Ultra-processed food: what the evidence actually says.

Few terms in nutrition have travelled as far in as little time as ultra-processed food. A decade ago the phrase was confined to a handful of Brazilian epidemiologists. Today it is in policy documents, school lunch debates and supermarket marketing. Some of that ascent reflects real evidence. Some of it has clearly run ahead of the data. This piece tries to separate the two.

§ Where the term comes from

The category was defined in 2009 by Carlos Monteiro and colleagues at the University of São Paulo as part of the NOVA classification system. NOVA sorts foods not by nutrient content but by extent and purpose of industrial processing. Group 1 is unprocessed or minimally processed foods. Group 2 is culinary ingredients like oils and salt. Group 3 is processed foods such as canned beans or tinned fish. Group 4 — ultra-processed foods — covers industrial formulations typically containing additives, isolated ingredients and components not found in domestic kitchens, designed to be hyper-palatable, shelf-stable and convenient.

It is important to be precise about what this does and does not mean. Tinned tomatoes are not ultra-processed. Wholegrain pasta is not ultra-processed. A flavoured strawberry yoghurt with stabilisers, modified starch and added sweeteners is. The definition is structural, not moral. Whether that structural definition tracks reliably with health outcomes is the empirical question.

§ The trial that changed the conversation

The single most important piece of UPF evidence is a small, careful trial conducted by Kevin Hall and colleagues at the US National Institutes of Health, published in Cell Metabolism in 2019. Twenty adults lived in a metabolic ward for four weeks. For two weeks they ate a diet built from ultra-processed foods; for the other two weeks they ate an unprocessed-food diet. Crucially, the two diets were matched for calories presented, macronutrients, sugar, fibre and sodium. Participants were told to eat as much or as little as they wanted.

On the UPF diet, participants spontaneously ate roughly 500 more calories per day and gained, on average, about a kilogram of body weight. On the unprocessed diet, they lost roughly the same. The macronutrient matching matters: this is not a story about hidden sugar or salt. Something about the form of the food drove people to eat more of it.

Hall has been careful, since, to point out that one twenty-person crossover trial does not settle a policy question. But it does establish causation in a controlled setting, which is exactly what the field had previously lacked.

"Macronutrients matched. Same calories on the table. People ate 500 more. The processing itself, not the nutrient profile, drove the gap."

§ What the cohort data shows

Outside the metabolic ward, the picture comes from large prospective cohort studies. The French NutriNet-Santé cohort, tracking more than 100,000 adults, found that each 10% increase in the proportion of ultra-processed food in the diet was associated with a roughly 12% higher risk of cardiovascular disease, with similar dose-response signals for type 2 diabetes and all-cause mortality. The 2024 BMJ umbrella review by Lane and colleagues, which pooled 45 meta-analyses across nearly ten million participants, found consistent associations between high UPF consumption and a range of cardiometabolic and mental-health outcomes.

Cohort studies cannot prove causation. People who eat more UPFs differ in many other ways. But the consistency of the signal, the dose-response pattern, the biological plausibility now anchored by the Hall trial, and the convergence across populations make this one of the better-supported associations in modern nutritional epidemiology.

§ Why the category leaks

The clearest critique of the UPF framework is that the category is too broad to be clinically useful. A plain industrial wholegrain bread, a packaged unsweetened soy milk and a can of cola are all NOVA Group 4. They are obviously not equivalent in their effect on the body.

Recent work, including a 2024 review by Dicken and Batterham, has begun unpicking this. Within the UPF category, sugar-sweetened beverages, processed meats, and refined carbohydrate snacks show the strongest associations with adverse outcomes. Packaged wholegrains, plain yoghurts and certain plant-based meat alternatives show much weaker associations, and in some analyses none at all. The category, in other words, is a useful flag and a poor scalpel.

§ What is probably driving the effect

Several mechanisms are being investigated. None is yet settled, and the truth is likely to be a combination.

• →Hyper-palatability and softness. Industrial formulations are engineered to be easy to eat quickly. Faster eating rates are reliably associated with greater intake before satiety signals catch up.
• →Energy density. UPFs tend to deliver more calories per gram, displacing lower-density whole foods from the same portion size.
• →Disruption of the gut–brain satiety axis. Emerging work suggests certain emulsifiers and additives may alter gut hormones such as GLP-1 and PYY, though this remains preliminary.
• →Displacement. The more of the diet that is UPF, the less room there is for fibre-rich whole foods, with downstream effects on the microbiome and blood-glucose response.

§ What this means in your kitchen

The trial and cohort evidence is strong enough to act on, and weak enough that the action should be proportionate. A useful rule of thumb, drawn from the data rather than from a slogan:

• →Cook most of what you eat. Whole or minimally processed foods, prepared at home or in a workplace canteen, should make up roughly four-fifths of the week.
• →Treat the highest-risk UPFs — sugar-sweetened drinks, processed meats, confectionery, refined snack foods — as occasional rather than daily.
• →Stop worrying about the lower-risk UPFs. Plain wholegrain bread, unsweetened yoghurt, frozen vegetables, canned beans and tinned fish are fine, often excellent, and trying to eliminate them produces stress with no measurable benefit.
• →Read the ingredients list, not the front of the pack. The presence of ingredients you would not find in a domestic kitchen — modified starches, hydrogenated oils, isolated protein fractions, multiple sweeteners — is the practical signal that you are in NOVA Group 4 territory.

§ What still has to be proven

Honest reporting of this area requires acknowledging what is unsettled. Causal mechanisms beyond Hall's calorie-overconsumption finding are not yet pinned down. The risk gradient within the UPF category needs more granular work. Long-term randomised trials at population scale are, for obvious practical reasons, unlikely ever to exist.

What we have is a careful controlled-feeding trial, a large and consistent cohort literature, and plausible mechanisms. That is enough to act on. It is not enough to panic on. The framing that serves you best is neither dismissive nor catastrophising. It is closer to: this is one of the few modifiable, large-effect levers in your diet, and the cost of taking it seriously is small. The cost of ignoring it, slowly compounded, is not.