The End of Dietary Guesswork? How Metabolomics is Revolutionizing Nutritional Epidemiology

By Mauro Proença — Jun 13, 2025
Forget squinting at a food diary and hoping someone remembered how many cookies they ate last Tuesday. Nutritional science might finally trade in its crystal ball for a blood test, thanks to the emerging power of metabolomics. The latest research suggests we are one step closer to cracking the diet and disease code with hard biochemical evidence instead of dietary guesswork.
Generated by AI

Much of nutritional epidemiology still relies on methods that resemble guesswork more than science. For those familiar with detective stories where the protagonist relies solely on intuition rather than objective evidence to uncover the true culprit, the method used by Miss Marple, Agatha Christie’s character, offers an almost perfect analogy for how nutritional epidemiology often produces astonishing results. This helps explain the emergence of outlandish claims, such as the idea that any ultra-processed food, regardless of its nutritional content, increases all-cause mortality. It also clarifies why eggs, in one study, appear strongly linked to mortality, while in another show no association.

Such methodological inconsistency has consequences that extend beyond academic circles, spreading misinformation, leading professionals to offer misguided advice, and leaving the public uncertain about what constitutes a healthy diet. It undermines trust in the field, reinforcing the popular notion that nutritionists seem just as confused as anyone else about what to eat.

Many factors contribute to these issues, but I believe the main ones are:

  • Reliance on observational studies to assess whether a food benefits or harms health.
  • The difficulty in conducting large-scale, long-term, randomized clinical trials — which would require literally kidnapping people and holding them hostage to control their diets and all confounding variables.
  • Excessive confidence in self-report tools, which do not accurately capture actual intake and are plagued by limitations and biases.

While the first two problems remain unresolved, a new paper published in PLoS Medicine may mark the first steps toward a potential revolution in the field. However, to grasp this potential revolution, we must first examine the systemic flaws in the tools currently in use.

Our current toolbox

In clinical practice and observational studies, a standard method for assessing food consumption is through self-report questionnaires, such as the 24-hour recall (R24), where individuals record everything they eat, along with the quantity, over 24 hours. Alternatively, there are food frequency questionnaires (FFQs) that record how often certain foods or food groups are consumed over a week and food diaries that record all meals and quantities over multiple days, typically two weekdays and one weekend day.

The entries enable the assessment of caloric intake and the identification of nutrients, such as trans and saturated fats, protein, and sugars. Though internationally validated and undeniably useful, these tools have critical limitations that can significantly influence research outcomes.

For example, a 2008 study in the Nutrition Journal used a brief FFQ and R24 to assess social approval bias —the tendency to respond in a socially desirable way —in reports of fruit and vegetable consumption.

Participants in the intervention group received a refrigerator magnet and a letter featuring colorful graphics of fruits and vegetables, along with a message highlighting the benefits of these foods and their role in disease prevention. The control group received a black-and-white letter without the message or additional materials.

As a result, the intervention group reported higher fruit and vegetable intake on the FFQ, with no significant differences in milk, potatoes, or sweets. In the R24 responses, they also reported a higher intake of all meals and snacks on the previous day. The authors concluded that participants may have either increased their actual intake or exaggerated their reports to align with the message and its implicit expectations.

A decade later, an article in the Journal of Clinical Epidemiology argued that memory-based dietary assessment methods are invalid and inadequate for supporting public policy. The authors noted that participants may lie or alter their diets when asked to report consumption — a phenomenon known as the Hawthorne effect. A 1997 study found that 46 out of 100 participants reported changing their diets during the study. Of these, 20 cited greater awareness of their eating habits and feelings of shame or guilt when recording certain foods or quantities. As one participant said, "I didn't dare write down all the cookies I usually eat."

These tools do not measure actual intake; at best, they capture perceptions of intake.

