Key takeaway
In 20 adults living in a hospital research unit for 28 days, an ultra-processed menu led them to eat 508 kcal/day more (± 106 SE, P = 0.0001) and gain 0.9 kg, while a menu of unprocessed foods led to a 0.9 kg loss, each over 2 weeks and with eating left to appetite. The study was small, short and fully provisioned, so it cannot show what happens in daily life or which feature of the foods drove the extra eating.
Study design
Randomised trial
Reading basis
Full text
Reading focus
Comparator & uncertainty
The research question
Diet advice from rival camps (low-carbohydrate, low-fat, plant-based and others) often shares one line: avoid ultra-processed foods. These are industrial formulations with little whole food, placed in the fourth group of the NOVA system, which sorts foods by the nature, extent and purpose of their processing. They have become more common worldwide, now make up the majority of calories eaten in America, and have been associated with poor health outcomes, including death. The authors state, however, that no randomised controlled trial had shown a benefit from cutting them or a harm from adding them, so cause and effect was unproven. This trial tested whether an ultra-processed menu and an unprocessed menu, matched on the calories, sugar, fat, sodium, fibre and macronutrients offered, change how much people eat and how their weight moves.
Who and how
This was a randomised crossover trial run on the Metabolic Clinical Research Unit at the National Institutes of Health (NIH) Clinical Center in the United States. It enrolled 10 men and 10 women, aged 18 to 50 years (mean 31.2, SE 1.6), with a body mass index above 18.5 (mean 27, SE 1.5) and stable weight over the previous 6 months. People with diabetes, anaemia, cancer, thyroid disease, eating disorders or other psychiatric conditions, food allergies or special diets were excluded. Everyone lived on the unit for 28 continuous days and was randomly assigned to an ultra-processed or an unprocessed menu for 2 weeks, followed immediately by the other, with no run-in or washout. Three meals a day came from 7-day rotating menus, with snacks and water available all day. Meals plus snacks supplied 81.3% of energy from ultra-processed foods on one menu and 88.1% from unprocessed foods on the other, and food was offered at twice each person's estimated energy needs. Participants could eat as much or as little as they wanted, for up to 60 minutes per meal. They were told this was not a weight-loss study, were blinded to daily weight and glucose readings, and cycled for exercise daily in three 20-minute bouts at 30 to 40% of heart rate reserve.
The primary endpoint was mean ad libitum (eat-to-appetite) energy intake over each 14-day period, calculated from weighed leftovers with nutrition software. Weight was measured at 6 am after the first void, with the 14-day change estimated by linear regression, and body composition was measured weekly by dual-energy X-ray absorptiometry (DXA). Liver fat was measured by magnetic resonance spectroscopy in 13 participants. Energy expenditure came from a respiratory chamber one day a week and from doubly labeled water, activity from a hip accelerometer, glucose from a continuous glucose monitor and an oral glucose tolerance test, and appetite from 100-point visual analogue scales. Eating rate was intake divided by meal duration. The study was powered to detect an intake difference of 125 to 150 kcal/day in 20 people (power 0.8, type I error 0.05). Analysis used analysis of variance with two-sided t-tests and significance set at P < 0.05; values are least-squares means ± standard error (SE).
What it found
- 1With food left to appetite, people ate 508 ± 106 kcal/day more on the ultra-processed menu (P = 0.0001). Neither the order of the diets (P = 0.64) nor sex (P = 0.28) changed this, and in the final week of each period the gap was still 459 ± 105 kcal/day (P = 0.0003). Intake on the ultra-processed menu fell significantly over the 2 weeks (P < 0.0001), while intake on the unprocessed menu did not (P = 0.23).
- 2The extra energy came from carbohydrate and fat, not protein: carbohydrate rose by 280 ± 54 kcal/day (P < 0.0001), fat by 230 ± 53 kcal/day (P = 0.0004) and protein was unchanged (-2 ± 12 kcal/day; P = 0.85). People ate more at breakfast (144 ± 39 kcal/day), lunch (248 ± 39) and dinner (108 ± 41). The offered protein share was slightly lower on the ultra-processed menu (14% versus 15.6% of calories), which the authors say could explain at most about 50% of the extra intake.
- 3Weight followed intake. Participants gained 0.9 ± 0.3 kg (P = 0.009) on the ultra-processed menu and lost 0.9 ± 0.3 kg (P = 0.007) on the unprocessed menu. Individual weight differences between the diets tracked individual intake differences (r = 0.8, P < 0.0001). Fat mass rose by 0.4 ± 0.1 kg (P = 0.0015) on the ultra-processed menu and fell by 0.3 ± 0.1 kg (P = 0.05) on the unprocessed one.
- 4Liking and hunger did not explain it, but eating speed may have. Ratings of pleasantness (P = 0.13), familiarity (P = 0.57), hunger (P = 0.5) and fullness (P = 0.67) did not differ. People ate faster on the ultra-processed menu (17 ± 1 kcal/min more; P < 0.0001), and those who sped up most also ate most extra (r = 0.45, P = 0.047). The foods actually eaten were more energy dense (1.36 versus 1.09 kcal/g; P < 0.0001).
- 5Energy expenditure rose a little but not enough to cancel the extra intake. Expenditure by doubly labeled water was 171 ± 56 kcal/day higher on the ultra-processed menu (P = 0.006), and intake exceeded expenditure by 417 ± 121 kcal/day (P = 0.003). Measured physical activity did not differ significantly.
- 6Short-term glucose and liver measures did not differ significantly. Glucose tolerance was similar, with a Matsuda insulin sensitivity index of 3.9 versus 4.5 (P = 0.1), average continuous glucose monitor readings were 99.1 versus 96.0 mg/dl (P = 0.10), and liver fat in 13 participants was not significantly changed.
