Ultra-Processed Foods: What the Evidence Actually Shows

Ultra-processed foods are the most argued-about subject in nutrition right now, and the real evidence is stronger than the sceptics allow but weaker than the headlines claim. In the most tightly controlled test yet — 20 adults living on a metabolic ward — people ate roughly 500 extra calories a day and gained weight when identical nutrients arrived in ultra-processed form. Cohort studies covering millions link high intake to cardiometabolic disease and earlier death. But the classification behind the term is genuinely contested among nutrition scientists, and the category is not uniformly harmful: some foods inside it show no association with disease at all.

The trial that changed the conversation

Nearly all diet research is observational, which is why one study dominates this field. Hall and colleagues ran a randomised controlled crossover feeding trial published in Cell Metabolism in 2019, with 20 weight-stable adults (mean age 31.2 years, mean BMI 27) admitted as inpatients to the NIH Clinical Center. Each participant spent two weeks on an ultra-processed diet and two weeks on an unprocessed diet, in random order. Crucially, the two menus were matched for presented calories, energy density, macronutrients, sugar, sodium and fibre. Participants were told to eat as much or as little as they wanted.

Energy intake was 508 (standard error 106) kcal per day higher on the ultra-processed diet (p = 0.0001), split between carbohydrate (280 kcal/day) and fat (230 kcal/day), with no difference in protein. Participants gained 0.9 kg (SE 0.3) over the two ultra-processed weeks and lost 0.9 kg over the two unprocessed weeks, and weight change tracked energy intake closely (r = 0.8). Because the diets were matched on the nutrients dietary guidelines usually target, something other than sugar, fat, salt or fibre content was driving the extra 500 calories.

Funding was NIH intramural. Of the 24 authors, one declared industry ties (speaking reimbursements from companies selling nutritional products, a scientific advisory council role, and consortium research funding from Abbott Nutrition, Nestec and Danone); the rest declared none. Two published corrections have been issued since the paper appeared. Twenty people over four weeks is a small study — it is the design, not the sample size, that gives it weight.

A second trial, run inside official dietary guidelines

The obvious objection to the NIH trial is that its ultra-processed menu, while nutrient-matched, was not what anyone would call a healthy diet. Dicken and colleagues answered that in a randomised 2 x 2 crossover feeding trial published in Nature Medicine in 2025. Fifty-five adults in England with a BMI between 25 and 40, all habitually getting at least half their calories from ultra-processed food, were provided with two eight-week ad libitum diets — one minimally processed, one ultra-processed — and both were built to comply with the UK Eatwell Guide. Fifty participants formed the intention-to-treat sample.

Both diets produced weight loss. The minimally processed diet gave a mean weight change of -2.06% (95% CI -2.99 to -1.13) and the ultra-processed diet -1.05% (95% CI -1.98 to -0.13). The difference favoured minimally processed food but was modest: -1.01 percentage points (95% CI -1.87 to -0.14; p = 0.024; Cohen’s d -0.48). The honest reading is that processing mattered, and that an ultra-processed diet meeting national nutrition guidance still produced weight loss. The first author disclosed book royalties on a title mentioning ultra-processed food and consultancy payments; the trial was publicly and charitably funded.

What the population data show

Long-term outcomes — heart attacks, diabetes, death — cannot be tested on a metabolic ward, so this evidence is observational.

The umbrella review

Lane and colleagues published an umbrella review in The BMJ in 2024 covering 45 pooled meta-analyses and roughly 9.9 million participants. Using pre-specified credibility criteria, it graded as convincing the associations between higher ultra-processed food exposure and cardiovascular-disease-related mortality (risk ratio 1.50, 95% CI 1.37 to 1.63) and type 2 diabetes (dose-response risk ratio 1.12, 95% CI 1.11 to 1.13), plus anxiety (odds ratio 1.48, 95% CI 1.37 to 1.59) and common mental disorders (odds ratio 1.53, 95% CI 1.43 to 1.63). Highly suggestive evidence covered all-cause mortality (risk ratio 1.21, 95% CI 1.15 to 1.27), obesity (odds ratio 1.55, 95% CI 1.36 to 1.77) and depressive outcomes (hazard ratio 1.22, 95% CI 1.16 to 1.28).

