Click for next page ( 20


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



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

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

OCR for page 19
Some Study Design Characteristics DIETARY METHODOLOGY Selection among available methods for assessing food and/or nutrient intake of households or individuals depends on available resources (funds, time, and personnel), the objectives of the study, and the target population (Marr, 1971~. The effectiveness of any survey method in answering questions about food or nutrient intake of a population is contingent upon a statistically adequate sampling frame to assure that appropriate respondents are used and that oversampling of targeted groups to meet specific needs is permitted. In general, information on food intake is obtained by (1) inference, (2) observation, and (3) verbal or written reports. Each has some inherent strengths and limitations, so a combination of methods may be needed. Inferred data are derived primarily from aggregate data provided by commodity reports, commercial surveys of movement of food products in and out of warehouses or markets, and national food balance reports. These data provide information on trends in food consumption for a population group. They do not provide information on average intakes or ranges of intakes of individual consumers. Aggregate data can serve as a surrogate for individual data if the above limitations are understood. However, aggregate data are most useful in validating observations generalized to the total population from individual data. Direct personal observation is generally so costly of money, time, and personnel that it is precluded in large-scale surveys and is used primarily to validate reported data. The observations may be either obtrusive (such as participant observation, which may in short-term studies result in modified 19

OCR for page 19
20 ASSESSING CHANGING FOOD CONSUMPTION PATTERNS behavior) or unobtrusive (such as monitoring store purchases and correlating them with specific individuals, which may constitute an invasion of pri- vacy). Both modified behavior and invasion of privacy are serious obstacles to the use of methods involving observation of food intake. AS a result of the limitations of inferred and observed data, reporting is generally regarded as the most feasible and cost-effective way of learning about food intake of individuals and groups. Reported data are obtained by oral interview or are recorded by the subject. Oral interviews involve retro- spective accounts with the interviewer available in person or by telephone. Food frequency information, a dietary history, and/or a 24-hour recall of food intake may be recorded by or for the subject in a written, telephoned, or tape-recorded format. Food intake records may be maintained in this manner for from one to usually no more than 7 days. Usefulness of a reporting method depends on validity and reliability of the information reported, on the cost and speed of data collection and analysis, and on the appropriateness for both long- and short-range goals of the survey. Conclusions are only as good as the quality of the core information on which they are based, so an ongoing effort to assess the validity of the reported data is essential. Although there is considerable information on the reliability of various techniques for obtaining written and oral reports, there is very little information on validity of the data obtained. Limited studies done on small subsamples of the population indicate a tendency to underre- port at upper levels of intake and to overreport at lower levels. This phenomenon, known as the "flat slope syndrome," casts doubt on the interpretation of data as regards extremes of intake. The extent to which actual and reported intakes vary appears to differ with the nutrient studied; therefore, the "flat slope syndrome" may need to be evaluated across the range of nutrients (Gersovitz et al., 19781. There is an urgent need for methodological studies for assessment of the validity of currently used dietary survey techniques in a range of cir- cumstances and for identification and validation of alternative and innova- tive methods for obtaining food intake information (Garn et al., 1978~. While this report suggests four replications, questions remain regarding the minimum number of days or other units of time for which reported observa- tions must be made in order to assess usual dietary intake and produce an accepted representation of actual intake. It has been reported in at least one study that validity of record keeping by adults decreases after 4 consecutive days (Gersovitz et al., 1978~. There does not seem to be a similar problem when records are kept for several days intermittently (Beaton et al., 1979~. Records of a single day's intake of individuals representing weekdays and weekend days over a long enough interval to detect cyclic changes have been found to give an acceptable estimate of usual food consumption

