MAHA’s food-policy arguments often treat ultra-processed foods as a contributor to chronic disease. The previous lesson established why an observed association does not, by itself, identify a cause. Here, the question becomes more specific: What happens when people are assigned different diets, and how far can those findings be extended to diseases that develop over years?
This lesson distinguishes three claims that are sometimes run together: ultra-processed diets may encourage overeating; sustained changes in eating may affect body weight; and ultra-processed food consumption may increase long-term disease risk. Evidence for one claim does not automatically prove the next.
What does “ultra-processed” describe?
The NOVA classification groups foods by the nature and purpose of their processing, rather than solely by their nutrient content. Its ultra-processed category generally describes industrial formulations containing ingredients or additives not ordinarily used in home cooking. That makes it a potentially useful way to study the food supply, but it is a broad category: two products can receive the same classification while differing substantially in nutrients, texture, and how people eat them.
Watch the short portions of this interview with the researcher who led the feeding trial discussed below. The first introduces the classification; the second explains the trial’s central comparison. Treat the interview as an explanation of the research, not a substitute for the published study.
World-Leading NIH Metabolic Scientist: Why You Eat 500 More Calories a Day Without Knowing It
In “World-Leading NIH Metabolic Scientist: Why You Eat 500 More Calories a Day Without Knowing It,” The Proof with Simon Hill interviews researcher Kevin Hall about ultra-processed foods and his randomized feeding trial.
Watch the classification for the NOVA categories and the difficulty of drawing a single boundary around ultra-processed foods. Then watch the feeding trial, focusing on what was assigned to participants and what they were free to choose.
The distinction between food supplied and food consumed is crucial. Researchers can offer two diets with similar calories and nutrient amounts while allowing participants to eat as much or as little as they wish. If intake then differs, that difference is a result to investigate—not a contradiction in the study design.
What the randomized feeding trial establishes
In the published trial, 20 adults lived in an inpatient research unit for 28 days. Each ate an ultra-processed diet for two weeks and an unprocessed diet for two weeks; random assignment determined which came first. This crossover design let researchers compare each person’s intake under both diets, reducing the influence of stable differences between people. Meals and snacks were plentiful, and participants chose how much to eat.

Read the methods and selected findings in Hall and colleagues’ paper. Pay particular attention to what the researchers matched in the presented meals, and to the differences that remained.
Read the trial report by Hall and colleagues to separate its strong experimental finding from the questions its design could not settle.
In “Results and Discussion,” begin with the design and intake result, ending after the opening paragraph under “Food Intake.” In the same section, read from the energy-density comparison; then read “Appetitive measurements and eating rate” and “Body weight and composition,” from the appetite measures through weight change. Finally, read the entire “Limitations of Study” subsection, beginning with cost and convenience and ending with the comparison to shorter laboratory studies. Focus on which features of the diets were controlled, which were not, and what the inpatient setting leaves unresolved.
Participants consumed an average of about 508 more kilocalories per day during the ultra-processed phase. They gained about 0.9 kg during that phase and lost about 0.9 kg during the unprocessed phase. The graph shows the direction of these short-term differences.

Random assignment makes the trial strong evidence that these assigned diets caused different short-term intake and weight outcomes under these conditions. It does not show that the diets contained the same consumed calories. Their provided calories and several nutritional features were designed to be similar; people then ate different amounts.
Nor does matching make “processing itself” the isolated cause. The diets still differed in such respects as added sugar as a share of total sugar, types of fiber, and saturated fat as a share of total fat. The researchers also observed faster eating and greater energy density in non-beverage foods during the ultra-processed phase. Those are possible explanations for greater intake, but this trial did not randomly assign eating speed or food texture independently. It therefore cannot determine which feature—or combination of features—produced the effect.
Finally, the trial was small and short, had no washout period between diets, and provided meals at no cost to participants. Its control over food and measurement is a strength for detecting short-term effects; the same setting cannot tell us how people would shop, cook, substitute foods, or maintain a diet over years. Two weeks of weight change is not a measured change in long-term diabetes, cardiovascular disease, or cancer incidence.
What longer observational studies add
To study diseases that take years to develop, researchers often follow large groups of people, record what they report eating, and compare subsequent health outcomes. Reviews have reported associations between higher ultra-processed food intake and outcomes including type 2 diabetes and cardiovascular disease. Such studies reach populations and time spans that an inpatient feeding trial generally cannot.
