bases. PREREQUISITES AND LIMITATIONS OF THE PROPOSED APPROACH . The subcommittee is aware that its report poses a number of questions regarding the immediate application of the pro- posed approach to the analysis of nutrient intake data. As indicated by the report of one member (see pp. 104-109), there was not complete agreement about the practicality of the approach. All members of the subcommittee are in agree- ment on the scientific validity of the proposed approach and on the nature of these unresolved issues relating to survey design and data base adequacy; however, opinion varied as to the probability of success in developing the Information needed to apply the approach and the time that will be needed for the required research. The subcom- mittee's judgments are presented under four major areas: acceptable precision of estimates of the prevalence of inadequate intake, estimation of usual food intake, compu- tation of nutrient intake, and definition of nutrient requirements. Acceptable Precision of the Estimates Consideration of the use of estimates of the prevalence of inadequate intakes and of the overall implication of the proposed approach led the subcommittee to discuss the ac- ceptable level of accuracy for prevalence estimates. In Chapter 8, estimates of the confidence intervals for esti- mates of the prevalence of inadequate intake are presented. These take into account all recognized potential sources of bias in the prevalence estimate with the exception of systematic bias in reporting intake or misestimation of
89 mean requirement. In the considered judgment of members of the subcommittee, errors even severalfold greater than those estimated in Chapter 8 would be acceptable for policy analyses. On this all members of the subcommittee agreed. The areas of differing opinion within the subcommittee concerned ideas about the magnitude of unmeasured errors. Estimation of Usual Food Intake The omission of homeless and institutionalized persons from the sampling frame for the NFCS has already been mentioned in Chapter 7. This omission will lead to sys- tematic underrepresentation of the population at highest risk of nutritional deficits, thereby biasing the results of the study to some extent. The subcommittee has recom- mended further research to estimate the magnitude of this bias and to reduce it. This appears to be amenable to sat- isfactory resolution Through applied research and appro- priate survey design, including collaboration with other national surveys. Even though NFCS may continue to under- sample or omit certain segments of the population, it may be possible to gain information about these segments from other surveys, thus complementing NFCS information without necessitating a radical change in the NFCS design. Any residual limitations of the design, and hence the need to draw upon other information, should be made clear to users. The subcommittee recognizes that subj eats may vary in the reliability with which they report food ingested. There may be random under- and overreporting by one sub- ject across days. A subject may consistently under- or overreport, but this may be random between subjects. Or, the entire group may systematically under- or overreport. The nature and extent of these effects in the NFCS are not known. In Chapter 7, the subcommittee considered the potential impact of variations in reporting and concluded that random variation in the reports of one subject would have no Impact on the prevalence estimate derived by the proposed method. Random variation between subjects would have an effect on the estimate. Within the plausible range of magnitude of these effects, however, the bias introduced in the prevalence estimate would be acceptable for policy applications of the results. The subcommittee concluded that a serious potential effect would derive from any appreciable systematic under- or overreporting across the entire population or subpopulation under study. In the
so absence of specific information about the magnitude of this phenomenon, if it exists in the NFCS, opinions about reli- ability of the prevalence estimates varied For those mothers who believed the problem to be large, there were also differences of opinion about the feasibility of improving the collection of data on dietary intake and analytical procedures to minimize the impact on the prev- alence estimates. There was unanimous agreement that the USDA should continue its efforts to assess the magnitude of reporting bias. If bias is found, the USDA should increase its efforts to Improve methods for collecting data on reduction of bias in dietary intake data and analytical methods to correct for known bias in derivations of the prevalence estimate. The subcommittee emphasizes that where other information documents 'he existence of ~ nutritional problem in the population, meaningful comparison of intakes between popu- lation groups can be made without use of the probability approach. Computation of Nutrient Intake As was discussed in Chapter 7, error can also enter into the estimates during the conversion of food intake data into nutrient composition data. This process is dependent on the sampling of foods for chemical analysis, the analytical methods used, the coding categories used to describe foods in the food composition data set, and the computation of nutrient intake. The greatest potential for error in this process is associated with the analytical methods selected and the representativeness with which foods are sampled. The subcommittee believes that the analytical methods for vitamin C and folate and for the vitamin A carotenoids may produce inaccurate data and that sensitivity analysis is required to determine the extent of this effect on the probable outcome of improved assays for these nutrients. The results of these sensitivity analyses should indicate whether it would be worthwhile to develop better analytical methods to correct errors in the existing data. Improve- ments of this kind are feasible. Error may also be introduced into the estimates during the computation of nutrient intake. The subcommittee con- cluded that imputed values may be a source of such error and calls for identification of imputed values in the data
91 sets on nutrient composition. The significance of the impact of imputed values should also be determined through sensitivity analysis. Selective assaying of foods shown by the sensitivity analysis to be responsible for errors in the final data could then be undertaken. Again the sub- committee believes that action can be taken to manage this issue. Members of the subcommittee agreed that variations in the bioavailability of nutrients will produce errors in estimations of the nutrient intake of a person. However, across individuals, the effect may be no more important than the random under- and overreporting across individ- uals, as considered earlier. Opinions varied as to the present importance of the problem and the potential for its solution. Nutrients for which errors are likely to have the greatest impact are iron, zinc, and folate. The subcommittee unanimously agreed that errors in the computer algorithm for computations of any nutrients, to the extent that they exist, would have an impact on the estimates and that these can and should be corrected. Definition of Nutrient Requirements As was discussed in detail in Chapter 3, information on the mean and approximate symmetry of the nutrient requlre- ment is needed in order to apply the probability approach. Analyses performed by the subcommittee and presented in Chapter 5 show that differences in the standard deviation and range of the distribution have a minimal effect on the estimates as long as the distribution is symmetrical. Therefore, imprecision in the description of the variance of requirement was not a major issue. Lack of information on mean nutrient requirement was a major concern for all members of the subcommittee; however, views on the pre- cision of the requisite estimates of mean nutrient require- ment varied. Some members believed that reasonable esti- mates of mean requirement can be constructed immediately for many nutrients across requisite age and sex groups and that these estimates will provide a rational basis for analysis. Others believe that more definitive and scien- tifically validated estimates are required. In the face of these differing opinions on the level of specificity and criteria of proof needed for mean requirement data, there was a lack of agreement on the likelihood of developing this information in the foreseeable future.