OCR for page 19
Some Study Design Characteristics 21 (Houser and Bebb, 19811. Probing to clarify records and enhance the accu- racy of recall has resulted in substantial increases in the completeness of records (Campbell and Dodds, 1967~. Analyses of food intake records using James-Stein indicators show that nutrient intakes regress toward the mean and that averages of 5 (and possibly fewer) days may give values that more closely approximate usual intakes than do individual day intakes (Samonds et al., 1978~. Other methodological questions for which there are only limited answers and which require further research are how to ensure im- mediate reporting, how to standardize estimates of serving sizes and de- scriptions of food items, the extent to which the homemaker can report for other family members, and the extent and nature of interviewer differences. Written records are useful with literate, motivated subjects but not with those who are illiterate or poorly motivated or those with impaired vision or lack of neuromuscular coordination. Oral records are useful with these groups, but they often necessitate either a costly personal interview or the use of a recording device that may be technically baffling or too impersonal to encourage continued participation. Interviews that provide an opportunity for probing and permit clarification of food identification have been found to increase reported intakes by as much as 25 percent and to prolong participa- tion, but they may encourage subjects to strive to respond in a perceived approved fashion. A limited study of the validity of food records telephoned each day to a nutritionist or a telephone answering service showed them to be as valid as a 24-hour recall or a 7-day food record (Raker, 1979~. Reports given to the nutritionist who could probe for some specific information were judged to be more valid than those from a recording device. The phone survey technique has the advantages of assuring that records are kept each day and of being equally applicable for literate and nonliterate subjects. However, telephone recordings have limitations similar to those of written records and limit the population surveyed to respondents who have telephones. Currently available methods for determining food intake and the inherent advantages and disadvantages of each are shown in Table 2. Dietary methodologies must be refined to identify feasible methods of minimizing their present limitations and disadvantages. ESTIMATION OF USUAL INTAKES The preceding discussion has defined many of the limitations of current methods for measuring food consumption patterns of individuals and popu- lation groups. Most methods in current use collect data relating to finite periods of time, i.e., 1, 2, 3, or 7 days. For the proposed system it will be necessary to collect data that describe the distribution of usual food intakes

OCR for page 19
22 ~4 - o o ;^ a' so em ._ | c Ct , to Cat 04 Ct a: o ._ E I o Us C) ;` 1 ~ | ~ | ~ ~1 ~ ~ 1 ~1 m ail 1 ~ 1 , - I V I , ~ ~ r I ~ ' ~ a, ~ ~ ~ ~ a, ~ a == ~ ED O ~ ~ aim e >a.= ~ ~ ~ O ~ _ an ~ ~ ~ O ~ .= ~ O I) ~ ~ to c) , ~ o. E - - o C ~ C c ~ ~ ~ Cat - C~ ~ I, ~ ~ C) ~ {; , 0 Y W ~ ' ~ ~ is, U o o ~ a: ~ .= ;, ;5 C D ~ ~ ~. I E ~ ~ ~ m~ ~ ~ _ ~ ~

OCR for page 19
23 ~ i, ~ ~ ~ c ~~ Y ~ 3 ~ ~ E y . _ E ~ ^2 ~ C _ ~ ' ' ' 0 O ~ O ~ o y e y _ D C ~ _ ~ e _ C .0 ~ ~ O ~ U U. _ ~ E o C; - ;~ 6 X C >~ 2tD C o _ ~ t_ o ~ it. ~ o ~ 1 In Ct _ ~ _ Cat Cut ~ ,, ~ ~ Cat o C,) o o ID ~ ~ ~ o C C ~ 3 ~ _ m ~

OCR for page 19
24 ASSESSING CHANGING FOOD CONSUMPTION PATTERNS of the population group over an extended period of time. Statistically, this measure may be described as an estimate of the inter individual (between individual) variation. The relationship between observed variance and inter- and intraindividual variances is described by the following equation: o2 = ob2 + w21n, where o2 = observed variance, 0b2 = intraindividual (between individual) variance, oW2 = intraindividual (within individual) variance, and n = number of repeated observations for each individual. The last term (ow21n) may be described as day-by-day variation in the intake of individuals. The magnitude of this component of variance may be estimated if the study design includes repeated measurements of 1-day intake for all, or an approximate representative sample of the individuals. Preferably these replications of 1-day data should be independent of one another (e.g., 1-day observations separated in time), but, with appropriate statistical treatment, related obser- vations (e.g., the individual days of a 3-day record) may be used in estimating the intraindividual variance. The design should include appropriate sampling to measure the influence of seasons, holidays, and weekends on the patterns of food intake. With knowledge of total or observed variance and of intraindividual var- iance, an estimate of interindividual variance (the measure of variation in usual intake) can be derived from the above equation. Repeated observa- tions on the same individuals is the preferred method for measuring usual pattern of intake and its variation. The reliability of the estimate of oW2 is a function of the total number of repeated 1-day observations. From the statistical point of view, collecting data from 100 people twice or 50 people three times (each pattern has 100 repeated days) leads to similar confidence limits in the estimate of intrain- dividual variance as long as the subjects observed are representative of the population. The decision on the desirable number of replications per subject should involve consideration of the logistics of repetitive examinations separated in time (i.e., as independent observations). The decision on the total number of additional measurements (total replications) must be based upon a consideration of needed confidence. Since interindividual variance is calculated from intraindividual variation and observed total variance, it follows that if the total sample is large and hence the total variance has been reliably measured, the confidence of the estimate of interindividual variation is inversely related to the ratio of intra/interindividual variation. That is, as the magnitude of intraindividual variation increases in relationship to interindividual variation, the reliability of the estimate of intraindividual variation, and hence the number of repli-

