Ultra-processed foods are associated with higher rates of seven major health outcomes, obesity and overweight among them, according to a dose-response meta-analysis of 51 prospective cohort studies covering 8,819,894 adults, published on 22 September 2026 in Family Medicine and Community Health. People in the highest intake groups had a 23% higher rate of obesity or overweight, and an 18% higher rate of death from any cause, than those in the lowest. Every study pooled was observational, and the authors graded the certainty of their own evidence as low or moderate.
What the pooled data show
The authors searched for prospective cohort studies published up to April 2026 in which ultra-processed food intake was measured by food frequency questionnaire, 24-hour recall or dietary history, in adults aged 18 or over with no restriction on baseline health. Fifty-one cohorts met the criteria.
Comparing highest intake with lowest, the pooled hazard ratios were 1.24 for cardiovascular events (95% confidence interval 1.14 to 1.36), 1.12 for cancer (1.04 to 1.21), 1.23 for obesity or overweight (1.17 to 1.30), 1.24 for metabolic syndrome or diabetes (1.08 to 1.42), 1.27 for depression or anxiety (1.15 to 1.40), 1.31 for digestive disease (1.16 to 1.48) and 1.18 for all-cause mortality (1.10 to 1.26). A hazard ratio of 1.23 means the rate in the high-intake group ran 23% higher over follow-up. The interval of 1.17 to 1.30 is the range of values the data are compatible with, and because it sits entirely above 1.0, chance alone is an unlikely explanation for the direction of the finding.
The dose-response models are more modest, and they are the more informative half of the paper. Each additional 100 grams of ultra-processed food per day was associated with a 14% higher rate of cardiovascular events (1.06 to 1.22), 4% for cancer (1.02 to 1.06), 3% for all-cause mortality (1.02 to 1.05) and 2% for metabolic syndrome or diabetes (1.01 to 1.04). Those increments compound across a large habitual intake, but each one is small enough that modest confounding could account for it.
Why the size of the association is not the whole story
None of this is a randomised trial. Every included study followed people who chose what they ate, which leaves two problems the design cannot solve. The first is confounding: high ultra-processed food intake travels with smoking, lower income, less physical activity and higher total energy intake, and statistical adjustment removes only what was measured, and measured well. The second is exposure measurement. Food frequency questionnaires and 24-hour recalls were built to capture nutrients, not degree of industrial processing, so sorting a recorded diet into processing categories after the fact introduces misclassification.
The obesity result carries a further difficulty. Weight change is both a plausible consequence of the diet and a reason people describe their diet differently when asked. That is the kind of loop a cohort cannot untangle, and it is why the authors framed their conclusion as reducing intake “may” be relevant for chronic disease prevention, a weaker claim than the effect sizes read on their own would support. Set against the drug trials that dominate our research coverage, this is evidence about risk across a population rather than about how any one person’s weight responds to a change.
What the analysis does not establish is that ultra-processed food is itself the cause rather than a marker of a wider eating pattern, nor which ultra-processed foods carry the risk: the category spans packaged bread, breakfast cereal, soft drinks and confectionery, and the meta-analysis pools them into one exposure. It also cannot say what happens when intake falls, because no cohort tested that. Answering it would take randomised feeding studies, which this synthesis did not include and which are hard to run for long enough to measure cancer or death.
Sources: Liu X, Luo S, Zeng Y, 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, 22 September 2026;14(Suppl 5):e003925. doi:10.1136/fmch-2026-003925. PROSPERO registration CRD42023421092. PubMed record (PMID 42773062)
