Fifty-five adults with obesity, prediabetes and fatty liver were given every meal they ate until each had lost about a tenth of their body weight. Insulin sensitivity in muscle rose by about half on all three diets, with nothing to choose between them. In the liver the diets separated: fat inside the liver fell 67 percent on the ketogenic diet against 45 percent on the Mediterranean and very low fat diets, and prediabetes resolved in 50 percent of the ketogenic group against 29 and 7 percent. Forty-two people finished, 14 per group, and every outcome measured was a marker rather than an illness.
Fifty-five adults with obesity, prediabetes and a fatty liver were handed every meal they ate. Forty-two of them stayed to the end, and by the time they were measured again each had lost about a tenth of their body weight: 10.4 percent on a ketogenic diet, 10.2 percent on a Mediterranean one, 10.1 percent on a very low fat plant-forward one. The weight loss was the same. The liver was not.
Intrahepatic triglyceride, the fat stored inside liver cells, fell from 19.4 percent of the liver to 6.2 percent in the ketogenic group, a relative drop of 67 percent. On the Mediterranean diet it fell from 18.1 to 10.3 percent and on the very low fat diet from 17.7 to 9.6 percent, 45 percent in each case. Insulin sensitivity in the liver, meaning how readily the liver answers insulin's signal to stop putting glucose into the blood, rose in all three groups, and rose two to three times further in the ketogenic group. HbA1c, the share of haemoglobin carrying sugar and so a record of average blood glucose over the previous two to three months, fell 0.6 points from 5.6 percent in the ketogenic group, against 0.1 points on the Mediterranean diet and 0.4 on the very low fat one. Prediabetes, which the trial counted as resolved only when fasting glucose, the two-hour glucose test and HbA1c had all returned to normal, resolved in 50 percent of the ketogenic group, 29 percent of the Mediterranean group and 7 percent of the very low fat group.
Where the diets did not differ is as much of the result. Insulin sensitivity in skeletal muscle rose by about 50 percent in every group, with nothing to separate them. Neither did LDL cholesterol or apolipoprotein B, the protein that carries LDL and the other artery-clogging particles through the blood: both fell in all three groups by amounts the trial could not tell apart, though the ketogenic diet carried 52 g of saturated fat and 749 mg of cholesterol per 2,000 calories against 19 g and 160 mg on the Mediterranean diet and 7 g and 16 mg on the very low fat one.
How we know
The trial recruited adults with what the authors call metabolically unhealthy obesity, which here meant three things at once: a body mass index between 30 and 50, prediabetes, and at least 5 percent of the liver made of fat on a magnetic resonance imaging scan. Nobody had diabetes. A randomised controlled trial assigns each participant to a group by chance, so that the groups differ in the treatment and not in who chose it, and 55 people were assigned this way: 17 to the ketogenic diet, 18 to the Mediterranean and 20 to the very low fat plant-forward. Forty-two completed the study and were analysed, 14 in each group, 9 women and 5 men in each. They averaged 43.2 years old, with a body mass index of 38.9. The work ran at Washington University School of Medicine in St Louis between February 2016 and April 2025, and is registered as NCT02706262.
Every meal and snack was supplied as a prepared entree. After about four weeks eating enough to hold weight steady, each participant's energy intake was cut onto a path towards losing 10 percent over 16 weeks, then raised again to hold the new weight for three to four weeks before the baseline measurements were repeated. That target was reached at 4.6 months on the ketogenic diet, 5.4 on the Mediterranean and 5.1 on the very low fat. Participants met the study dietitian weekly, and more than 95 percent of the supplied meals and snacks were logged as eaten. Per 2,000 calories the ketogenic diet carried 20 g of carbohydrate, the Mediterranean 254 g and the very low fat plant-forward 352 g.
Insulin sensitivity was measured with a hyperinsulinaemic-euglycaemic clamp, a procedure that infuses insulin at a fixed rate and then feeds in as much glucose as it takes to hold blood glucose level: the amount of glucose required is the measure of how well the insulin is working. Liver fat was measured by magnetic resonance imaging. Two outcomes were named as primary endpoints before the trial began, insulin sensitivity in skeletal muscle and an index of insulin sensitivity in the liver, and each was tested at a stricter threshold to allow for testing two things at once. The paper names intrahepatic triglyceride, HbA1c, body composition and blood lipids as secondary outcomes, and states that everything other than the two primary endpoints was evaluated without any adjustment for the number of comparisons made.
Thirteen of the 55 randomised did not complete. A sensitivity analysis using all available data from all 55 reached the same answer on both primary endpoints as the main analysis of the 42. No serious adverse events occurred and nobody withdrew because of one. The work was funded by the National Institutes of Health and the Foundation for Barnes-Jewish Hospital, and the paper is open access at Cell Metabolism, published on 27 August 2026.
What a marker is not
Every number above is a marker. Liver fat, insulin sensitivity, HbA1c and blood lipids are all measured because they stand in for something further off and harder to reach: a diagnosis of type 2 diabetes, a failing liver, a heart attack, a death. They stand in for it because they usually track it, not because they are it. Nobody in this trial was followed to any of those, and across 42 people over five months nobody could have been.
That distinction bites hardest where the trial found nothing. LDL cholesterol fell 10 mg/dL in the ketogenic group, 22 in the Mediterranean and 8 in the very low fat, and apolipoprotein B fell 6, 15 and 9, with the differences between the groups well inside the range chance alone produces. On a diet taking 23 percent of its energy from saturated fat, that is a real and somewhat surprising result. It is also a five-month one. Sai Das, a nutrition professor at Tufts University, told Scientific American that the liver and triglyceride findings were highly promising while pointing to the short timeline and the small sample, and Jeff Volek, a kinesiologist at Ohio State University, praised the design.
