Table of Contents
- Introduction
- Co-Hosts for this Episode
- Overview
- Premium Content
- Detailed Study Notes
- Transcript
Introduction
Food frequency questionnaires (FFQs) have been widely employed in nutrition research to assess dietary intake patterns among study participants. However, debates surrounding the reliability of FFQs have persisted both inside and outside the academic community.
These debates primarily revolve around issues related to measurement error, recall bias, and the appropriateness of FFQs for diverse populations.
One prominent concern is the potential for measurement error in FFQs. These questionnaires rely on self-reported data from participants, which can introduce inaccuracies due to memory limitations and social desirability bias. Participants may not accurately recall their food consumption frequencies and portion sizes, leading to imprecise estimates of nutrient intake.
Recall bias is another critical issue in the reliability debate. Participants may selectively remember or misreport the consumption of certain foods or nutrients, leading to an overestimation or underestimation of actual dietary intake.
Two concepts are crucial to understand: validity and reproducibility. FFQs are validated by cross-referencing the FFQ data with other dietary assessment tools (or other methods). It’s also important to consider if an FFQ gives reproducible results when used on multiple occasions.
When we ask “are FFQs reliable?”, we must first understand the conceptual exposure of interest: average intake over time. Second, we must consider what nutrients we are looking at. And third, in what population.
In this episode, Danny & Alan discuss the reliability of FFQs and how to have a deeper, more accurate understanding of their use. They take a look at valid critcisms of FFQs, as well as some of the more ill-informed criticisms.
Co-hosts for this Episode
Dr. Alan Flanagan has a PhD in nutrition from the University of Surrey, where his doctoral research focused on circadian rhythms, feeding, and chrononutrition.
This work was based on human intervention trials. He also has a Masters in Nutritional Medicine from the same institution.
Dr. Flanagan is a regular co-host of Sigma Nutrition Radio. He also produces written content for Sigma Nutrition, as part of his role as Research Communication Officer.
Danny Lennon has a master’s degree (MSc.) in Nutritional Sciences from University College Cork, and he is the founder of Sigma Nutrition.
Danny is currently a member of the Advisory Board of the Sports Nutrition Association, the global regulatory body responsible for the standardisation of best practice in the sports nutrition profession.
Overview
- What Are FFQs?
- Advantages of FFQs?
- Assessing FFQs
- What Are The Main Criticisms or Limitations?
- Validation of FFQ Method
- Importance of Population Specificity
Premium Content
Not a Premium subscriber? Subscribe here!
Login
Comments
Excellent episode. One thing that confused me (though I may be missing something obvious) is that at 47:30, you said that because we get correlation coefficients of 0.2-0.7 when validating an FFQ, this means the FFQ is giving us an underestimate (iirc you said this somewhere towards the start of the episode too). But I don’t see how that follows – there’s nothing to stop one variable from being a substantial overestimate of the other with a correlation coefficient of e.g. 0.7 because it’s only describing strength of association. If you’re looking at the correlation coefficient between variables x and y, you can transform variables by scaling (e.g. x -> 2x) and the correlation coefficient stays the same; that is to say, an overestimate could be transformed into an underestimate and vice versa without changing the correlation coefficient, so the coefficient doesn’t tell you about relative over- or underestimates.
I’m probably explaining this really badly, but hopefully it gets the point across. (I don’t disagree that FFQs usually underestimate intake btw, just with the idea one can derive this from the correlation coefficient.)
Hey Lee,
Thanks for the kind words. Apologies for the confusion, perhaps we could have phrased it better. But essentially, the point about underestimation is not inferred directly from the correlation coefficient. The correlation coefficient, as you correctly note, is describing the level of agreement. We know the direction (i.e. FFQs underestimate intake relative to others) from the absolute numbers seen in validation studies. For example, a food record says someone eats 100g, but when filling out the FFQ it may then be recorded as ~70g. So the statement about FFQs underestimating intake is not made on the basis of deriving that from a correlation coefficient of X, rather these are two separate but related points. Hopefully this clarifies things!