A small but unusually controlled feeding trial has found that a ketogenic diet reduced liver fat by about 67% in adults with obesity, prediabetes and fatty liver, compared with about 45% for two other diets. The result was published in Cell Metabolism and has attracted attention because all groups lost roughly the same amount of weight, allowing the researchers to ask whether the macronutrient mix mattered beyond weight loss alone.

The study is not a simple verdict that one diet is best. It randomized 55 people and analysed 42 who completed the intervention. Every meal was provided, participants met a dietitian weekly and recorded adherence exceeded 95%. The trial lasted about four to five months. Those design choices make the comparison unusually clean, but they also limit how directly the findings apply to people cooking for themselves over years.

The headline number is still striking. After matched weight loss of about 10%, liver fat measured by magnetic resonance fell far more in the ketogenic arm. Measures of liver insulin sensitivity improved two to three times more, and half of the keto participants no longer met the criteria for prediabetes at the end of the study. The question is how much of that can be attributed to ketosis itself, and how much should be treated as a promising signal from a small trial.

What the trial actually tested

The study, led by researchers at Washington University School of Medicine in St Louis, compared three diets with very different carbohydrate contents. The ketogenic diet supplied about 4% of energy from carbohydrate, 23% from protein and 73% from fat. The Mediterranean diet supplied about 50% carbohydrate, 15% protein and 35% fat. A very-low-fat, plant-forward diet supplied about 70% carbohydrate, 15% protein and 15% fat.

All three plans were calorie-restricted so that participants would lose approximately 10% of their starting weight. That design is central to interpreting the results. Many diet studies compare weight loss achieved by different means, which makes it hard to separate the effect of weight loss from the effect of the food itself. Here, the weight-loss target was similar across groups.

The trial was registered as NCT02706262. It ran from 2016 to 2025 and used a parallel-group design with random allocation. Participants were not masked to their diet, which would have been impossible, but the team responsible for enrolment and the assessors analysing outcomes were blinded to allocation.

The primary outcomes were insulin sensitivity in skeletal muscle and in the liver. Liver fat, de novo lipogenesis, glucose and hormone profiles, HbA1c, lipids and body composition were among the secondary measures. This was a mechanistic study, not a trial of heart attacks, strokes or diabetes incidence.

Who was in the study

Participants were adults aged 18 to 55 with a body mass index between 30 and 50. They had metabolically unhealthy obesity, meaning they also had prediabetes and hepatic steatosis. Prediabetes was defined by at least one of several glucose or HbA1c thresholds. Fatty liver was confirmed by magnetic resonance imaging showing intrahepatic triglyceride of at least 5%.

The people who completed the trial were, on average, about 43 years old with a BMI near 39. Each diet group had 14 completers: nine women and five men. Fifty-five people were randomized in total, so 13 did not complete the intervention. The paper reports that the main findings were similar when those participants were handled in a sensitivity analysis, but the small final sample remains an important limitation.

This population matters. The results cannot be assumed to apply to people with normal liver fat, to those without prediabetes, to older adults or to people with type 2 diabetes requiring medication. The trial was designed to test a specific metabolic question in a high-risk group, not to produce universal dietary guidance.

The participants were also highly selected in another way. They accepted a study in which all food was supplied and weekly dietitian meetings were part of the protocol. That is very different from the conditions under which most people try to change their diet.

How the diets were delivered

Every participant received all of their food. That is unusual in nutrition research and gives the study strong internal validity. The researchers could control the exact macronutrient content, monitor adherence and avoid the vague reporting that undermines many free-living diet trials. Weekly dietitian meetings reinforced the plan and helped participants manage the transition.

Recorded adherence exceeded 95%. That figure should be read as adherence to the provided meals within the trial setting. It does not tell us how easily the same eating pattern could be sustained outside a research programme. The authors themselves note that real-world adherence is likely to be the weaker link.

The ketogenic arm was restricted to about 20 grams of carbohydrate per 2,000 kilocalories. That is a very low intake by any standard and produces ketosis in most people. The plant-forward arm supplied about 351 grams of carbohydrate per 2,000 kilocalories, with only 35 grams of fat. The Mediterranean arm sat between them.

Fat quality was not the main experimental variable. The ketogenic diet contained more saturated fat than the other arms, about 52 grams per 2,000 kilocalories, compared with 19 grams in the Mediterranean diet and 7 grams in the plant-forward diet. Because carbohydrate and fat changed together, the trial cannot tell us whether ketosis, carbohydrate restriction, fat quantity or particular fat sources drove the difference.

