The Harder Test: What Randomized Trials Actually Show

The randomized evidence on ultra-processed foods asks a more demanding question than a cohort study can answer: if researchers change the diet, does a health-related outcome change too? This chapter focuses on the small controlled-feeding trial that provides the clearest experimental signal, then compares it with the other randomized studies in the review. The practical lesson is precise: randomization can strengthen a causal inference, but only for the dietary contrast that was actually delivered and measured.

A randomized trial starts by assigning people to different conditions rather than waiting for them to choose their own diets. That assignment helps balance both known and unknown differences between groups, at least in expectation. A controlled-feeding trial goes further by supplying the food, reducing the variation introduced by shopping, cooking, portioning, and reporting. These features make the design powerful for testing short-term effects. They also make it expensive, artificial, and difficult to extend over years.

What the controlled-feeding trial tested

The trial included 20 adults whose weight was stable. Each participant received an ultra-processed diet and an unprocessed diet in a randomized crossover sequence, with each diet period lasting two weeks. Researchers provided three main meals and snacks, and participants were told to eat as much as they wanted during the meal periods. Because the same people experienced both conditions, the comparison did not depend only on differences between two unrelated groups. Each person served as a partial control for themselves.

The contrast was deliberately large. During one period, 81.3% of dietary energy came from foods classified as ultra-processed; during the other, the share was 0%. The diets were designed to be similar in several broad respects, including calories offered, energy density, macronutrient distribution, sugar, sodium, and total fiber. That design makes the result more informative than a simple comparison between people who buy different foods in daily life. It does not make the diets identical in every relevant feature.

Participants consumed about 508 additional kilocalories per day during the ultra-processed period. They gained about 0.9 kilograms during that period, while losing about 0.9 kilograms during the unprocessed period. Fat mass increased by about 0.4 kilograms during the ultra-processed period; no significant change in fat mass was observed during the unprocessed period. Carbohydrate and fat intake were higher on the ultra-processed diet, while protein intake did not differ significantly.

The trial also examined metabolic outcomes. The review reported no significant differences at the end of the two periods in glucose, insulin, or glycated hemoglobin. Total cholesterol and HDL cholesterol were lower after the unprocessed period, while LDL cholesterol and triglycerides did not differ. In the review’s GRADE assessment, body weight and total and HDL cholesterol received moderate certainty, while energy, carbohydrate, and fat intake received low certainty. Those labels describe the certainty of the evidence for the reported outcomes; they do not certify a universal effect of every food classified as UPF.