Describing Dietary Intake

NOTE: Links for "Row" in the text below will take you to the relevant portion of the Summary Table.

Data Capture Recommendations

24-hour Dietary Recalls

Choose the 24HR for studies in which your objective is to describe the dietary intake of a group (see the 24-hour Dietary Recall Profile).

  • If your objective is to estimate [glossary term:] mean usual intake of a group (Row 1), or the difference in mean usual intake among two or more groups (Row 2), a single administration of the 24HR is sufficient. It is, however, important to consider the time frame to which the 24HR data apply. For example, if you are interested in intakes over a full week, distribute the administration of the 24HR across the days of the week so that on each day, including weekdays and weekends, you collect the 24HR from some of the individuals in your sample (Learn More about Day-of-Week Effects). Similarly, if you are interested in intake over a year, distribute the adinistration of the 24HR across the seasons of the year (Learn More about Season Effects). Although collecting 24HRs for two weekdays and one weekend day from each participant is a common strategy, this is not necessary. Rather, a single 24HR spread across the sample to capture all the days of the week and seasons of the year as per the time period of interest for the study is sufficient.

    • When estimating means of usual intake, consider collecting [glossary term:] recovery biomarker data for at least a subsample to aid in adjusting for [glossary term:] measurement error, including random error and [glossary term:] bias. However, recovery biomarkers are currently known for only a few dietary components (Learn More about Biomarkers).

  • If your objective is to describe the distribution of usual intake among a group (Row 3) (for example, for the purpose of estimating the proportion of individuals above or below some threshold or percentile of usual intake), repeat your administration of the 24HR in at least a subsample to enable adjustment for [glossary term:] within-person random error (see Key Concepts about Measurement Error). As noted above, although collecting 24HRs for two weekdays and one weekend day from each participant is a common strategy, this is not necessary. However, repeat administrations to a given individual should not be scheduled for consecutive days or on the same day of the week. Also, consider the time frame over which the 24HRs (both the initial 24HRs and the repeat assessments) are completed in relation to the period of interest (e.g., week, year). Ensure that 24HRs in your sample are distributed across all days of the week and in all seasons of the year. Further, the subsample that completes a repeat 24HR should be representative of your full sample.

    • In studies in which only one recall is available per person, it may be possible to obtain an estimate of the ratio of [glossary term:] within-person variation to the sum of within- and [glossary term:] between-person variation from an external study on a comparable sample. This information can then be used to generate estimates of within- and between-person variation that can be applied to estimate distributions of usual intake.

  • If your objective is to estimate distributions of usual intake of [glossary term:] episodically-consumed dietary components (dietary components that are consumed on a given day by less than 5% to 10% of the population of interest, such as whole grains, dark green vegetables, n-3 fatty acids), you may need more than two administrations of 24HRs to capture at least two consumption days for the component among a reasonable proportion of the population. The number of administrations that is necessary depends on how episodically a given dietary component is consumed. Components that are consumed more episodically (i.e., less frequently or by fewer people) may require a larger subsample with repeat 24HRs. Further research is needed to guide specific recommendations on the number of 24HRs needed for different dietary components in relation to frequency of consumption, which may vary across populations with different consumption patterns. That said, the prudent approach is to collect two 24HRs from the entire sample.
  • The existing evidence suggests that it is not necessary to combine an FFQ with 24HRs to estimate means and distributions of usual intake. However, frequency data may be of use in estimating the tails of the distribution, particularly for episodically-consumed dietary components. Again, further research is needed to inform such approaches.
  • For studies in which your objective is to estimate the mean or the distribution of acute intake, or intake for a single day (for example, the proportion of a group with intake within recommendations for alcohol consumption on a given day) (Row 4), a single administration of a 24HR is sufficient.
  • If you are interested in estimating [glossary term:] total nutrient intake, either usual or acute, beyond those from foods and beverages, you will need to also collect data on intake of [glossary term:] dietary supplements (Learn More about Dietary Supplements and Estimating Total Nutrient Intake).

Food Frequency Questionnaires & Screeners

FFQs and screeners are generally not recommended for studies aimed at describing the dietary intake of a group because the estimates are likely to be biased. However, if efforts are made to reduce this bias, it is possible to use FFQs and screeners for this purpose, although the resulting estimates are generally likely to be more biased than those derived from 24HRs (see Food Frequency Questionnaire Profile and Screener Profile).