Why weight loss was matched

The most important feature of the trial is that all three groups lost about the same amount of weight. The ketogenic group lost 10.4% on average over about 4.6 months. The Mediterranean group lost 10.2% over about 5.4 months. The plant-forward group lost 10.1% over about 5.1 months. The difference between groups was not statistically meaningful.

That matching changes the question. Instead of asking which diet produces more weight loss, the researchers asked what happens to liver and metabolic markers when weight loss is held roughly constant. It is a harder and more interesting question, because clinicians already know that weight loss improves fatty liver. What has been less clear is whether the composition of the diet adds an effect.

The answer appears to be yes for liver fat and hepatic insulin sensitivity. Muscle insulin sensitivity improved by roughly 50% in all three groups, with no significant difference between them. Liver measures behaved differently. That suggests the liver responded to the diet's macronutrient content in a way that skeletal muscle did not.

The result is not evidence that weight loss is irrelevant. Every group lost weight and improved. It is evidence that, after similar weight loss, a very-low-carbohydrate ketogenic diet produced larger changes in some liver-related outcomes.

The 67% liver-fat finding

Liver fat was measured with magnetic resonance imaging, specifically the proton density fat fraction technique. In the ketogenic group, mean intrahepatic triglyceride fell from 19.4% to 6.2%, an absolute drop of 13.2 percentage points. Expressed as a mean of individual relative changes, that is a reduction of about 67%. In the Mediterranean group, liver fat fell from 18.1% to 10.3%. In the plant-forward group, it fell from 17.7% to 9.6%. Both comparator groups recorded relative reductions of about 45%.

The difference between the ketogenic arm and the other two was statistically significant. The paper reports an analysis-of-covariance p value of 0.011 for intrahepatic triglyceride. That does not make the finding immune to uncertainty, but it makes it unlikely to be a random fluctuation within this trial.

The word "cut by 67%" needs careful handling. It is a relative mean reduction, not a 67-percentage-point fall, and it applies to the ketogenic group's average experience over roughly four to five months. It also comes from 14 people who completed that arm. A different group might see a smaller or larger response.

Hepatic de novo lipogenesis, the process by which the liver makes fat from carbohydrate and other substrates, also fell most in the ketogenic group. That is biologically plausible. A very-low-carbohydrate diet supplies less substrate for that pathway and lowers insulin, which regulates it. The study measured these changes rather than merely inferring them.

Beyond fat: insulin and glucose

The trial's co-primary outcomes were insulin sensitivity in muscle and liver. Muscle insulin sensitivity improved by about 50% in every group, regardless of carbohydrate intake. Liver insulin sensitivity told a different story: it improved two to three times more in the ketogenic group than in the other two.

That distinction is scientifically important. Insulin resistance in the liver and in muscle are related but not identical. Liver insulin resistance is closely tied to fat accumulation, glucose production and lipid handling. Muscle insulin resistance is more about glucose uptake after a meal. The finding that one improved similarly across diets while the other did not suggests distinct mechanisms.

Twenty-four-hour glucose and insulin profiles also favoured the ketogenic arm. Washington University's release reports a 20% reduction in 24-hour glucose with keto, compared with about 8% in the other groups. Twenty-four-hour insulin fell 74% in the ketogenic group, compared with 44% and 27% in the Mediterranean and plant-forward groups. Lower insulin levels are consistent with lower carbohydrate intake and improved hepatic insulin sensitivity.

HbA1c, a measure of average glucose over several months, fell from 5.6% to 5.0% in the ketogenic group. It fell by 0.1 percentage points in the Mediterranean group and by 0.4 points in the plant-forward group. The between-group difference was statistically significant. Fasting insulin and fasting triglycerides also fell more with keto.

These are laboratory measures, not clinical endpoints. The trial was too short and too small to show that a ketogenic diet prevents diabetes, heart disease or liver cancer. It shows that several risk markers moved in a favourable direction.

Prediabetes remission, carefully read

One of the most widely reported findings was that prediabetes was "reversed" in 50% of the ketogenic group, compared with 29% of the Mediterranean group and 7% of the plant-forward group. The trial defined remission strictly: participants had to meet normal thresholds for fasting glucose, two-hour glucose after an oral tolerance test, and HbA1c at the end of the intervention.

That is a meaningful result, but it is not proof of a permanent cure. It means that, at the end of a four-to-five-month feeding trial, half of the keto participants no longer met the study's laboratory criteria for prediabetes. The paper does not show what happened after they returned to their own food.