The part rarely quoted: under the GRADE framework, only four of those 45 pooled analyses were rated moderate quality. Twenty-two were low and 19 very low. The cardiovascular mortality finding — the strongest in the review by credibility class — was itself graded very low quality. Several of the review’s authors are affiliated with a research centre that has received funding from food companies, disclosed at length in the paper.

Individual cohorts

Srour and colleagues followed 105,159 adults in the French NutriNet-Sante cohort for a median of 5.2 years (The BMJ, 2019), with 1,409 cardiovascular events. Each absolute 10-point increase in the ultra-processed share of the diet was associated with a hazard ratio of 1.12 (95% CI 1.05 to 1.20) for overall cardiovascular disease, 1.13 (95% CI 1.02 to 1.24) for coronary heart disease and 1.11 (95% CI 1.01 to 1.21) for cerebrovascular disease. The associations survived adjustment for saturated fat, sodium, sugar, fibre and overall dietary pattern. The authors stated plainly that causality remains to be established.

The most instructive cohort result comes from EPIC. Cordova and colleagues tracked 266,666 adults across seven European countries for a median 11.2 years (The Lancet Regional Health – Europe, 2023); 4,461 developed multimorbidity of cancer and cardiometabolic disease. Total ultra-processed intake carried a hazard ratio of 1.09 (95% CI 1.05 to 1.12) per standard deviation. Broken into subgroups, though, the risk was concentrated: animal-based products 1.09 (95% CI 1.05 to 1.12) and sweetened beverages 1.09 (95% CI 1.06 to 1.12), while ultra-processed breads and cereals (0.97, 95% CI 0.94 to 1.00) and plant-based alternatives (0.97, 95% CI 0.91 to 1.02) showed no association at all.

NOVA, and why nutrition scientists argue about it

Every study above uses the NOVA system, which sorts foods by degree and purpose of industrial processing rather than nutrient content: unprocessed or minimally processed foods; processed culinary ingredients such as oil and sugar; processed foods; and ultra-processed foods, meaning industrial formulations typically containing additives and ingredients rarely found in a home kitchen.

That last group is enormous, and the arguments against it are serious. In a formal 2022 debate in the American Journal of Clinical Nutrition — with the unusual twist that the same two researchers wrote both sides — the case against NOVA held that it rests on poorly defined processes and on the presence of additives from a chemically heterogeneous group, which makes misclassification easy; that the claim of processing-driven overconsumption has little support beyond what nutrient composition, energy density and food matrix already explain; and that observational studies have not demonstrated an effect independent of those factors. The case for NOVA held that the association with poor diet quality and chronic disease is consistent enough to act on now, while mechanisms are worked out.

The practical consequence is that the label captures wholemeal supermarket bread, many plain yogurts, fortified plant milks and infant formula alongside confectionery and soft drinks. The EPIC subgroup results are the clearest sign that this breadth matters.

Candidate mechanisms

If ultra-processed diets drive overeating, the leading explanations are about physical form and eating behaviour rather than any single ingredient:

• Eating rate and texture. In a crossover study of 50 healthy-weight adults (American Journal of Clinical Nutrition, 2022), harder-textured lunches were eaten more slowly and produced 21% less food weight and 26% less energy consumed. Intake ranged from 483 kcal (95% CI 432 to 531) for the hard minimally processed meal to 789 kcal (95% CI 726 to 853) for the soft ultra-processed one — around 300 kcal from a single lunch.
• Energy density. More calories per gram means more calories consumed before the physical volume of a meal registers.
• Hyper-palatability. Combinations of fat, sugar and salt that rarely occur together in whole foods appear to weaken the normal signals that end a meal.
• Reduced satiety per calorie. Soft, low-fibre, quickly eaten foods deliver less chewing time and less gastric stretch for the same energy.
• Displacement. A diet dominated by these foods contains less of everything else — fibre, legumes, vegetables, whole grains.
• Additives and packaging. Emulsifiers, non-nutritive sweeteners and materials migrating from packaging are hypothesised contributors. In humans, at dietary intakes, this remains a hypothesis rather than a demonstrated cause of disease.