Limits & uncertainty
- 1The authors note the setting limits generalisation: 20 people, 28 days, living in a metabolic ward, with food provided free and no choice of meals. The study says nothing about how cost, convenience and personal choice shape what people eat at home.
- 2The authors note there was no run-in or washout between diets. Comparing the final week of each period (459 ± 105 kcal/day) partly addresses this, but comparisons of blood markers with baseline are muddied by the large differences in intake and weight between the diets.
- 3The authors note the trial was not designed to find the cause. The menus were matched on offered calories, macronutrients, sugar, sodium and fibre, yet still differed in added sugar share (about 54% versus 1%), insoluble fibre share (about 16% versus 77%), saturated fat share (about 34% versus 19%) and omega-6 to omega-3 ratio (about 11:1 versus 5:1), and the non-beverage foods were about 85% more energy dense. 'Ultra-processed' here is a bundle of features.
- 4The authors note measurement limits: DXA tends to underestimate body-fat changes, fluid shifts linked to sodium intake (r = 0.63 with fat-free mass change, r = 0.64 with weight change) blur body-composition readings, and doubly labeled water has limited precision. On the unprocessed menu the energy-balance estimate did not match the measured body-energy change, which the authors think may reflect overestimated metabolisable energy.
- 5Our observation: the study was powered for the intake difference, not for the many secondary outcomes (glucose, liver fat, hormones, expenditure), and the report does not describe adjustment for multiple comparisons. A non-significant glucose or liver result in 2 weeks, with daily exercise built in, should not be read as no effect.
What to take from it
This is one of the few randomised tests showing that the way food is made, and not only its nutrient totals, can change how much people eat. On a ward, a mostly ultra-processed menu led to roughly 500 kcal/day more intake without people rating it as tastier or feeling hungrier. If you want to eat less without strict counting, moving more meals toward foods that are less processed, less energy dense and slower to eat is a reasonable thing to try.
It does not show that every packaged food is harmful, that two weeks in a ward predict long-term weight at home, or that processing itself is the mechanism rather than features such as eating speed and energy density. The authors also point to cost and effort: ingredients for 2000 kcal/day of the unprocessed meals were estimated at $151 a week versus $106 for the ultra-processed meals. Confidence is moderate for the short-term effect on intake (a randomised crossover in which all 20 participants finished) and low for long-term weight in everyday life. Anyone managing a health condition should discuss diet changes with their clinician.
Reading your own data
Weigh at the same time each morning after using the toilet, as the researchers did at 6 am, and follow the weekly average rather than single days, because the authors tie part of the weight and fat-free mass changes to sodium-related fluid shifts. Note how processed your meals were and roughly how long each took, since faster eating tracked with eating more; a photo or food log is enough. If you wear a continuous glucose monitor, compare weekly averages, as average glucose did not differ significantly here. The authors say fluid shifts can blur even DXA body-fat readings, so treat small changes on a home body-fat scale with scepticism.
How to read this evidence
Randomisation can support causal inference, but check allocation, blinding, missing outcomes, follow-up and the actual comparator. A trial label alone does not establish high certainty.
Sources & attribution
Ultra-Processed Diets Cause Excess Calorie Intake and Weight Gain: An Inpatient Randomized Controlled Trial of Ad Libitum Food Intake
Cell Metabolism · 2019-07-02
DOI: 10.1016/j.cmet.2019.05.008 ↗PubMed · 31105044 ↗MoveVitals editorial interpretation · educational content · no individual clinical review recorded
Sources checked 2026-11-01
Practical guides
Read in context
Early time-restricted eating linked to more weight loss and lower insulin than late eating
Pooling 41 randomised trials (2287 participants, median intervention 8 weeks), time-restricted eating was associated with 2.15 kg lower body weight than usual diets (95% CI -2.56 to -1.74), and eating early beat eating late by 1.15 kg (95% CI -1.86 to -0.45). Most trials were short, only 6 study arms tested early eating, and the outcomes were body and blood measures, not heart attacks or deaths.
BMJ Medicine
An 8 a.m. to 4 p.m. eating window added to calorie cutting gave no clear extra weight loss
Among 139 patients with obesity on a calorie-restricted diet, adding an eating window of 8:00 a.m. to 4:00 p.m. led to a mean 12-month weight change of -8.0 kg, versus -6.3 kg with calorie restriction alone. The net difference of -1.8 kg (95% CI -4.0 to 0.4; P = 0.11) was not statistically significant, and the interval cannot rule out either a real benefit or none.
The New England Journal of Medicine
Mediterranean diet with olive oil or nuts linked to fewer heart events (corrected trial)
Among 7447 Spanish adults aged 55 to 80 at high cardiovascular risk, those assigned to a Mediterranean diet with extra-virgin olive oil (HR 0.69, 95% CI 0.53 to 0.91) or with nuts (HR 0.72, 0.54 to 0.95) had fewer major cardiovascular events than those told to cut dietary fat. The trial had randomisation errors, which is why the 2013 report was withdrawn and the estimates recalculated, so treat this as strong but imperfect evidence.
The New England Journal of Medicine
Low-fat vs low-carb: similar weight loss at 12 months, and no genotype or insulin match
In 609 adults without diabetes, 12-month weight change was -5.3 kg on a healthy low-fat diet and -6.0 kg on a healthy low-carbohydrate diet, a 0.7 kg gap (95% CI -0.2 to 1.6 kg) that was not statistically significant. A tested genotype pattern and baseline insulin secretion did not predict who did better on which diet. Both arms received 22 coaching sessions and 79% finished, so this describes coached dieting in adults aged 18 to 50.
JAMA

