Examining the Association between Diet as an Independent Variable & a Dependent Variable
NOTE: Links for “Row” in the text below will take you to the relevant portion of the Summary Table.
Data Capture Recommendations
Prospective Studies
When researchers are interested in whether diet as an [glossary term:] independent variable may affect a future dependent variable, such as health status, they conduct an observational [glossary term:] prospective study (Row 5).
- For such studies, an approach incorporating multiple administrations of 24HRs in combination with a frequency instrument is ideal and provides maximum analytic flexibility. If you are interested in only one or a few specific dietary components that are concentrated in relatively few food sources, a screener could serve as the frequency instrument, but in most cases an FFQ covering the total diet is preferable. In this case, the 24HR and FFQ data are combined to produce a prediction of usual intake using [glossary term:] regression calibration. The frequency data may be especially helpful for estimating [glossary term:] associations between [glossary term:] episodically-consumed dietary components and another variable (see the 24-hour Dietary Recall Profile and the Food Frequency Questionnaire Profile).
- If it is not possible to administer a combination of instruments in prospective studies aimed at examining associations between diet as an independent variable and a dependent variable, we recommend 24HRs among the full sample together with multiple administrations on at least a subsample. Conduct repeat administrations on non-consecutive days (i.e., separate each 24HR by some number of days). Depending on the number of days reported, the lack of frequency data may considerably reduce the statistical [glossary term:] power available to detect associations between episodically-consumed components and the variable of interest.
- If it is not possible to collect 24HRs from all study participants, use an FFQ as your main instrument and collect 24HR data from a subsample (this is called an [glossary term:] internal calibration sub-study). This will allow you to use regression calibration techniques to adjust for [glossary term:] bias in the FFQ data. Consider collecting [glossary term:] recovery biomarker data for at least a subsample to aid in adjusting for bias for some nutrients (Learn More about Biomarkers). If an internal calibration sub-study is not possible, data from an [glossary term:] external calibration study may be useful if the study populations are similar.
- A screener also may be acceptable if you are interested only in one or a few food groups that are not widely dispersed throughout the food supply (see the Screeners Profile). Collection of less biased dietary data, such as from a 24HR, in an internal calibration sub-study allows adjustment for bias in the screener data. If an internal calibration sub-study is not possible, data from an external calibration study or other [glossary term:] scoring algorithms may be useful if the study populations are similar (Learn More about Scoring Algorithms for Screeners).
- Use of an FFQ or screener alone is possible but not recommended because absolute values are likely to be biased. Energy-adjusted values are likely to be less biased, but as mentioned above, [glossary term:] energy adjustment is not possible with screeners.
- Carefully select your FFQ or screener 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.
Current evidence specific to assessing relationships between diet as an independent variable and a dependent variable suggests a strategy of using one FFQ along with four to six 24HRs collected throughout the baseline year (Learn More about Combining Dietary Assessment Instruments). Ideally, the 24HRs should be collected over the year and spread across the sample to capture all the days of the week and seasons of the year. To measure dietary change over time, consider administering additional dietary assessment instruments throughout the duration of the study.
Cross-sectional Studies
When researchers are interested in whether diet as an independent variable may affect a concurrent dependent variable, such as health status, they conduct a [glossary term:] cross-sectional study (Row 5).
- Analyses using cross-sectional data assume that current usual intake of the dietary component(s) of interest could reasonably be related to a concurrent dependent variable. This type of analysis might be used, for example, to examine current usual intakes of dietary iron in relation to iron-deficiency anemia.
- The recommendations provided above for data capture in prospective studies apply.
Retrospective Studies
When researchers are interested in whether past diet as an independent variable may affect a dependent variable such as health status, they conduct a [glossary term:] retrospective study (also known as a case-control study) (Row 6).
- For this type of study, the only choice is an instrument that queries past diet. Thus, the only possible instruments are an FFQ or a screener. In this case, estimated values are likely to be biased and energy adjustment can be used only with an FFQ to aid in adjusting for [glossary term:] measurement error. In addition, [glossary term:] differential error between the comparative groups may occur.
Data Analysis Recommendations
- In prospective and cross-sectional studies in which the 24HR and FFQ are collected from all participants, recently developed statistical methods for combining data from multiple instruments use [glossary term:] regression calibration to remove the within-person variation in the 24HR data and include information from the FFQ as [glossary term:] covariates to help predict usual intake (Learn More about Regression Calibration). This enables a less biased assessment of the association between diet as the [glossary term:] independent variable and the dependent variable. It is possible to use [glossary term:] recovery biomarkers, if available, as reference instruments to adjust for [glossary term:] measurement error, including [glossary term:] random error and bias, for some nutrients (Learn More about Biomarkers).
- In prospective and cross-sectional studies in which the FFQ or screener is the main instrument and 24HR or recovery biomarker data are available from an internal calibration sub-study or an external calibration study, use statistical techniques to adjust regression coefficients for bias, with the 24HR or recovery biomarker data used as a reference. The most commonly used method is regression calibration, which is used to adjust the estimated relative risk (usually in the direction of deattenuation). In addition, energy adjustment is highly recommended to reduce random error and bias and increase statistical [glossary term:] power. Energy adjustment is not possible with screeners because they do not assess energy.
- NCI has developed [glossary term:] scoring algorithms for a range of screeners. These scoring algorithms enable conversion of responses on the screener to estimates of intake and can be used to reduce bias.
References and Resources
The following references and resources provide additional information on the topics discussed in this section.
References
Carroll RJ, Midthune D, Subar AF, Shumakovich M, Freedman LS, Thompson FE, Kipnis V. Taking advantage of the strengths of 2 different dietary assessment instruments to improve intake estimates for nutritional epidemiology. Am J Epidemiol 2012 Feb 15;175(4):340-7. [View Abstract]
Freedman LS, Schatzkin A, Midthune D, Kipnis V. Dealing with dietary measurement error in nutritional cohort studies. J Natl Cancer Inst 2011 Jul 20;103(14):1086-92. [View Abstract]
Resources
National Cancer Institute. Short Dietary Assessment Instruments.
National Cancer Institute. Usual Dietary Intakes.
National Cancer Institute. Measurement Error Webinar Series.
NCI’s Healthy Eating Index Web site.
National Health and Nutrition Examination Survey (NHANES). NHANES Dietary Web Tutorial.