92 There was agreement in recommending that multiple levels of requirement, reflecting a series of states of nutriture, be developed and applied. These might relate, for example, to the prevention of scurvy on the one hand and to the main- tenance of tissue stores of ascorbic acid on the other. There was recognition that ascorbic acid may have other effects, perhaps unrelated to its nutrient properties, such as anticarcinogenicity. Requirements to achieve this effect are not known. Other nutrients may require similar cons~d- erations. All members agreed that widespread chronic dis- eases in the United States require more attention than defi- ciency diseases, which are rare in the general population. One member was not certain whether the probability approach could be used for chronic diseases. All other members believe it is important to pay continuing attention to the potential for dietary deficiency in the U.S. population. CONCLUSIONS The subcommittee has stressed the many important uses of NFCS data. In addition to estimating the prevalence of in- adequate intakes, the data permit examination of food use and of dietary patterns. For example, the data are used to establish the patterns of food use associated with nutri- tional deficiency and with undesirably high levels of nutri ent intake (e.g., fat intake). Information about food use, as distinct from nutrient intake, is essential in setting attainable nutritional standards for food assistance programs, for designing meal patterns to meet these pro- grams, and for designing and implementing nutrition educa- tion and other nutrition intervention programs intended to ameliorate nutritional problems detected by whatever means in the U.S. population or groups at particular risk. Infor- mation about food consumption is essential also in the development of food safety regulations. Thus the NFCS data base is important in supporting many national activities mandated by law. There is no clear replacement, now or in the future, for NFCS data for these types of users. In recommending design and interpretational enhance- ments for assessing the prevalence of inadequate intakes, the subcommittee recognizes that this is but one use of the NFCS and not necessarily the most vital use. Therefore, it has repeatedly cautioned that all uses of the data should be considered in making final survey design decisions. -
93 The subcommittee was specifically required by its mandate to evaluate methods to assess observed dietary intake. All except one mother of the subcommittee believed that for sev- eral nutrients, it should be possible to apply the proba- bility approach, as recommended, at this time. All the above limitations to the immediate application of the probability approach apply just as much to any other analytical method that is based on a comparison of intakes to requirements (e.g., use of RDA-based cutoffs, nutrient density). Furthermore, to be applied properly, these other methods would also require even more information than the probability approach and this information may be even more difficult to obtain. For this reason, the probability approach has been identified as the preferred method. After examining the use of fixed cutoff points to analyze dietary adequacy, the subcommittee concluded that attempts to estimate absolute prevalence of inadequate intake using the RDAs or any fixed proportion thereof have no scientific validity and that the results of this type of analysis can- not be meaningfully interpreted. The one exception to this conclusion is analysis for nutrients for which the entire or almost the entire distribution of usual intake is above the RDA. In such cases, it is possible to conclude that inadequate intake is not a major public health problem for that particular nutrient and category of age and sex. All the recommendations of the subcommittee can be implemented immediately or as a part of a phased approach to implementation. The interpretational approach does not mandate major changes in survey design and therefore should not delay the survey itself. However, the interpretation of the data requires development of the average requirement estimates needed to implement the probability approach. This is beyond the mandate of the subcommittee and could be a task for the National Research Council's Committee on Dietary Allowances. Some preparatory work can be initiated immediately by the USDA. Sensitivity analyses to determine which error sources have a meaningful impact on estimates of prevalence of inadequate intakes, as exemplified within the present report, are recommended for continuation through the next few years. These would serve to resolve disagreements of opinion about the Importance of different potential sources of error. More focused sensitivity analyses for nutrients with a potential for having an important impact
94 will guide the longer term research efforts needed for a full application of the approach. Appropriately analyzed, dietary data can be used as a basis ,-or judging the presumed adequacy of intake. The subcommittee has proposed an appropriate approach to such interpretation. This application is distinct from an assessment of the state of health consequent to inadequate intake. As discussed in a previous report (NRC, 1984), biochemical and clinical observations included in other national surveys are more appropriate for the assessment of nutritional status per se. Nevertheless , dietary data, such as those collected in NFCS, are required in inferring a dietary causation of observed health effects and in considering dietary actions that might ameliorate them.
