Do studies on ultra-processed food intake suffer from confounding and measurement bias?

What is confounding, and why does it matter for UPF studies?

Confounding happens when a third factor—like overall diet quality, income, or exercise—is linked to both UPF intake and the health outcome, making it look like UPF is the cause when it might not be. For example, people who eat more UPFs also tend to smoke more, exercise less, and have lower fruit and vegetable intake [3][4][7]. If a study doesn't adjust for these, the apparent harm of UPFs could be overstated. The good news is that the strongest studies here—large prospective cohorts and meta-analyses—do adjust for many confounders. The Framingham Offspring Study, for instance, adjusted for age, sex, education, alcohol, smoking, and physical activity, and still found a 7% higher risk of hard cardiovascular disease per daily serving of UPF [9]. Similarly, a meta-analysis of cancer studies adjusted for obesity and total energy intake and still found a 13% higher risk of overall cancer per 10% increase in UPF [1]. So while confounding is a real concern, the adjustments used in the best studies are reasonably thorough.

Measurement bias: How accurate are the diet tools used?

Measurement bias is a major issue because most UPF studies rely on food frequency questionnaires (FFQs) that were not designed to classify foods by processing level. A systematic review of 99 studies noted that most observations came from FFQs 'not specifically validated for UPF,' which can misclassify foods—for example, calling a homemade soup 'unprocessed' and a canned soup 'ultra-processed' even if their nutritional profiles are similar [8]. This misclassification can dilute or exaggerate associations. However, the same review found that despite this limitation, the associations between UPF intake and health outcomes were consistent across countries and study designs [8]. The Korean study on type 2 diabetes used a validated FFQ and still found a 34% higher risk of diabetes in the highest UPF quartile [6]. So while measurement error exists, it likely biases results toward the null (making it harder to find an effect), meaning the true associations may be even stronger than reported.

Do these biases invalidate the findings?

No, but they do mean we should interpret results cautiously. The evidence is strongest for outcomes like cardiovascular disease, type 2 diabetes, and certain cancers, where multiple large studies converge. For example, the PURE study across 21 countries found a 82% higher risk of inflammatory bowel disease for those eating 5+ servings of UPF per day vs. less than 1 serving, after adjusting for confounders [5]. The EPIC cohort analysis on type 2 diabetes also found a positive association, though it noted that some UPF subgroups (like breads and cereals) appeared protective, highlighting the complexity of the NOVA classification system [10]. A key limitation is that most studies are observational, not randomized controlled trials (RCTs), so they cannot prove causation. The only RCT cited here was a small feeding trial that showed UPF promotes weight gain [4], but it was short-term. The authors of the meta-analyses consistently call for 'further rigorously designed prospective and experimental studies' [2] and note that 'these findings reflect associations rather than causal relationships' [3]. So while confounding and measurement bias are present, the consistency, dose-response, and biological plausibility (e.g., UPFs are often high in sugar, salt, and additives) suggest the associations are real, not artifacts.

About These Sources

This answer is built on 10 peer-reviewed studies — published from 2021 to 2026, 3 from 2024 or later, 8 in Q1 journals, collectively cited 1,031 times — selected as the most relevant from 15 studies that passed quality screening, drawn from 65 papers retrieved from a database of over 500 million.

Sources used in this answer

Related research questions