  • It may be acceptable to use an FFQ for estimating mean usual intake if the bias is adjusted for with the help of more accurate measures. This could involve collecting data from a subsample (this is called an [glossary term:] internal calibration sub-study) using a less-biased measure, such as a 24HR and/or a recovery biomarker. Alternatively, data from an external source (called an [glossary term:] external calibration study) can be used.
  • A screener also may be acceptable if the focus of your study is limited to one or a few specific dietary components that are concentrated in relatively few food sources. As with an FFQ, the data from the screener should be adjusted for bias using a less-biased source of data.
  • In research using FFQs or screeners, the specific instrument you use should be carefully considered to ensure that it is tailored to the population and captures the majority of food and beverage sources within the food supply for the dietary component of interest (Row 1 and Row 2).
  • FFQs and screeners are not recommended for estimating distributions of usual intake (Row 3).
  • FFQs and screeners are not designed for capturing intake for a single day (Row 4) and thus, should not be used to estimate acute intake.
  • As with 24HRs, if you are interested in estimating total nutrient intake beyond those from foods and beverages, you will need to also collect data on intake of dietary supplements.

Data Analysis Recommendations

24-hour Dietary Recalls

Data collected using 24HRs are affected by within-person [glossary term:] random error, primarily driven by [glossary term:] day-to-day variation in intake. The implications of this error and the strategies needed to address it depend on the objective of the study.

  • If you are interested in estimating [glossary term:] mean usual intake (rather than the distribution of usual intake), you can use data from a single administration of a 24HR and you do not need to adjust for within-person random error. That is because the random error has a mean of zero and so would not be reflected in the average of all 24HRs. As noted above, the administration of the 24HR across the sample should be spread across the time period of interest, e.g., a week or year. In addition, you may include [glossary term:] covariates to account for [glossary term:] nuisance effects. Covariates also can be used to account for other factors related to the study design, such as interview mode (e.g., telephone compared to in-person 24HR).
  • If you are interested in estimating the distribution of usual intake, you could collect a large number of 24HRs from each participant and use the average intake over the repeats in place of estimated usual intake. However, this option is typically prohibitively expensive and burdensome given the number of 24HRs needed per participant. Fortunately, statistical modeling makes it possible to use a limited number of repeat 24HRs to separate the within-person variation (the major source of which is day-to-day variation in the case of 24HRs) from the [glossary term:] between-person variation and then remove the within-person variation to arrive at an estimate of usual intake for the group (Learn More about Statistical Modeling).

    • As noted above, two 24HRs are needed on at least a subsample to estimate distributions of usual intake. The number of administrations and the size of the subsample will depend on how episodically the dietary component is consumed. In studies in which only one recall is available per person, it may be possible to obtain an estimate of the ratio of within-person variation to the sum of within- and between-person variation from an external study on a comparable sample. This information can then be used to generate estimates of within- and between-person variation in the single 24HR study that can be applied to estimate distributions of usual intake.

  • A number of statistical methods have been proposed to estimate usual intake distributions using data from repeat 24HRs, including the National Cancer Institute (NCI) method, which was developed by researchers at the NCI and elsewhere. Programs (e.g., SAS macros and sample applications) are available to implement the NCI method in the following situations:
    • Univariate analysis: estimating the distribution of usual intake of a single dietary component (for example, the distribution of usual intake of added sugars).
    • Bivariate analysis: estimating the joint distribution of usual intake of two dietary components (for example, the distribution of the usual percentage of energy intake from fat).
    • Multivariate analysis: estimating the joint distribution of usual intake of more than two dietary components (for example, the distribution of usual intake on a multi-component diet quality index, such as the Healthy Eating Index (HEI).
  • Using the NCI method, it is possible to include data on frequency of consumption derived from an FFQ as covariates in statistical models using 24HR as the primary source of intake data. This may be useful for estimating the tails of the distribution, particularly for episodically-consumed dietary components. Additional studies are needed to inform the utility of an FFQ for this purpose. The NCI method allows the option to include covariates to account for nuisance effects (e.g., day of week for which the 24HR was collected).
  • Other methods also have been developed to estimate usual intake distributions using data from repeat 24HRs (see References and Resources below).
  • Statistical methods exist to estimate the [glossary term:] measurement error structure of dietary instruments using recovery biomarker data on at least a subsample. Once that structure is known, statistical techniques can be used to help adjust measurement error, including random and systematic error, in the estimation of means of usual intake.
  • Methods also are available for incorporating data on intake from [glossary term:] dietary supplements to estimate total usual nutrient intake.