OCR for page 19
Some Study Design Characteristics 25 cate 1-day intake estimates required for the population must increase if a reliable estimate of interindividual variation is to be obtained. The statistical problem of confidence limits is considered in survey design. Beaton and co-workers (1979) have provided estimates of the partitioning of variance for several nutrient sectors. It is apparent that this varies with the nutrient and with the food item under study. Thus, design requirements are a function of questions to be asked of the data. It is extremely important that all sources of consumed food, including food supplements, alcoholic and nonalcoholic beverages, water, and medi- cations, be incorporated into the data. The several methods for collecting individual food consumption data should be evaluated. In practice, no one method is likely to be perfect; therefore, it is necessary to select and adapt the best method to meet the needs of the particular study to be undertaken. Combinations of 24-hour recall with two or more 3-day diet records, food frequency questionnaires, telephone interviews, and/or extended diary rec- ords could be considered. It should be remembered, however, that for the proposed monitoring system, repeated observations that permit measurement of the usual food intake and the extremes or variation by and between individuals are essential components. Internal quality control checks must be an integral part of this or any system. Validity checks are extremely important and a concerted effort should be made to develop and implement procedures to determine the validity of various methods of assessing dietary intake, including combina- tions of methods. Research is required to develop satisfactory systems for determining the validity of measurements obtained. ASSESSMENT OF NUTRITIONAL STATUS The conventional method of determining nutritional status of individuals and population groups relies primarily on biochemical assessment. The proposed food consumption monitoring system can provide only a probabil- ity estimate of nutritional status of a population group (by comparison of the usual nutrient intake with appropriate nutrient requirement figures). It is not possible from consumption data alone to identify the specific individuals in the population who are in a particular nutritional state. Both prevalence and severity of nutritional inadequacy and excess are of concern in population assessment. In biochemical assessment, different levels of nutrients (in cells, tissues, and fluids) are associated with different probabilities of impaired function. Frequently, cutoff points are selected for categorization of deficiency or excess. As the cutoff point for adequacy is raised or lowered, the prevalence of deficiency appears to increase or de-

OCR for page 19
26 ASSESSING CHANGING FOOD CONSUMPTION PATTERNS crease. However, it is the interpretation that has changed, not the prevalence or severity of functional impairment. It is possible to establish a series of cutoff points related to the deficiency (or excess) of a nutrient, and it is possible to develop a series of prediction curves expressing the expected prevalence of varying degrees of deficiency (or excess) in the population. This can be done with biochemical data and with combination dietary and nutrient requirement data. Using the proposed system of estimating usual nutrient intake from usual food intake and relat- ing this to usual nutrient requirement, information obtained from dietary data ought to be similar in accuracy to information obtained from biochemi- cal data. When the estimates of nutritional adequacy indicate the probability that a particular population segment has a significant prevalence of inadequ- acy (or excess) for one or more nutrients, it may be desirable to study that population group in greater detail. Such studies must include biochemical assessment if the intent is to identify specific individuals who are in poor nutritional state. RELATIONSHIP OF FOOD CONSUMPTION PATTERNS TO HEALTH STATUS INDICATORS The possible relationship between food consumption patterns and health status provides the rationale for FDA interest in monitoring food consump- tion. Yet the relationships are complex and not easily documented. This section will attempt to provide some perspective on the types of indices and methods used in the proposed system, which will give a picture of food- related differences in health status in the population. Three types of relationships between flood consumption and health status are of concern. Firstly, health status is affected by food consumption through the intervening variable, nutritional state. For these types of health outcomes, dietary data have predictive value in population terms even though they are not direct measures of nutritional status. Selection of specific measures of nutritional status depends on the level of sensitivity required, e.g., degree of iron saturation. A second type of relationship is one in which consumption of specific foods, substances, or combinations of these is linked to health status but not through the intervening variable, nutritional status. Most toxicological problems that are due to chance contamination of a specific production lot of a food are of this type. For this kind of linkage to be monitored effectively, it is essential that the food consumption data base be adaptable to very fine disaggregation by specific commodity product type, brand, or other factors. For some of these relationships, special ad hoc studies will have to be undertaken as the information required is likely to be highly specific to certain foods or population segments.

