Science for a Healthier Tomorrow
A dose-response meta-analysis of 51 prospective cohorts — 8.8 million adults followed for 2 to 32 years — found each extra 100 grams of ultra-processed food per day associated with roughly 14% higher cardiovascular-event risk and 3% higher all-cause mortality. The authors graded the certainty of most of these findings low.
What this means: Across 51 observational studies following about 8.8 million adults, people who ate more ultra-processed food had more cardiovascular events — heart attacks and strokes, including fatal ones. Each extra 100 grams a day was associated with roughly 14% higher risk of those events and 3% higher all-cause mortality. These are associations in free-living populations, not experimental results, and the study’s own authors rate their confidence in them as low. What remains uncertain is what drives the pattern — the food itself, the people who eat more of it, or how intake was measured: here, the hypertension link vanished under two of the three measurements, yet the one that showed an association is the one that made the headlines.
The pattern, and how it was measured
In the pooled analysis, the highest ultra-processed-food eaters had 24% higher cardiovascular-event risk than the lowest (HR 1.24, 95% CI 1.14–1.36), 18% higher all-cause mortality (HR 1.18), 24% higher metabolic syndrome and diabetes (HR 1.24), and 23% higher obesity or overweight (HR 1.23). Further pools linked highest intake to more depression and anxiety (HR 1.27), digestive diseases (HR 1.31) and cancer (HR 1.12), some resting on as few as four cohorts.
The dose-response estimate behind the week’s headlines: each additional 100 grams per day — two servings, on the paper’s own 50-gram serving definition — was associated with 14% higher cardiovascular-event risk (HR 1.14, 1.06–1.22) and 3% higher all-cause mortality (1.02–1.05). These are relative hazards, and neither the main text nor the supplement reports observed absolute event rates — the supplement’s GRADE tables carry only assumed-baseline placeholder values, not observed event rates — so “14% higher” cannot be translated into extra events per 1,000 people.
The analysis was published September 22 in Family Medicine and Community Health under a registered protocol (PROSPERO CRD42023421092); reviewers identified 10,699 records and screened down to 51 studies of adults — 25 European, 16 from the Americas, 8 Asian and 2 Oceanian cohorts, 46% male overall. “Ultra-processed food” followed the NOVA classification.
The hypertension result that didn’t make the headlines
This week’s coverage across roughly 15 outlets ran on three numbers: 14% for heart events, 11% for high blood pressure, 16% for metabolic syndrome and diabetes. The middle number is where headlines and paper part ways.
The 11% is the gram-per-day estimate for hypertension — HR 1.11 per extra 100 g/day (95% CI 1.02–1.21), barely clearing significance. Yet in the same paper, the highest-versus-lowest comparison found no significant hypertension association (HR 1.09, 0.99–1.20), and as a share of the diet — 10% of daily energy — there was no trend at all (HR 1.02, 0.98–1.05). The authors graded the hypertension evidence very low certainty, and the underlying studies disagreed sharply (I²=81%).
This metric dependence runs through the whole paper. For heart events: 14% per 100 g/day versus 3% per each additional 10% of daily energy. For metabolic syndrome and diabetes: 2% per 100 g/day versus 16% per 10% of energy. In each case the circulating figure came from the metric showing the larger association — for hypertension, the only metric showing one. Not necessarily a distortion, but metric choice materially changes the association’s size.
Low certainty, by the authors’ own rating
The authors’ GRADE assessment rates confidence low for cardiovascular events, cancer, all-cause mortality, metabolic syndrome and diabetes, depression and anxiety, and digestive diseases; very low for hypertension; and moderate only for obesity and overweight. The low ratings track the design ceiling: every pooled study is observational; diet was self-reported, with acknowledged under-reporting; and confounding cannot be ruled out — the authors themselves note that ultra-processed-food intake “often co-occurs with broader lifestyle and behavioural patterns.” Publication bias was not assessed for any outcome — the authors cite too few studies per outcome, though their own counts list 12 cardiovascular and 11 cancer cohorts.
NOVA’s breadth adds its own blur: the category lumps wholegrain bread, unsweetened yoghurt and nut products with sodas and processed meats — the authors themselves cite large-cohort examples in which chocolate and nuts showed inverse associations. That breadth limits what any single estimate can mean.