A cohort can establish that a recorded dietary exposure preceded a recorded diagnosis. Researchers can also compare groups after adjusting for measured factors such as smoking or physical activity. Yet the previous lesson’s cautions remain: people with different diets may differ in unmeasured ways, dietary reports may misclassify intake, and people may change their diets as health problems emerge. Even a carefully adjusted association does not identify a particular processing step or additive as the cause.
This short news interview offers a concrete example of both the value and the limits of a prospective cohort. It concerns depression, not the metabolic diseases measured in many food-policy arguments; use it to inspect the study design, not to transfer its reported risk estimate to another condition.
Researcher explains connection between ultra-processed foods and depression
PBS NewsHour’s “Researcher explains connection between ultra-processed foods and depression” reports on a long-running observational study and interviews a researcher about its methods and limits.
Watch the reported association, noting the population studied. Then watch the comparison for how researchers addressed timing and some alternative explanations, followed by the population limit for what the sample cannot establish about other groups. The brief headline description of risk should not be read as an individual’s absolute probability of developing depression.
The study’s size and follow-up help address questions a two-week trial cannot. Starting with participants who did not have the outcome also strengthens the time ordering. Adjustment for measured lifestyle factors is useful, but it cannot guarantee that all relevant differences have been removed. Nor can a finding in one study population automatically be generalized to everyone.
The distinction between relative and absolute risk matters when such findings appear in news coverage. A hypothetical increase from a 4% risk to a 6% risk is a 50% relative increase but a 2-percentage-point absolute increase. Neither number should be inferred for a particular person from the interview alone.
Reading the two kinds of evidence together
A systematic review of randomized trials helps put the prominent inpatient result in context. At the time of that review, the authors found one controlled feeding trial and three studies based on educational interventions. The latter did not simply assign and provide two distinct diets: advice often included other changes, and a recommendation to reduce ultra-processed foods did not always produce a clear difference in consumption. A trial can randomize advice while leaving uncertain whether the groups actually ate meaningfully different diets.
Read this systematic review for the gap between controlled feeding results and evidence about sustained dietary change or long-term disease.
Before “Materials and methods,” read the observational-evidence rationale. In “3.4 Certainty of evidence” and “4 Discussion,” begin at the four-trial comparison. Later in “4 Discussion,” read the feeding-trial trade-off. In the discussion of NOVA’s scope, read the category distinction. Near the end of “4 Discussion,” read from short-term outcomes through triangulation. Focus on why randomization alone cannot rescue an intervention that does not change exposure, and why short-term measures cannot stand in automatically for disease outcomes.
The review also highlights a problem with treating the entire ultra-processed category as one exposure. If products within it differ, a population-wide association need not imply equal risk from every product. A useful next study might compare clearly specified foods or dietary patterns, measure what people actually consume, and examine whether effects persist when features such as nutrient profile or eating rate differ.
The most defensible assessment therefore combines, without conflating, the designs:
- Feeding trial: supports a causal effect of the tested diet packages on short-term calorie intake and weight under inpatient conditions.
- Prospective observational studies: can link reported intake to later disease outcomes across larger populations, but remain vulnerable to confounding and exposure-measurement error.
- Together: make a dietary contribution to long-term health worth serious investigation; they do not yet quantify a universal chronic-disease effect for all ultra-processed foods or establish that processing, rather than another shared feature of the tested diets, is the mechanism.
Evidence would become stronger if independently conducted studies reproduced the intake effect, distinguished among types of ultra-processed foods, and found consistent long-term associations after careful measurement and control of alternative explanations. Evidence against a broad claim would include reliable studies showing that the association disappears when relevant diet and lifestyle differences are accounted for, or that purportedly similar foods have markedly different effects. Either result would sharpen the claim rather than settle every question about the food supply.
Conclusion
A randomized feeding trial answers a relatively narrow question well: in one controlled comparison, people ate more and gained weight during the ultra-processed diet phase. Observational research asks a broader, longer-term question and finds associations with disease, but cannot assign causation on its own. The unresolved bridge is which foods and features produce which effects over time, for whom, and under ordinary conditions.
The next lesson applies the same distinction between a specific causal claim and wider concern about health to the claim that vaccines cause autism.