OCR for page 19
Some Study Design Characteristics 27 Food consumption patterns and/or nutritional status can also be con- tributing or facilitating factors but not the sole etiologic factor in the pathogenesis of diseases such as coronary heart disease, hypertension, or other chronic degenerative diseases. This third kind of relationship is characteristic of several of the leading causes of morbidity and mortality in the United States. Generally, the contributory role of diet is not well quan- tified. However, since diet may be the only factor that can be easily manip- ulated, it is important to know if differences in food consumption are as- sociated with differing incidences of these diseases. In making this determi- nation, it is important to control the other variables (such as heredity, occu- pation, etc.) known to affect the disease. Irrespective of which of the three types of relationship is postulated, the disease or condition of concern must meet certain basic criteria (Institute of Medicine, 1973), which include the following: 1. have significant functional impact on those affected; 2. be relatively well defined and easy to diagnose in clinical and nonclinical settings; 3. have a prevalence rate that is high enough to permit the collection of adequate data from a limited population sample; 4. have a natural history that varies with utilization or consumption of food and/or nutrients; and 5. have potential for documentation of influences of nondietary vari ablest Identification of a clear relationship between food consumption and health status indicators will suggest means of dietary intervention for prevention, and often treatment. A systematic approach for relating a health status problem to food con- sumption data is essential. A logical progression of steps for such linkage is given below, recognizing that modification will be appropriate for specific conditions. There are six steps, which may be concurrent: 1. conducting retrospective studies to establish increased relative risk of disease (or decrement in a health status indicator) associated with differ- ences in food consumption; 2. monitoring populations which have a particularly high or low preva- lence or incidence of the disease indicator; 3. verifying the continued association in identified populations; 4. identifying populations with food consumption patterns indicating they may be at risk and initiating prospective studies in these populations in order to confirm the relationship and determine proportion of risk attribut- able to food consumption;

OCR for page 19
28 ASSESSING CHANGING FOOD CONSUMPTION PATTERNS 5. monitoring food consumption patterns of populations to detect those populations at risk; and 6. developing and implementing intervention strategies. For some conditions, in particular acute and chronic toxicities, steps 2 to 5 may not be needed before intervention is undertaken. In cases where the association between health status and food consumption is firmly estab- lished, steps 1 to 4 may not be necessary. However, the general model provides a framework for an orderly approach to development of cost- effective intervention strategies. Retrospective studies to establish an association of health status with food consumption are dependent on identifying probable associations through laboratory, clinical, or epidemiological studies. Appropriate study design must consider variables other than food consumption that may be related to the etiology or expression of the health state. Dietary methodologies de- scribed earlier will be used to determine current food consumption and to test its relationship to acute toxicities, perinatal events, infant morbidity or mortality, and the like. Different methodology is required to determine long-term food consumption patterns and their effect upon chronic tox- icities, chronic disease, growth, etc. A projected need to use data generated by the proposed consumption monitoring system for this determination would have significant design and sampling implications. An important aspect of the proposed system for relating food consumption to health status involves the use of existing health data on identified popula- tion segments to determine unusual health patterns. The usual pattern of food intake of these populations can then be determined in a special study or by oversampling the identified segment in the ongoing survey. Health and food intake data are preferably obtained from the same indi- viduals. If this is not possible, data should be collected on individuals of the same sample cell characteristics to provide as close a relationship as possible between usual food intake information and health status. When indications of a relationship between patterns of usual food intake and health status in a population group are observed, confirmation may be necessary through the use of special studies to obtain significant measures of nutritional and health status. Biochemical and anthropometric data as well as extensive dietary and medical histories of the individuals in the population under study may be required.