What Does “Harm” Mean When the Evidence Is Mixed?
Ultra-processed foods are the subject of this series, but the useful question is not simply whether they are harmful. The harder and more informative question is what happens when people consume dietary patterns containing more foods classified as ultra-processed: are those patterns associated with weight-related outcomes, do controlled experiments show a causal effect, what mechanisms might explain it, and how far should the findings influence personal or public-health decisions? The aim is to answer those questions without treating a processing label as a verdict.
The debate matters because ultra-processed foods are no longer a marginal part of the food supply. They are common in modern diets, and many are convenient, inexpensive, durable, and engineered to be easy to eat. The category includes products as different as sweetened drinks, packaged snacks, processed meats, industrial breads, ready-to-eat meals, and some flavored dairy products. A pattern containing more of these foods may also contain more sugar, salt, refined starch, saturated fat, or total energy—and may replace foods with different textures, nutrients, and preparation demands. That overlap is exactly why the health question is important and difficult.
Two recent systematic reviews provide a useful starting point. A review prepared for the 2025 Dietary Guidelines Advisory Committee examined dietary patterns with varying amounts of foods classified as ultra-processed across life stages. It found a relationship between higher-UPF patterns and greater adiposity or overweight risk in children and adolescents and in adults and older adults, but graded both conclusions as Limited. A second review searched specifically for randomized trials of changing UPF consumption. It found only four eligible studies, involving 455 participants, and reported no significant effect for 30 of 42 evaluated outcomes. Together, these reviews offer neither a clean dismissal nor a universal causal verdict. They show a recurring signal, one important short-term experimental result, and substantial uncertainty about what the category itself contributes.
The word “harm” hides four different questions
The first question is about association. If a study follows people over time and finds that those who consume more UPF-containing foods tend to gain more body fat or develop overweight, it has identified a relationship between an exposure and an outcome. That relationship can be meaningful even when it is not causal. People who eat more packaged or ready-to-eat products may differ in activity, income, education, baseline health, sleep, smoking, or access to fresh foods. Researchers can adjust for some of these factors, but adjustment does not guarantee that every relevant difference was measured accurately or that the statistical model captured it fully.
The second question is about causation under an intervention. A randomized trial assigns people to different dietary conditions, reducing the chance that pre-existing differences explain the result. But the trial still identifies the effect of the intervention that was actually delivered. If the two diets differ in texture, energy density, added sugar, fat, fiber form, eating speed, and convenience as well as processing classification, a result cannot automatically be attributed to processing alone. Randomization makes the comparison stronger; it does not make the exposure concept simpler than it is.
The third question is about mechanism. A proposed mechanism might connect a softer texture to faster eating, weaker satiety signals, higher energy intake, and eventually weight gain. Another might emphasize nutrient composition, displacement of minimally processed foods, additives, processing-generated compounds, the gut environment, or packaging. These possibilities can make a finding biologically plausible. They do not, by themselves, show that the pathway operates in every UPF, dominates other pathways, or produces long-term disease.
The fourth question is about action. Even when evidence is incomplete, people and institutions must decide what to eat, what to recommend, and which policies to test. An action can be reasonable without being based on a precisely quantified causal risk. But the strength of the action and the confidence of the language should match the evidence. “There is enough reason to favor feasible improvements in diet quality” is a different statement from “this processing category causes a known percentage increase in lifetime disease risk.”
These questions are connected, but they are not interchangeable. Confusing them produces the two familiar extremes: dismissing all observational evidence because it is not an experiment, or treating a repeated association and a plausible mechanism as proof of a category-wide harm.
The map is deliberately sequential. Before asking whether a result is alarming, we need to know what counted as ultra-processed, what it was compared with, which outcome was measured, and whether the design can support the conclusion being drawn. A headline usually compresses all of those steps into one word: “harm.” Good evidence reading puts the steps back.
What these reviews can actually tell us
The USDA review’s most developed conclusion concerns adiposity—measurable body fat and related indicators such as BMI, waist circumference, and fat mass—and the risk of overweight or obesity. In children and adolescents, the review considered a substantial body of prospective cohort studies. The direction of findings was generally similar: dietary patterns higher in foods classified as UPF tended to be associated with greater adiposity or overweight risk. In adults and older adults, the review considered 16 articles, 15 prospective cohorts and one randomized trial, and reached a similar conclusion. Yet both conclusions received a Limited grade.