Food Frequency Questionnaires & Screeners

  • If you are interested in estimating the mean usual intake of a group or the difference between two groups in mean usual intake using FFQ or screener data, draw upon less-biased intake data to adjust estimates for bias.
  • To reduce bias in estimates of mean intake derived from FFQs, use dietary data captured with 24HRs and collected from a subsample (called an internal calibration sub-study). Alternatively, 24HR data may be available from an external sample (called an external calibration study) and may be used to adjust for bias in the frequency data to some extent (see Key Concepts about Validation). The most common method used for making this adjustment is regression calibration (Learn More about Regression Calibration). It also is possible to use recovery biomarker data to adjust for bias for some nutrients in cases in which such data are available. Energy adjustment also is recommended to reduce bias in frequency data (Learn More about Energy Adjustment).
  • To reduce bias in estimates of mean intake derived from screeners, use dietary data captured with 24HRs and collected from a subsample in an internal calibration sub-study, or use externally-derived scoring algorithms. The NCI has developed scoring algorithms for a range of [glossary term:] screeners (see the Screeners Profile). These scoring algorithms enable conversion of responses on the screener to estimates of mean intake and can be used to reduce bias in estimates of mean intake (Learn More about Scoring Algorithms for Screeners). In contrast to FFQs, because screeners do not capture the total diet, energy adjustment to further reduce bias is not possible.
  • FFQs often query usual intake of a finite list of commonly consumed dietary supplements. These data are used in conjunction with estimates from foods and beverages collected on an FFQ to estimate total usual nutrient intake.

References and Resources

The following references and resources provide additional information on the topics discussed in this section.

References

Bailey RL, Dodd KW, Goldman JA, Gahche JJ, Dwyer JT, Moshfegh AJ, Sempos CT, Picciano MF. Estimation of total usual calcium and vitamin D intakes in the United States. J Nutr 2010 Apr;140(4):817-22. [View Abstract]

Bailey RL, McDowell MA, Dodd KW, Gahche JJ, Dwyer JT, Picciano MF. Total folate and folic acid intakes from foods and dietary supplements of US children aged 1-13 y. Am J Clin Nutr 2010 Aug;92(2):353-8. [View Abstract]

Basiotis PP, Welsh SO, Cronin FJ, Kelsay JL, Mertz W. Number of days of food intake records required to estimate individual and group nutrient intakes with defined confidence. J Nutr 1987 Sep;117(9):1638-41. [View Abstract]

Dodd KW, Guenther PM, Freedman LS, Subar AF, Kipnis V, Midthune D, Tooze JA, Krebs-Smith SM. Statistical methods for estimating usual intake of nutrients and foods: a review of the theory. J Am Diet Assoc 2006 Oct;106(10):1640-50. Review. [View Abstract]

Harttig U, Haubrock J, Knüppel S, Boeing H; EFCOVAL Consortium. The MSM program: web-based statistics package for estimating usual dietary intake using the Multiple Source Method. Eur J Clin Nutr 2011 Jul;65 Suppl 1:S87-91. [View Abstract]

Murphy SP, Barr SI, Poos MI. Using the new dietary reference intakes to assess diets: a map to the maze. Nutr Rev 2002 Sep;60(9):267-75. Review. [View Abstract]

Nusser SM, Carriquiry AL, Dodd KW, Fuller WA. A semiparametric transformation approach to estimating usual daily intake distributions. J Am Stat Assoc 1996;91(436):1440-9.

Souverein OW, Dekkers AL, Geelen A, Haubrock J, de Vries JH, Ocké MC, Harttig U, Boeing H, van 't Veer P; EFCOVAL Consortium. Comparing four methods to estimate usual intake distributions. Eur J Clin Nutr 2011 Jul;65 Suppl 1:S92-101. doi: 10.1038/ejcn.2011.93. [View Abstract]

Tooze JA, Midthune D, Dodd KW, Freedman LS, Krebs-Smith SM, Subar AF, Guenther PM, Carroll RJ, Kipnis V. A new statistical method for estimating the usual intake of episodically consumed foods with application to their distribution. J Am Diet Assoc 2006 Oct;106(10):1575-87. [View Abstract]

Tooze JA, Kipnis V, Buckman DW, Carroll RJ, Freedman LS, Guenther PM, Krebs-Smith SM, Subar AF, Dodd KW. A mixed-effects model approach for estimating the distribution of usual intake of nutrients: the NCI method. Stat Med 2010 Nov 30;29(27):2857-68. doi: 10.1002/sim.4063. [View Abstract]

Yanetz R, Kipnis V, Carroll RJ, Dodd KW, Subar AF, Schatzkin A, Freedman LS. Using biomarker data to adjust estimates of the distribution of usual intakes for misreporting: application to energy intake in the US population. J Am Diet Assoc 2008 Mar;108(3):455-64; discussion 464. Erratum in: J Am Diet Assoc. 2008 May;108(5):890. Kipnis, Victor [added]. [View Abstract]

Resources