Ultra-Processed Foods: What the Research Actually Says, and What It Doesn't

“Ultra-processed” has become the nutrition word of the decade. It is on front pages, in parliamentary inquiries and on the back of every packet that wants to tell you it is not one. The research behind it is real and some of it is alarming. But the term is also doing a lot of work it was not designed to do, and the gap between what the studies show and how they are reported has grown wide. This piece lays out the evidence by strength, so you can decide what to change.

What “ultra-processed” means

The term comes from the NOVA classification, developed by Carlos Monteiro’s group at the University of São Paulo. NOVA sorts food by the extent and purpose of industrial processing, not by its nutrients. It has four groups:

1. Unprocessed or minimally processed foods: fruit, vegetables, meat, eggs, milk, grains, legumes, nuts.
2. Processed culinary ingredients: oils, butter, sugar, salt.
3. Processed foods: group 1 foods with group 2 ingredients added, such as canned vegetables, cheese, freshly baked bread, tinned fish.
4. Ultra-processed foods: industrial formulations made mostly from substances extracted from foods or synthesised, typically with additives such as emulsifiers, flavourings, colours and sweeteners that you would not use in a home kitchen. Soft drinks, packaged snacks, reconstituted meat products, instant noodles, most mass-produced breads and breakfast cereals, and many “ready meals” sit here.

Group 4 is the one that draws the headlines. In the United States, analysis of national dietary survey data found ultra-processed foods supplied about 57% of adults’ calories in 2017 to 2018, up from around 54% in 2001. In the UK the share is similar.

The strongest evidence: the BMJ umbrella review

In February 2024, The BMJ published an umbrella review led by Melissa Lane at Deakin University. An umbrella review is a review of reviews. It pooled 45 separate meta-analyses, drawn from 14 review articles and representing just under 10 million participants, and graded the evidence for each health outcome using standard criteria for credibility.

The findings, by grade:

Convincing evidence (the highest grade):

• Cardiovascular disease-related death: roughly 50% higher risk with high versus low intake.
• Type 2 diabetes: about 12% higher risk.
• Anxiety and common mental disorders combined: 48% to 53% higher risk.

Highly suggestive evidence:

• All-cause mortality: about 21% higher.
• Heart disease-related death: about 66% higher.
• Obesity: about 55% higher.
• Depression: about 22% higher.
• Sleep problems: about 41% higher.
• Wheezing: about 40% higher.

The remaining outcomes, including various cancers, asthma, gastrointestinal conditions and cardiometabolic markers, had weaker or less consistent evidence.

The limitation that matters most

Almost everything in the umbrella review is observational. Researchers measured what people ate, followed them for years and compared outcomes. That design cannot prove that ultra-processed food caused the harm. People who eat a lot of it also tend to earn less, smoke more, exercise less and eat fewer vegetables, and no statistical adjustment removes those differences completely.

There is also a measurement problem. Most of the cohorts used food questionnaires designed before NOVA existed, so researchers had to retrofit the classification. A “sandwich” could be group 3 or group 4 depending on the bread.

This is why one small trial carries so much weight.

The one controlled trial

In 2019, Kevin Hall’s team at the US National Institutes of Health did something nobody had done: they admitted 20 adults to a research unit for four weeks and fed them. For two weeks the meals were ultra-processed; for the other two they were minimally processed. The order was randomised. Critically, the two diets were matched for calories offered, sugar, fat, fibre, sodium and macronutrients. Participants could eat as much or as little as they wanted.

On the ultra-processed diet, people ate about 500 more calories a day and gained an average of 0.9 kg over two weeks. On the minimally processed diet they lost about the same. The difference was not explained by the nutrition labels, because the labels were matched. Something about the food itself, whether texture, energy density, eating speed or palatability, drove people to eat more.

Twenty people for four weeks is a small study, and it was done in a controlled unit rather than in real life. But it is a true experiment, it was designed to rule out the obvious confounders, and it produced a clear effect. It is the reason most researchers now believe the observational link to obesity is at least partly causal.

Not all ultra-processed foods behave the same

This is the point most coverage misses. NOVA groups by processing, so it puts whole-grain breakfast cereal in the same box as a fizzy drink. When researchers break group 4 into sub-categories, the picture changes.

A 2024 analysis in The BMJ followed more than 110,000 US health professionals for over 30 years. Overall, the highest ultra-processed intake was linked to a 4% higher all-cause mortality rate, a much smaller figure than the headline numbers above. The risk was concentrated in specific sub-groups: ready-to-eat meat, poultry and seafood products, sugar-sweetened and artificially sweetened drinks, dairy-based desserts, and highly processed breakfast foods. Ultra-processed breads and cereals, by contrast, were not associated with higher mortality.

The same study found that overall diet quality predicted mortality better than the ultra-processed share did. In other words, eating a lot of vegetables, whole grains and legumes mattered more than whether your bread came in a plastic bag.

What to actually change

Put the two strands of evidence together and a reasonable set of priorities emerges.

Cut first: processed meats (sausages, deli meats, bacon-style products), sugar-sweetened drinks, and ready-to-eat sweet or salty snacks. These show the most consistent harm across both the umbrella review and the sub-group analyses, and they are easy to replace.

Worry less about: packaged whole-grain breads, plain yogurts, canned beans and vegetables, and plain breakfast cereals without added sugar. Some are technically group 4; the evidence does not single them out.

Think in shares, not items. The cohorts compared people eating roughly 20% of calories from ultra-processed food with people eating 70% or more. Moving your own share down is what the evidence supports. Hunting every emulsifier is not.

Cook when you can, and buy simply when you cannot. The Hall trial suggests the problem is partly how these foods are engineered to be eaten quickly and in volume. Meals you assemble yourself, even from frozen or canned components, do not have that property.

The honest summary

The evidence that a diet dominated by ultra-processed food is bad for the heart, metabolism and mood is strong and getting stronger. The evidence that every ultra-processed product is harmful is not there. The sharpest tool you have is not a classification system but a rough look at where your calories come from, and a shift toward food that needs a plate.