Baraldi · BMJ open 2018 · Cross-sectional study · n=23,847

Consumption of ultra-processed foods and associated sociodemographic factors in the USA between 2007 and 2012: evidence from a nationally representative cross-sectional study.

Cited 507 times in the scientific literature.

Level 4 - case-series / case-control

Cross-sectional survey analysis of nationally representative data (NHANES)

PubMed 29525772 · doi:10.1136/bmjopen-2017-020574 · record verified 2026-08-28

What was done

Analyzed dietary data from 23,847 individuals aged ≥2 years participating in the cross-sectional National Health and Nutrition Examination Survey (NHANES) across three cycles (2007–2008, 2009–2010, 2011–2012). Dietary intake was collected via at least one 24-hour recall, and foods were categorized according to processing level using the NOVA classification. Linear regression models assessed associations between the dietary caloric contribution of ultra-processed foods and sociodemographic variables as well as time trends.

What was found

Ultra-processed foods contributed almost 60% of total dietary energy intake between 2007 and 2012. Consumption decreased significantly with increasing age and income level, was higher among non-Hispanic whites and non-Hispanic blacks than other racial/ethnic groups, and was lower among college-educated individuals compared to those with lower education levels. Ultra-processed food intake increased significantly across survey cycles by nearly 1 percentage point per cycle, with significant increases specifically observed in males, adolescents, and individuals with a high school education.

Why it matters

This study documents that the majority of calories in the American diet come from ultra-processed foods. It identifies key sociodemographic disparities, highlighting that younger, lower-income, and less-educated groups have the highest intake.

Limits

The cross-sectional design cannot establish temporality or track individual dietary patterns over time. Dietary intake relied on self-reported 24-hour recalls (sometimes only a single recall), which are subject to recall bias and measurement error. Specific numeric effect estimates and confidence intervals for sociodemographic subgroups were not provided in the abstract.

Cited by