That grade is not a way of saying that nothing was found. It says the body of evidence is not strong enough to make the conclusion stable. The USDA assessment identified concerns about potential confounding, exposure misclassification, missing data, indirectness, and applicability. Effect sizes varied, and some confidence intervals included the null. Many studies used one dietary assessment, and food-frequency questionnaires were not always designed to classify foods by processing level. Some studies examined dietary patterns that contained many foods likely to be ultra-processed without having been designed specifically to measure variation in UPF. The review also did not pool the results into one summary estimate.
The life-stage boundaries are equally important. The USDA committee could not draw a conclusion for infants and young children up to 24 months because the five included prospective studies raised substantial concerns about consistency and directness. It found too little evidence to assess pregnancy, represented by one prospective cohort, or postpartum weight change, represented by two. “No conclusion” here means that the evidence was insufficient or too limited to answer the question. It does not mean that UPFs were shown to be harmless in those groups. Nor can evidence from adults with stable weight, or from children already carrying overweight or obesity, be transferred automatically to every population.
The randomized-trial review asks a different question and therefore changes the kind of confidence we can have. It found one small controlled-feeding trial in which weight-stable adults received an ultra-processed diet and an unprocessed diet in randomized two-week periods. Participants ate as much as they wanted while researchers supplied the meals. During the ultra-processed period, they consumed more energy and gained weight; during the unprocessed period, they lost weight. That result is important because it shows that a highly controlled contrast between two diet packages can change short-term intake and body weight.
But the trial does not supply a lifetime risk estimate, and it does not prove that every ultra-processed product has the same effect. The diets differed in more than the label used to classify them, and the study was small and brief. The review notes concerns about the lack of blinding, the absence of an adaptation or elimination period, and differences in features such as added sugar, insoluble fiber, saturated fat, and the form of fiber used. The result is best read as a strong clue about a short-term dietary effect under those conditions, not as a complete explanation of the broader observational pattern.
The other randomized studies make the picture less tidy. Three used counseling or educational interventions rather than supplying all food. In the pregnancy study, the intervention did not actually reduce UPF consumption. In the study of women with obesity, some measures improved in the intervention group, but the analysis did not provide a clean between-group comparison, and UPF intake also fell in the control group. In children and adolescents with overweight or obesity, the intervention did not produce significant differences in the reported BMI, glucose, insulin, lipid, uric-acid, or fat-mass outcomes. High or differential dropout and additional recommendations about diet and physical activity made attribution harder. Across the randomized review, most evaluated outcomes were not significantly different.
This is why “observational versus randomized” is too simple a summary. The cohort evidence tells us that a pattern recurs across many populations, while the controlled-feeding trial shows that one carefully constructed diet contrast can alter intake and weight quickly. The educational trials show how difficult it is to reduce a complex exposure in ordinary life and then identify its independent effect. Each design illuminates a different part of the problem.
A more useful way to read the debate
The most defensible opening judgment is therefore layered. Higher-UPF dietary patterns are associated with greater adiposity and overweight or obesity risk in the children, adolescents, adults, and older adults covered by the USDA conclusions. The evidence supporting those associations is Limited, not strong. A small controlled-feeding study provides experimental support for a short-term increase in energy intake and weight under a particular diet contrast. The randomized evidence as a whole remains too sparse and methodologically limited to establish a definitive causal link between the degree of food processing and adverse health outcomes across the category.
That judgment leaves room for concern without pretending that uncertainty has disappeared. It also keeps the subject in view. The question is not whether people should care about what they eat; the evidence does not justify that dismissive conclusion. The question is what the processing category adds beyond the nutrient composition, energy density, texture, eating behavior, food access, and social context that travel with it. The supplied reviews do not settle that question, and they do not support a universal threshold at which UPF becomes harmful.
For a reader encountering a new claim, five checks are more useful than a confident slogan:
- What exactly was classified as ultra-processed? Look for the foods, the classification system, and the amount or percentage being compared.
- What was the comparator? A low-UPF pattern, a particular unprocessed version of the same food, usual intake, or a multi-component program are not equivalent comparisons.
- What outcome was measured, and for how long? Short-term energy intake, body weight, body fat, chronic disease, and mortality answer different questions.
- What can the design identify? A cohort can show a relationship; a controlled trial can test a delivered diet contrast; neither automatically isolates every proposed mechanism.
- What would change the conclusion? Better repeated measurements, larger and longer trials, matched energy density, validated classification, and more diverse populations would make the answer more precise.
Those checks do not make the debate less consequential. They make the consequences easier to judge. The next step is to look closely at the category itself, because the meaning of any UPF estimate depends on which very different foods have been placed inside it and which features the classification leaves entangled.