What We've Learned: Measuring Intakes

To accurately monitor food and nutrient intakes and identify those at risk as well as those who are meeting recommendations, it is necessary to precisely estimate intakes. That has consistently posed a challenge because people don't always report accurately and because dietary assessment instruments contain some degree of error. In addressing these problems through our methodologic research, we've learned a few things:

• Underreporting is a common problem in dietary surveys and in epidemiological studies. Survey respondents whose reported energy intakes are implausibly low -- "low energy reporters" -- are less likely to report a broad range of foods, but when they do, are more likely to report them less frequently and in smaller quantities.
• Our Food Frequency Questionnaire (FFQ), the Diet History Questionnaire, performed best overall in a 1997-1998 comparison to two other FFQs in wide use at that time (the 1995 NCI-Block Health Habits and History Questionnaire and the Willett, purple version) using four 24-hour recalls as the reference instrument. The DHQ was developed with attention to cognitive ease and nutrient database enhancements.
• Analyses using our NHANES Food Frequency Questionnaire (formerly called Food Propensity Questionnaire) show that food frequency data can be used to boost the predictive power of 24-hour recall instruments, thereby providing a way to estimate the distributions of usual intake of episodically consumed foods.
• Cognitive testing of diet questions can make such questions easier for respondents to answer and enhance the accuracy of the responses. The formatting of diet questions affects the rate of unanswered questions. For FFQs, the consecutive question format is superior to the grid format.
• Short questionnaires estimate group mean dietary intakes fairly well among highly educated study populations, but additional work is needed to adapt these instruments to diverse populations.
• Short questionnaires and complete food frequency questionnaires are generally moderately correlated with true intakes. However, studies using these instruments to examine the association between diet and disease would require many more subjects than previously thought in order to attain adequate statistical power.
• The use of biomarkers as reference instruments in dietary intake validation studies suggests a large degree of measurement error and bias in both 24-hour dietary recalls and FFQs. These errors and biases result in attenuation of estimated relative risks in nutritional epidemiology studies. This attenuation is more severe than previously thought. Thus, FFQs and 24-hour recalls may not be able to detect moderate but important associations in diet and cancer studies.