The cohort study included 718 mostly white (93%) adults between the ages of 50 and 74. The participants provided up to six 24-hour dietary recalls (97% provided at least 4) and 2 samples of urine and blood over a 12-month period. Ultraprocessed food intake was estimated via dietary recalls and averaged 50% of participants’ total energy.

The investigators measured more than 1,000 metabolites in the participants’ blood and urine. They found that 191 blood and 212 urine metabolites were correlated with ultraprocessed food intake. Common metabolites were amino acids, xenobiotics (substances that don’t naturally occur in the body), and carbohydrates. Using machine learning, the researchers selected 28 blood metabolites and 23 to 33 urine metabolites as robust predictors of ultraprocessed food intake. The specific number of urine metabolites depended on when urine was collected, either once in the morning or over 24 hours.

Metabolites in blood and urine that were especially robust predictors of ultraprocessed food intake included the following:

N6-carboxymethyllysine (higher levels correlated with higher UPF intake)
S-Methyl-L-cysteine sulfoxide, or SMCSO (lower levels correlated with higher UPF intake)
N2,N5-diacetylornithine (lower levels correlated with higher UPF intake)
Pentoic acid (lower levels correlated with higher UPF intake)

The 28 blood metabolites and 23 to 33 urine metabolites were used to create a polymetabolite score designed to be predictive of ultraprocessed food intake.

To validate the polymetabolite score, the investigators analyzed the blood and urine samples of 20 participants from a tightly controlled crossover trial. In the trial, the participants (average age of 20, average BMI of 22) ate diets that contained 80% and 0% energy from ultraprocessed foods for 2 weeks each.^[2]

The researchers confirmed the efficacy of the polymetabolite score because they found that the scores were higher when the participants were eating the diet high in ultraprocessed food than when they were eating the diet devoid of ultraprocessed foods.

Using metabolites in blood and urine to predict ultraprocessed food intake Original paper