Where this sits in the field
The evidence on ultra-processed food and disease was, until now, all observational. The pattern traces back to single-cohort landmarks — the NutriNet-Santé cohort reported in 2019 that each 10-percentage-point rise in the ultra-processed share of people’s diets came with 12% higher cardiovascular disease risk (HR 1.12, 1.05–1.20, about 105,000 French adults followed a median 5.2 years). The 2024 BMJ umbrella review — 45 pooled analyses, about 9.9 million participants — found direct associations between ultra-processed food and 32 of those 45 analyses (71%), its strongest tier including cardiovascular mortality (RR 1.50, 1.37–1.63), itself graded very low certainty. The new meta-analysis adds the largest dose-response pooling yet without lifting that ceiling.
One experiment sits apart. In a 2019 inpatient trial at the National Institutes of Health, 20 adults lived on matched menus — one ultra-processed, one unprocessed — for two weeks each, and ate roughly 508 extra calories a day on the ultra-processed menu, gaining about 0.9 kg (about 2 lb) while losing a similar amount on the unprocessed one. That trial shows ultra-processed diets can causally drive overeating — but its endpoint was intake and weight, not heart disease, and no trial has yet tested ultra-processed diets against cardiovascular events. The association is mechanism-consistent; the cardiovascular link itself remains untested experimentally.
When every contributing study measures self-reported diets in free-living people, the same unmeasured factors (income, activity, healthcare access) can move every estimate the same way.
Funding and access
The study reports public funding — the Natural Science Foundation of China (grant 82360162) and Jiangxi Province programs — and no food-industry funding is stated in the text we reviewed. The open-access text itself carries no competing-interest statement, but the PubMed record for the paper lists ‘Competing interests: None declared.’ Its DOI address did not resolve when we checked — an apparent registration defect on the publisher’s side — so we link the PubMed and PubMed Central records, which we read in full.
Terms explained
- Ultra-processed food (NOVA): a classification grouping industrial formulations built mostly from refined ingredients and additives — soft drinks, packaged snacks, reconstituted meats. Wholegrain bread and unsweetened yoghurt fall in the same category as soda.
- Hazard ratio (HR): the rate of a health event in one group relative to another over follow-up. HR 1.14 means about 14% more events per extra 100 g/day — a relative comparison, not a count of affected people.
- Dose-response: whether more of an exposure tracks with more of an outcome. A dose-response pattern is often read as supporting causation, but it can also arise when other factors rise along with the exposure.
- GRADE: a structured system for rating confidence that an estimate reflects the true effect. “Low” certainty means the true value is probably near the estimate but could be materially different.
- Residual confounding: leftover distortion from unmeasured factors — sleep, income, smoking, healthcare access — that differ between groups and can partly or wholly explain an observed link.
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 2026;14(Suppl 5), published September 22, 2026. PubMed (PMID 42773062): https://pubmed.ncbi.nlm.nih.gov/42773062/ · Open-access full text (PMC13599868): https://pmc.ncbi.nlm.nih.gov/articles/PMC13599868/
- Lane et al. “Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses.” BMJ 2024;384:e077310. DOI: https://doi.org/10.1136/bmj-2023-077310 · PubMed (PMID 38418082): https://pubmed.ncbi.nlm.nih.gov/38418082/
- Kevin D. Hall, Ayuketah A, et al. “Ultra-Processed Diets Cause Excess Calorie Intake and Weight Gain: An Inpatient Randomized Controlled Trial of Ad Libitum Food Intake.” Cell Metabolism 2019;30(1):67–77.e3. PMID 31105044 · registered NCT03407053. Two published errata correct details without affecting the primary outcome.
- Srour B, et al. “Ultra-processed food intake and risk of cardiovascular disease: prospective cohort study (NutriNet-Santé).” BMJ 2019;365:l1451. PMID 31142457 · doi.org/10.1136/bmj.l1451
- Example coverage from the week’s press wave: Daijiworld, “100g more ultra-processed food daily linked to 14% higher heart risk: Study” (September 25, 2026).
