Is Ultra-Processed Food 45% or Just 23% of Calories?

In three Swedish national dietary surveys, ultra-processed food supplied 45% of adults’ calories under one classification and 23% under a stricter one — and higher intake tracked with poorer diet quality, with a negligible adult weight difference and none at all in children.

Three Swedish national dietary surveys recorded what 6,716 people ate. Ultra-processed food supplied 44.9% of adults’ energy, 51.8% of adolescents’ and 46.3% of young children’s under the main classification, or 22.5%, 25.9% and 29.2% under a stricter one. Higher intake tracked with poorer diet quality on most measures, the adult weight difference was small, and in children there was none.

What this means

The same Swedish data give a 45% ultra-processed diet or a 23% one depending on documented classification choices, so a headline of this kind describes a method as much as a food supply. Higher intake tracked with worse diet quality and a slightly higher adult BMI, at a size the authors call unlikely to be clinically meaningful; adolescents and 4-year-olds showed no association. Diet and weight were measured together, so nothing here shows that ultra-processed food contributes to weight gain.

What three surveys found

The Riksmaten surveys are Swedish national dietary studies: adults aged 18 to 80 in 2010-2011, students in grades 5, 8 and 11 in 2016-2017, and children aged 18 months and 4 years in 2021-2024. Diet analyses covered 1,797 adults, 3,099 adolescents and 1,820 young children; weight analyses, after exclusions, 1,671 adults, 2,865 adolescents and 665 four-year-olds.

Foods were sorted by NOVA, the four-level classification whose top group is ultra-processed, with a written protocol that disaggregated recipes. Two classifiers worked independently, agreeing on 86.9% of items; the authors ran lower-bound, middle-bound and upper-bound scenarios and designated the middle one primary.

That middle-bound number is the one quoted: 44.9% of adults’ energy (SD 13.3), 51.8% of adolescents’ (15.8) and 46.3% of young children’s (15.5). The lower bound gives 22.5%, 25.9% and 29.2% on the same data; the upper bound 51.5%, 63.3% and 52.7%. In adults that span runs 22.5% to 51.5%, created by classification decisions alone; the authors say comparing their figures across protocols is unsafe.

Worse diet quality, barely different weight

Diet quality and micronutrient density were modeled per 10 percentage points of ultra-processed energy. In adults, higher intake went with 72 kcal a day more total energy (95% CI 46.2 to 97.2), higher energy density, and lower fiber, whole-grain and most micronutrient densities. Two results cut against a simple junk-food reading: saturated fat density showed no association in adults or adolescents, and salt density was lower with higher intake in all three groups. In young children, iron and vitamin E density were higher.

Among adults, each 10 percentage points more ultra-processed energy went with a BMI 0.61% higher (95% CI 0.04 to 1.18; p=0.036) in the primary model. Overweight or obesity was 12% more prevalent in the top tertile than the bottom (PR 1.12, 95% CI 1.00 to 1.27; p=0.060) and 16% more prevalent after adjustment for total energy (PR 1.16, 1.03 to 1.30). The authors call that magnitude unlikely to be clinically meaningful. No model showed a significant association in adolescents or 4-year-olds: after energy adjustment, the prevalence ratio was 0.94 (0.79 to 1.13) in adolescents and 1.34 (0.73 to 2.47) in 4-year-olds.

What the category actually contains

Bread was the largest single source of ultra-processed energy for adults, at 9.5% of total energy. Of 2,218 reported food items, 990 (44.6%) were entirely ultra-processed under the primary scenario, and only 8.5% of those met the authors’ criteria for generally encouraged choices. A category that puts bread at the top of the list, yet judges most of its members not generally encouraged, is not a synonym for junk food — its boundaries decide the answer.

What the design cannot settle

The design is cross-sectional: diet and weight were measured together, once. That cannot separate cause from consequence: people with a higher BMI may eat differently, and the authors note that neither reverse causation nor selective under-reporting by people with obesity can be excluded. With no follow-up, nothing here speaks to weight change.

Participation was 37% for adults, 60% for adolescents and 11% for young children, with no survey weights applied; the samples over-represent more educated households, the authors say. These are national samples, not a clinical population.

One sensitivity analysis matters. Excluding people whose reported energy intake looked implausible made the weight associations stronger, not weaker; the authors read that as attenuation by reporting bias, concealing an association rather than manufacturing one. That is an interpretation, not a test.

Where this sits

The association’s direction holds in prospective cohorts. A 2026 meta-analysis of 51 cohorts covering 8,819,894 adults reported a hazard ratio of 1.23 (1.17 to 1.30) for overweight or obesity. A 2024 BMJ umbrella review of 45 pooled analyses found direct associations with 32 outcomes and graded the evidence convincing for cardiovascular mortality (RR 1.50, 1.37 to 1.63), a type 2 diabetes dose-response link (RR 1.12, 1.11 to 1.13), prevalent anxiety outcomes and combined common mental disorder outcomes; obesity was “highly suggestive” (OR 1.55, 1.36 to 1.77). Both are open access, and the figures quoted here were verified against the full texts as well as the abstracts.

What this adds is a demonstration that the exposure number is a choice: one dataset, 22.5% or 44.9% of adult calories — a caution that applies to every “ultra-processed food is X% of your diet” figure. Whether the category predicts health outcomes beyond nutrient profile is openly debated, and nothing here settles it. For whether a processed food can be nutritious, see is-healthy-processed-food-as-good.

Funding and interests

The first author’s PhD position was funded by the Swedish Research Council for Health, Working Life and Welfare (2024-00771) and the Swedish government and county councils’ ALF agreement (ALFGBG-1006385); the authors state the work had no other funding. Rikard Landberg, an author, is Nutrition Journal’s editor-in-chief; he had no role in peer review or the editorial decision, which went to another editor. ChatGPT-5 helped develop statistical code and improve structure and readability; the first author takes responsibility.

Terms explained

NOVA classification — a four-level system that sorts foods by industrial processing, not nutrients; its top level, ultra-processed, is this study’s exposure.
Energy percentage (E%) — the share of total calories from a given food group; 44.9 E% means 44.9% of calories.
Prevalence ratio — one group’s share with a condition divided by another’s; 1.16 means the top group’s prevalence was 16% higher.
Classification scenario — a written rule set for where borderline foods land; here strict, primary and loose versions over identical data.

Sources

Sällström T, Wadell AT, Wanselius J, Wallengren O, Åkesson A, Landberg R, Lindroos AK, Patterson E, Moraeus L, Karlsson T, Bärebring L. “Intake of ultra-processed foods in Sweden and its association with diet quality and weight status: cross-sectional analyses from three national dietary surveys.” Nutrition Journal. 2026;25(1):110. Published Oct. 13, 2026. doi:10.1186/s12937-026-01408-5 · PMID: 42870269 · https://pmc.ncbi.nlm.nih.gov/articles/PMC13663064/
Lane MM, et al. “Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses.” BMJ. 2024. doi:10.1136/bmj-2023-077310 · PMID: 38418082 · https://pmc.ncbi.nlm.nih.gov/articles/PMC10899807/ — full text read.
Liu X, et al. “Ultra-processed food consumption and risk of multiple chronic diseases among 8,819,894 adults from 51 prospective cohorts: a dose-response meta-analysis.” Family Medicine and Community Health. 2026. doi:10.1136/fmch-2026-003925 · PMID: 42773062 · https://pmc.ncbi.nlm.nih.gov/articles/PMC13599868/ — full text read.