View charts showing means and percentile distribution of energy (Kcal) intake in the U.S. population, as reported in the National Health and Nutrition Examination Survey (NHANES III), and the Continuing Survey of Food Intakes by Individuals (CSFII, 1994-1996).
Excerpted from Wakimoto P, Block G. Dietary intake, dietary patterns, and changes with age: an epidemiological perspective. J Gerontol 2001; 56 Spec No 2(2):65-80.
A discussion of validation considers the issues related to how researchers get to "the truth" about dietary intake for individuals. For additional information about the development and validation of Block questionnaires and screeners, go to Validation.
To access charts showing means and percentile distribution of energy (kcal) intake for adults in the U.S. population, for Whites, Blacks, and Mexican-Americans, and for men and women, go to Caloric Intake Charts.
The method selected for dietary assessment in research depends on the purpose of the assessment, the available funding, and the burden it is appropriate to impose on the respondents and the study. Commonly used methods include:
This is the method of choice if the purpose of the study is simply to describe the average intake of a group. That is, it provides a mathematically accurate group mean. However, it does not provide a good estimate of an individual's usual intake for any individual member of the group. It can also be used to examine whether two groups have similar mean intakes--for example, the baseline levels of an intervention and a control group.
For epidemiologic research, the purpose is to relate an individual's usual intake to that individual's health status or outcome. For this type of research, a single 24-hour recall is not appropriate. For an individual food intake varies considerably from day to day, and intake on any single day is not a good estimate of his/her usual intake.
If dietary intake over a number of days is collected, the data begins to approximate an individual's usual intake. Thus, if enough days are collected on each individual, this could be an appropriate measure to use for research in which the individual's usual intake is the unit of interest.
The question then arises as to how many days are needed. There has been extensive research on this topic, as discussed in these and other works:
Generally, at least three days of diet data are required for the most stable nutrient, percent of calories from fat. Other macronutrients require more, and micronutrients such as vitamin C and vitamin A require many more days. Except for very large well-funded studies, anything larger than three days is not generally feasible. In addition, the expense can be substantial. Researchers might wish to contact the University of Minnesota, Nutrition Coordinating Center ( http://www.ncc.umn.edu), or the Cincinnati Center for Nutrition Research (Suzanne.summer@cchmc.org) for information on costs for administering and coding 24-hour recalls. Multiple days of diet records are considered by some to be a gold standard for collection of individual dietary data. Dr. Gladys Block and others, however, believe that good, probed 24-hour recalls are superior. Diet records, completed by the respondent, are dependent on the literacy and commitment of the population. It is also the case that respondents may choose not to consume certain foods during the recording period, either because they are complex to record or because they are embarrassing to admit. Most important, it has been repeatedly seen that many respondents may record their food intake well at the beginning of the day; however, as the day progresses they tend to forget or get too busy. The result is that for many respondents, what was supposed to be a moment-by-moment recording becomes a non-probed recall at the end of the day.
Most nutrition research is interested in the usual intake of the individual. Consequently, single 24-hour recalls are not appropriate. Multiple recalls are expensive and burdensome. Furthermore, all recall or recording methods that obtain information on current diet are completely inappropriate for case-control studies, in which it is necessary to ask about intake sometime in the past, at least before symptoms of the disease were noticed.
Consequently, many researchers have turned to food frequency questionnaires as the most appropriate method. Food frequency questionnaires (FFQs) ask about usual frequency of consumption of a list of foods. The time frame may vary, but is often the past year or the past six months, to "smooth out" issues of seasonality. The food list can be chosen in a variety of ways (see our method, under NHANES analyses). Some FFQs ask about the individual's portion size for each food, whereas others assign a single portion amount, or ask the respondent to divide their own portion by a standard amount in order to estimate their frequency of consumption of the standard amount. FFQs generally take between 20 minutes (for the Block Brief 2000) to as much as 1-1/2 hours.
Do food frequency questionnaires always overestimate? No! Some FFQs overestimate, and some underestimate, and some are close to a good point estimate. It is important to realize that there is no such thing as "the food frequency questionnaire." They can and do differ in critical areas: number of food items included in the food list; method by which food items are selected; whether portion size is asked and the manner in which it is assessed; and the exact nature of the quantification of the instrument.
The full-length Block questionnaires (e.g. Block98, Block2005) have been shown to come quite close to the point estimates produced by multiple days of diet recalls and records. Very long food lists will tend to overestimate, while short lists will tend to underestimate. The Block Brief 2000 intentionally has only two-thirds as many foods as the full-length Block FFQ. It is obvious, then, that total calories and macronutrients will be underestimated by the Brief2000. However, for its purpose of ranking people along the distribution of intake, it is still effective.
Are food frequency questionnaires usable for examining an individual's diet? There is controversy about this question. Part of the controversy is due to the fact that food frequency questionnaires differ. As noted above, some are longer, some are shorter, and their quantifications differ. The Block full-length FFQs produce estimates that are reasonable estimates of an individual's intake, for most individuals. As with any FFQ, some people will produce unrealistically low estimates, and others will produce unrealistically high estimates.
For some research or public health purposes a full-length questionnaire is not practical. Therefore, screening tools have been developed, usually to assess just one or two nutrients or food groups. They generally do not include portion size, but just ask about frequency.
If a questionnaire is to be subjected to a validation study, certain criteria apply. It is generally not appropriate to simply compare one questionnaire with a longer version of the questionnaire. The correlations will be high, because the two instruments are based on similar principles; but we will not know which if either accurately represents "the truth." A person could exaggerate his carrot intake both times, for example. In general, the "gold standard" against which an instrument is compared should be quite different in type, in the kinds of answers that are required. Most commonly, a food frequency questionnaire is compared against multiple days of diet recalls or records. Several days of recalls or records should ideally be used since, as noted above, a single day will not provide a good estimate of the individual's usual intake, so will not provide a good gold standard against which to compare an FFQ.
As noted above, the purpose of an FFQ is to estimate an individual's usual intake. Correlations between an FFQ and a single day of dietary data will usually be very modest, as a result, due to the inappropriateness of the so-called gold standard data. At least two days of dietary data should be used, in which case it is possible to carry out adjustments to take into account the day to day variability of estimates from few-day data. For any given questionnaire being validated, the correlations will be increasingly higher as more days of reference data are used to compare against. Validation studies that use 4 days of reference data will produce higher correlations than validation studies that use 3 or 2 days, for exactly the same FFQ. Thus, in comparing validation studies, the number of days in the reference data should always be considered.