Beyond this, there is the impact of data extrapolation: the assumption that dietary patterns captured by these tools accurately represent an individual’s typical diet over time. In observational studies, particularly amid growing interest in ultra-processed foods, researchers often overlook this limitation. As a result, studies frequently report statistically significant associations that are likely driven by unmeasured confounding factors.

However, we may now be witnessing the first steps of a new approach that, if replicated and adapted to new variables, could transform or at least reframe this research.

The New Study

The research published in PLoS Medicine had two main objectives. The first objective is to identify metabolites associated with the intake of ultra-processed foods over a twelve-month period, with the goal of developing a scoring system capable of estimating, under real-life conditions, the consumption of these foods based on blood and urine test results. The second was to test, in a randomized controlled crossover clinical trial, whether these scores could distinguish diets providing 0 percent and 80 percent of their energy from ultra-processed foods.

The scoring system drew on blood samples, 24-hour urine samples, and first-morning urine samples, along with dietary data provided by 718 participants, to create a plasma and urinary metabolomic profile.

When analyzing the correlation between metabolites and the intake of ultra-processed foods, the researchers found statistically significant correlations in:

  • 191 of the 952 metabolites in blood samples
  • 293 of the 1,044 metabolites in 24-hour urine samples
  • 237 of the 1,043 metabolites in first-morning urine
  • In aggregate, 470 metabolites were detected in both serum and urine.

To create the scores that would classify the diet as "rich or not in ultra-processed foods" (rich in this instance, 58.2% of total energy from UPFs), the researchers selected five serum metabolites for serum and twenty-one urinary metabolites. The analyses showed that these scores provided moderate discrimination between participants with higher and lower consumption of ultra-processed foods.

When testing the scoring system in a clinical trial, researchers recruited 20 healthy adults who followed diets composed of either 80% or 0% UPFs for two weeks. The participants then cross over to the alternative diet. Metabolite levels after each diet phase were compared.

The results showed substantial variation in metabolite levels within individuals between the 80% and 0% UPF phases. While the overall poly-metabolite scores captured these differences, only a few individual metabolites showed consistent changes across diet phases. Moreover, a threshold appeared to exist, at roughly a 30% UPF diet, at which the metabolic pattern changed. 

These findings suggest that metabolomic profiles reflect dietary patterns characterized not only by high UPF intake but also by low consumption of natural foods such as fresh fruits and vegetables.

As is common in many observational studies and small clinical trials, this research has significant limitations. In addition to using online dietary recalls, which introduce potential measurement errors, the small sample size of this proof-of-concept study limits its accuracy and generalizability. Perhaps the most significant limitation lies in the lack of a widely accepted cutoff point for defining what constitutes a diet rich in ultra-processed foods. The study uses 0% ultra-processed food energy intake as a theoretical reference for an ideal diet; however, this does not reflect the scientific consensus or real-world eating patterns.

Despite these limitations, the authors emphasize that these findings represent an initial baseline and that the metabolite-based scores can be refined and subject to more diverse datasets. Despite my typical skepticism toward new findings, this is one of the rare cases where I feel moderately optimistic. I recognize there is still a long way to go and that the effectiveness of these scores will be questioned or perhaps refuted by future studies.

In my view, the journey toward truly evidence-based nutritional epidemiology is just beginning. But as Agatha Christie’s novels remind us, even the most complex mysteries can be solved when we replace guesswork with rigorous investigation.

Sources:  Effects of social approval bias on self-reported fruit and vegetable consumption: a randomized controlled trial. Nutrition Journal. DOI: 10.1186/1475-2891-7-18

Controversy and debate: Memory-Based Methods Paper 1: The Fatal Flaws of food frequency questionnaires and other memory-based dietary assessment methods. Journal of Clinical Epidemiology. DOI: 10.1016/j.jclinepi.2018.08.003

 

Identification and validation of poly-metabolite scores for diets high in ultra-processed food: An observational study and post-hoc randomized controlled crossover-feeding trial. PLoS Medicine. DOI: 10.1371/journal.pmed.1004560

 

 

Category