I know what you ate last summer
"The Art of Forgetting: Why Recalling What We Eat Isn’t As Simple As It Sounds"
What did you have for dinner last night? And for breakfast? Did you snack in between? Even if you remember all this, do you recall how much you ate? Was the cucumber sandwich 100g or 150g? How much smoked salmon was on the canapés?
Answering these questions correctly is not just a party trick—it’s essential for understanding the link between diet and health. If we want to know what makes a diet “healthy,” we first need to know what healthy people eat. And as simple as that sounds, it’s anything but—partly because “healthy” is such a subjective term, and partly because measuring dietary intake is a very complicated process.
The challenges of measuring diet
There are many different ways to find out what people eat - and one of the most common and most reliable methods is the so-called "24h recall". A researcher asks a participant what they have consumed in the last 24h and uses prompts and structured questions to help participants to remember as much as possible. They are often conducted using web-interfaces such as Intake 24 or MyFood 24. Most people remember reasonably well the last 24h - and with a little prompting and pictures, such recalls can be surprisingly accurate (event though we know of course that eye witness accounts are not always very trustworthy).
Since dietary assessment is such a crucial task, there have been decades of research on how to do it best. There are different methods - broadly divided into forward-looking (food diaries) and backward looking (recalls and questionnaires). Questionnaires - food-frequency questionnaires (FFQs) to be precise - are very popular but have their own problems as they cover a very long time and include a limited number of foods.
Energy intake - a special case
Energy intake is obviously of particular interest: people consuming more energy than they need tend to gain weight - those who consume less tend to lose weight. Nutritional energy is an odd thing anyway - but this is a story for another day. Here, I'd like to focus on how we can measure energy intake - and what it tells us about nutrition research.
Isotopes to the rescue
The most reliable way to measure energy intake in "free-living" people is a method called "double-labelled water". It works by tracking how quickly the body eliminates two isotopes: deuterium (²H) and oxygen-18 (¹⁸O) - which can make up water. Participants drink water enriched with these isotopes, which are naturally occurring and perfectly safe in the amounts consumed. Deuterium exits the body through water loss (e.g., sweat, urine), while oxygen-18 exits both as water and as carbon dioxide from respiration. By analysing urine or saliva samples over several days, scientists can measure the difference in isotope elimination rates, which directly correlates to the amount of carbon dioxide the body produces. Since energy expenditure is closely linked to carbon dioxide production, the DLW method provides an accurate measure of how many calories someone is burning. While incredibly precise, it’s also expensive and resource-intensive, making it impractical for widespread use in large studies. However, it remains invaluable for validating dietary intake data and uncovering the extent of misreporting in nutrition research.
Incidentally - energy is one of the few aspects of diet that can be measured accurately with biomarkers (the other is protein intake). This means energy intake is also one of the best parameters to investigate the reliability of dietary assessment methods.
Two decades ago, the OPEN study investigated how reliable food-frequency questionnaires are - and unsurprisingly, they are not very reliable (some would say not suitable) to measure energy intake reliably. However, how reliable are other methods? A large study has investigated how well 24h recalls work for energy intake - and the results are ... well interesting is the best description.
How bad is misreporting?
The study made use of national surveys - NHANES in the US and the NDNS in the UK, arguably some of the best national surveys. They found that the level of misreporting was >50%. This is an incredibly useful study for scientists as it also offers methods that can help to address some of these problems - but it also offers an insight into the problems of nutrition research, and why the tools we use are probably not suitable to answer the questions we have.
The energy we eat is made up of carbohydrates, fat and proteins (and for some people alcohol). This is already a simplification, because all of these three groups are already incredibly complex - we know that some fatty acids are bad, others are good for health; and carbohydrates include anything from sugar in diet coke to complex carbohydrates in artisan sourdough bread. But there is already a lot of controversy discussion about the benefits or dangers of carbohydrates. Much of these discussions are based on data from epidemiological studies - which rely on, you guessed it, self-reported dietary intake data.
Unsurprisingly, people with a higher BMI tend to underreport how much they; we have shown in the past that people who are overweight underreport sugar intake - resulting in the very nice but unfortunately wrong impression that sugar consumption leads to weight loss.
But it gets more complicate: people with a higher BMI overreport their protein intake and underreport their carbohydrate intake. What does this mean? The authors of the study describe it very well:
Applying the tool to two large surveys suggested that more than 50% of the dietary reports had implausible energy intakes and probably therefore erroneous intake of macro- and micronutrients. Ultimately, the main benefit of this tool is that it may highlight the true level of dietary misreporting when using existing methods and drive us towards innovating radical approaches that do not rely so much (or at all) on self-report.
Such a study should really be a wake-up call for nutrition research. In an aptly-titled feature, Science dedicates this study and the ensuing issues a long article. But is it likely that we see a sea change? Will researchers abandon old methods and focus on more reliable approaches - from regression calibration to address measurement error to using biomarkers? There is no doubt that we need better methods if we truly want to understand the link between diet and health - and make reliable recommendations. Let’s hope that will happen now.