Limitations: what this evidence does not establish

Popular coverage tends to collapse all of the above into a claim that ultra-processed food causes chronic disease. The research does not support that in its strong form.

• The hard outcomes are observational. No trial has randomised anyone to years of one diet and counted heart attacks. High ultra-processed intake also tracks with smoking, lower income, less physical activity and lower overall diet quality, so residual confounding is a live explanation, as the cohort authors themselves note.
• The trials are small and short. Twenty inpatients for four weeks, and 50 in the intention-to-treat sample over 16 weeks. Both measured intake and body weight, not disease.
• Evidence quality is mostly low. Only four of 45 pooled analyses in the BMJ umbrella review reached moderate GRADE quality; the rest were low or very low.
• Processing has not been isolated. No study has separated processing itself from energy density, texture and palatability — and the texture experiment suggests those may account for much of the effect.
• Exposure measurement is imprecise. Most cohorts estimate intake from food-frequency questionnaires, and different research groups classify the same product differently under NOVA.
• No threshold is known. Nothing in this literature identifies a safe percentage of the diet, and no trial has tested reducing intake to a target.
• Incentives cut both ways. Food-industry funding is a real problem in nutrition research; so is the momentum behind a popular framework that its own proponents concede lacks mechanistic proof.

Where this leaves a reasonable reader

Two randomised feeding trials point the same direction, and the cohort evidence, imperfect as it is, is consistent across countries and outcomes — enough to justify shifting the balance of a diet without pretending the mechanism is settled.

• Treat the categories the subgroup data actually implicate — sugary and artificially sweetened drinks, and ultra-processed animal-based products — differently from the ones that showed no association, such as wholemeal bread and plant-based alternatives.
• Pay attention to texture and eating speed, not just ingredient lists. Firmer, chewier, higher-fibre meals reduced intake substantially in controlled testing.
• Watch energy density, which is measurable on a label, rather than trying to classify every product by NOVA group.
• Remember that the Nature Medicine trial’s ultra-processed arm still produced weight loss when it followed national nutrition guidance. Meeting basic dietary guidelines is not made pointless by processing.

This article is educational and is not medical advice; anyone managing a specific condition should discuss dietary changes with their own clinician. The research is also moving quickly — the second major feeding trial only appeared in 2025, and the mechanistic work the BMJ review called urgent has not been done yet.

Sources

• Hall KD, 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. PubMed
• Dicken SJ, et al. Ultraprocessed or minimally processed diets following healthy dietary guidelines on weight and cardiometabolic health: a randomized, crossover trial. Nature Medicine. 2025. PubMed
• Lane MM, et al. Ultra-processed food exposure and adverse health outcomes: umbrella review of epidemiological meta-analyses. BMJ. 2024;384:e077310. PubMed
• Srour B, et al. Ultra-processed food intake and risk of cardiovascular disease: prospective cohort study (NutriNet-Sante). BMJ. 2019;365:l1451. PubMed
• Cordova R, et al. Consumption of ultra-processed foods and risk of multimorbidity of cancer and cardiometabolic diseases: a multinational cohort study. The Lancet Regional Health – Europe. 2023;35:100771. PubMed
• Monteiro CA, Astrup A. Does the concept of ultra-processed foods help inform dietary guidelines, beyond conventional classification systems? YES / NO. American Journal of Clinical Nutrition. 2022;116(6):1476-1488. YES | NO
• Teo PS, et al. Texture-based differences in eating rate influence energy intake for minimally processed and ultra-processed meals. American Journal of Clinical Nutrition. 2022;116(1):244-254. PubMed