Ultra-processed food linked to higher rates of seven chronic conditions in 51 studies of 8.8 million adults

Ultra-processed food has become the nutrition story of the decade. Studies linking it to one illness or another now appear almost weekly, each with its own population, its own way of counting and its own headline percentage. For a reader trying to work out how worried to be, the pile is hard to add up.

A team led by Xiao Liu of Sun Yat-sen University in China has done the adding. Their meta-analysis, published in September in Family Medicine and Community Health, a journal in the BMJ group, pools 51 long-term studies covering 8.8 million adults and seven kinds of chronic illness at once. Earlier poolings mostly compared heavy eaters with light ones. This one, the first by its authors’ account to do both across so many outcomes, also estimates how risk changes with each extra portion. The answer points the same way for every disease examined, and the authors themselves grade most of it as weak evidence.

How is ultra-processed food defined?

The term comes from a classification called NOVA, devised by Carlos Monteiro and colleagues at the University of São Paulo, which groups foods according to the extent and purpose of industrial processing and not by their nutrients. Ultra-processed is its fourth and last group.

The summary that accompanied the new study describes these products as ready-to-eat or heat-and-serve products that have undergone extensive industrial processing: soft drinks, packaged snacks, reconstituted meats, instant noodles, most breakfast cereals. Monteiro’s group offers a label test. A product belongs in the category if its ingredient list includes food substances never or rarely used in kitchens, such as hydrogenated oils or high-fructose corn syrup, or additives whose job is to make it look and taste more appealing.

That test is simple to apply and sweeps widely. A supermarket wholegrain loaf with an emulsifier in it lands in the same group as a cola.

Seven conditions, all more common with ultra-processed food

The studies gathered here all followed adults forward in time, asked them what they ate and recorded who later fell ill. Of the cohorts, 16 studies were conducted in the Americas, 25 in Europe, eight in Asia, and two in Oceania, and follow-up ranged from 2 to 32 years.

Compared with people who ate the least, those who ate the most ultra-processed food had higher rates of everything the team looked at. Eating a lot of ultra processed foods was associated with higher risks of cardiovascular events (24%), cancer (12%), obesity/overweight (23%), metabolic syndrome/diabetes (24%), depression/anxiety (27%), digestive diseases (31%), high blood pressure (11%) and death from any cause (18%). Cardiovascular events here means heart attacks, strokes and deaths from heart disease. Metabolic syndrome is a cluster of raised blood pressure, blood sugar, blood fats and waist size.

“High” and “low” meant different things in different studies, so the team also put everyone on a common scale of grams per day, treating 50 grams as one serving. On that scale, each additional 100 g/day consumed was associated with greater risks of cardiovascular events (14%), cancer (4%), death from any cause (3%), and metabolic syndrome/diabetes (2%). One hundred grams is two of the study’s servings. It is a measure of weight, so a can of soda counts for several times as much as a bag of chips.

The risk rose steadily with the amount, with no level below which it disappeared. The authors write that a safe consumption level remains undefined.

The direction of these results is not new. An umbrella review, meaning a review of earlier reviews, appeared in The BMJ two years ago. Led by Melissa Lane at Deakin University in Australia, it collected every earlier meta-analysis it could find. The search identified 45 unique pooled analyses, and direct associations were found between exposure to ultra-processed foods and 32 (71%) health parameters, from heart disease to anxiety. The new study adds the dose estimates and a larger pool.

Does every ultra-processed food carry the same risk?

Almost certainly not, and the new paper says so. Because NOVA sorts by processing, the authors note, foods with markedly different nutritional profiles and potential health effects may be grouped within the same UPF category, UPF being the common abbreviation. They name wholegrain bread, unsweetened yogurt and nut products alongside sugary drinks, confectionery and processed meat, and conclude that not all foods classified as UPFs are likely to have equivalent health effects. Too few of the 51 studies split the category for the team to test this themselves.

Others have. Harvard researchers tracked diet and heart disease in three large groups of American health professionals and broke ultra-processed food into its component types. In their words, UPF groups demonstrated divergent associations. Sugar-/artificially-sweetened drinks and processed meats were associated with higher CVD risk, CVD being cardiovascular disease. Bread and cold cereals, yogurt and dairy desserts ran the other way, with lower risk among people who ate more of them.

A second Harvard analysis, of 74 563 women and 39 501 men followed for more than thirty years, looked at deaths. Overall the heaviest consumers had a 4% higher all cause mortality, a far smaller figure than the new pooled 18%, and no associations were found for cancer or cardiovascular mortality. The product group that stood out was the same: meat/poultry/seafood based ready-to-eat products (for example, processed meat) consistently showed strong associations with mortality outcomes.

Put those together and the pooled percentages look like an average over foods with very different records. The Chinese team makes the same point about its own numbers, saying they should be interpreted as average associations across populations and methods. A reader who eats supermarket bread and plain yogurt and a reader who lives on soda and hot dogs are both “high consumers” in this literature.

What trials of ultra-processed diets have shown

Cohort studies cannot assign people a diet, so they cannot fully separate the food from the people who eat it. Two feeding trials have tried.

At the US National Institutes of Health, Kevin Hall’s team kept 20 weight-stable adults on a research ward for a month and fed them two weeks of ultra-processed meals and two weeks of unprocessed ones, in random order. Meals were designed to be matched for presented calories, energy density, macronutrients, sugar, sodium, and fiber, and people could eat as much as they liked. On the ultra-processed diet they ate about 500 calories a day more and gained close to a kilogram. On the unprocessed diet they lost the same amount.

A British trial tested a harder question: does processing matter when the diet is already a healthy one? Samuel Dicken and colleagues at University College London noted that no trial has assessed its health impact within the context of national dietary guidelines, then gave 55 adults in England two eight-week diets that both followed official advice, one built from minimally processed food and one from ultra-processed products. People lost weight on both, and about twice as much on the minimally processed diet.

Both trials measured eating and weight over weeks. Neither can say anything about heart attacks or cancer. What they add is a plausible route from these foods to one of the outcomes, weight gain, that does not depend on who chooses to eat them.

Why the authors rate the ultra-processed food evidence as low certainty

The authors grade their own evidence, and the grades are modest. Using the standard system for rating medical evidence, they report low certainty for cardiovascular events, cancer, diabetes, depression, digestive disease and death, moderate certainty for obesity/overweight and very low certainty for hypertension. Low certainty means the true effect may be substantially different from the estimate.

Several weaknesses lie behind those grades. All included studies relied on self-reported instruments, mostly food questionnaires that were not designed to tell a factory-made cake from a homemade one. The studies disagreed with each other considerably on the size of the effect. They did not all allow for the same outside influences. Age was the most commonly adjusted confounder, meaning a factor that could itself explain the link, whereas adjustment for other confounders varied across studies. People who eat a great deal of packaged food also tend to have less money, less time and less healthy habits in general, and the paper concedes that residual confounding also remains possible because all included studies were observational. Residual confounding is the influence of differences between people that the analysis did not fully remove.

The team could not check whether studies with unexciting results had gone unpublished, because each disease had too few studies for the usual test.

For guidance, the paper stays general. “Given that UPF related chronic diseases often cluster as multimorbidity, reducing consumption of UPFs and replacing them with minimally processed alternatives may be relevant for chronic disease prevention and primary care management,” the authors conclude. That is a statement about populations. How it applies to one person’s diet depends on what the ultra-processed part of it consists of, and on their health, which their doctor or dietitian assesses.

The link between ultra-processed food and chronic disease is broad and consistent, and by its authors’ own grading it is weakly supported.