Das also noted that the Mediterranean group's body composition numbers ran the more favourable way, losing more fat and less lean tissue. The trial's own table records those gaps as too small to distinguish from chance: body fat fell 2.7 percentage points in the ketogenic group against 3.3 and 3.4 in the other two, fat-free mass by 3.2 kg against 2.6 and 3.0, and neither comparison separated the groups. The share of the lost weight that came from fat was 71 percent, 74 percent and 75 percent. With 14 people in each arm, a gap of that size is exactly what a trial this small cannot resolve either way.
Why it matters
Diet trials usually cannot answer the question people actually ask of them, because two things move at once. What you eat changes, and how much of you there is changes, and a diet that happens to produce more weight loss produces better numbers while the composition takes the credit. This trial held the weight loss equal on purpose, supplied every meal so that adherence was not the variable, and then looked at what was left. In the muscle, and in the cholesterol, what was left was nothing. In the liver, it was a lot.
The authors are careful about what that does not settle. They cannot say whether a gentler carbohydrate restriction, one that never tips the body into ketosis, would do the same work, and they cannot say whether the liver benefit depends on the diet being eaten while the weight is coming off rather than started once it has gone. Both are the sort of question a five-month trial in 42 people opens rather than closes. What it does establish is narrower and firmer than the headline it will attract: in adults carrying obesity, prediabetes and a fatty liver, the composition of the food changed what a tenth of body weight was worth, and where it changed it was the liver.
This is a small mechanistic trial, not a population result. Fifty-five people were randomised and 42 finished, 14 in each group, and the authors say plainly that differences between the diets may have been missed at that size. Every outcome it reports is a marker: fat inside the liver, insulin sensitivity, HbA1c, blood lipids. Nobody was followed to a heart attack, a diabetes diagnosis, liver disease or a death, and at 14 people per group nobody could have been. The intervention ran about five months, which is long enough to move those markers and too short to say what a diet carrying 52 g of saturated fat and 749 mg of cholesterol per 2,000 calories does to the arteries over years. LDL cholesterol, meaning the cholesterol carried on the particles that build up in artery walls, and apolipoprotein B did not separate between the groups by five months, and that is a finding about five months rather than evidence of safety at five years. All food was supplied as prepared meals and participants saw a dietitian every week, so more than 95 percent of the meals were logged as eaten: the trial tests what these diets do to the body, not whether anyone can keep to them unaided. It ran at a single centre, Washington University School of Medicine, in adults aged 18 to 55 who all had obesity, prediabetes and a fatty liver at entry and none of whom had diabetes. The authors add two limits of their own: the trial cannot say whether a gentler carbohydrate restriction that never reaches ketosis would do the same, and cannot say whether the liver benefit depends on the diet being eaten while the weight is coming off rather than afterwards. The senior author declares advisory board fees from AbbVie, 89Bio and Boehringer Ingelheim, an investigator-initiated grant from Merck, and support for an industry-initiated trial from Viking Therapeutics.
What any of these diets does to a person. The trial reports averages across three groups of 14, over about five months, and an average is not a forecast for anyone. It does not establish that the ketogenic diet prevents type 2 diabetes, liver disease or a cardiovascular event, because it measured none of those: it measured markers that usually track them. It does not show the ketogenic diet is safe for the heart over years, since the absence of an LDL cholesterol difference at five months is a result about five months. It does not compare the diets at equal calories with no weight loss, because every group lost about 10 percent by design. It says nothing about anyone with diabetes, anyone under 18 or over 55, or anyone without a fatty liver, all of whom were outside the entry criteria. And the authors state two questions of their own that it leaves open: whether a less severe carbohydrate reduction that never induces ketosis would produce the same liver changes, and whether the same benefit would appear if the diet were started after the weight was lost rather than during.
Did the ketogenic diet produce more weight loss than the others?
No, and that was the point of the design. Weight loss was matched at 10.4 percent on the ketogenic diet, 10.2 on the Mediterranean and 10.1 on the very low fat plant-forward diet, reached at 4.6, 5.4 and 5.1 months. Holding it equal is what allows the trial to separate what the composition of the food did from what losing weight did.
What is intrahepatic triglyceride?
It is fat stored inside the cells of the liver, measured here by magnetic resonance imaging as a percentage of the liver. Having at least 5 percent was one of the conditions for taking part, and everyone in the trial met it at entry. It fell from 19.4 to 6.2 percent in the ketogenic group, from 18.1 to 10.3 percent on the Mediterranean diet and from 17.7 to 9.6 percent on the very low fat one.
Did all that saturated fat raise cholesterol?
Not over the roughly five months this trial ran. LDL cholesterol, meaning the cholesterol carried on the particles that build up in artery walls, and apolipoprotein B both fell in all three groups by amounts the trial could not tell apart, although the ketogenic diet carried 52 g of saturated fat and 749 mg of cholesterol per 2,000 calories against 7 g and 16 mg on the very low fat diet. That is a result about five months, and the trial was not built to follow anyone longer.
How small is 14 people per group?
Small enough that the authors say differences between the diets may have gone undetected for want of statistical power. Of 55 people randomised, 42 completed and were analysed, 14 in each group, and a sensitivity analysis using all available data from all 55 reached the same conclusion on both of the outcomes the trial had named in advance as its primary endpoints. Where this trial reports no difference between the diets, that is an absence of evidence of a difference rather than evidence that none exists.
This is research reporting, not medical advice. We describe what studies found. We never tell you what to do. Talk to a doctor before changing anything about your health. Read our full position.
The evidence, read properly, every three days.
We read the paper so you do not have to. Free, and no supplement ads, ever.
Powered by Buttondown