It is also worth noting that all three groups had some participants who met the criteria again or remained above them. Twenty-nine percent in the Mediterranean arm is not a trivial remission rate. The difference between 50% and 29% is based on small numbers: roughly seven people versus four people in groups of fourteen.

The word remission is more accurate than reversal. It conveys improvement without implying that the underlying susceptibility has disappeared. People who lose weight and then regain it are likely to see the metabolic benefits fade. The study did not test long-term maintenance.

Cholesterol and safety results

A common criticism of ketogenic diets is that high fat intake might worsen blood lipids. This trial did not find a between-group difference in LDL cholesterol, apolipoprotein B or total cholesterol. Those measures fell in all groups. HDL cholesterol favoured the ketogenic diet, and fasting triglycerides fell more sharply there.

That does not mean the lipid story is settled. The trial reported no significant difference in 24-hour triglycerides, even though fasting triglycerides fell more with keto. In the ketogenic group, post-meal triglyceride levels were higher, offsetting some of the fasting advantage. The study also did not test different types of fat or large numbers of people with established cardiovascular disease.

The paper reported no serious adverse events and no withdrawals because of an adverse event. That is reassuring in a short, supervised trial, but it is not a long-term safety guarantee. The ketogenic arm included 17 randomized participants and 14 completers. Rare or slow-moving problems would not be detectable in such a sample.

Funding came from the National Institutes of Health and the Foundation for Barnes-Jewish Hospital. The senior author, Samuel Klein, reported advisory fees from several pharmaceutical companies, an investigator-initiated grant from Merck and trial support from Viking Therapeutics. These disclosures do not invalidate the study, but they belong in a full accounting of its context.

Why this is not a verdict on all diets

The most tempting conclusion is that keto is better than Mediterranean or low-fat eating. The study does not support that broad claim. It supports a narrower claim: in 42 adults with obesity, prediabetes and fatty liver, all food provided, a ketogenic diet produced larger reductions in liver fat and larger improvements in some hepatic metabolic markers after similar weight loss.

The Mediterranean and plant-forward diets also worked. Both reduced liver fat by about 45%, improved muscle insulin sensitivity and produced meaningful weight loss. Calling them failures would be misleading. They were active comparators, and the differences emerged in specific liver outcomes rather than across every measure.

The trial also cannot isolate ketosis. It is possible that a lower but non-ketogenic carbohydrate intake would produce similar benefits. The authors explicitly list that as an open question. It is also possible that the benefit depends on following a ketogenic diet during active weight loss rather than during maintenance. The design did not separate those phases.

Because all meals were provided, the study says little about feasibility. Real-world adherence is difficult for every diet, and highly restrictive plans may be harder to sustain. The authors and outside commentators both note that larger, longer trials are needed before clinical guidelines change.

What clinicians and patients should take away

For someone with fatty liver and prediabetes, the study reinforces a familiar first message: losing about 10% of body weight improves liver fat and insulin sensitivity. That happened in every arm. It also suggests that, if weight loss is comparable, very-low-carbohydrate eating may offer additional liver-specific benefits.

That suggestion should not be turned into medical advice without context. People with diabetes, kidney disease, gallbladder problems, pregnancy or a history of disordered eating should discuss restrictive diets with a clinician. The trial excluded many people who would need individualized guidance.

The results are best read as a mechanism study that generates hypotheses. They tell researchers where to look next: hepatic fat synthesis, insulin signalling and the interaction between carbohydrate intake and weight loss. They do not tell a patient exactly what to eat for the next decade.

Public health messages also need to avoid turning one small trial into a culture-war weapon. Diets are not identities. The practical question is what a person can sustain while achieving and maintaining a healthy weight and good metabolic markers. This study speaks to the biological part of that question more loudly than the behavioural part.

What researchers still need to answer

Several gaps are obvious. The sample was small, the duration was short and the participant group was narrow. The trial did not follow people after the supervised feeding ended. It did not test clinical outcomes, different fat qualities or less extreme carbohydrate restriction. It did not include a no-treatment control group.

A larger trial could test whether the liver-fat advantage persists at 12 or 24 months. It could compare a ketogenic diet with a moderate-carbohydrate diet that produces the same weight loss, helping isolate the role of ketosis. It could also examine whether the benefit survives weight maintenance and whether it applies to people with established type 2 diabetes or more advanced liver disease.

Mechanistic work should continue as well. The study suggests that hepatic de novo lipogenesis and insulin sensitivity are involved, but it does not fully explain why muscle and liver responded differently. Understanding that difference could lead to therapies that mimic the liver benefit without requiring extreme carbohydrate restriction.

For now, the 67% figure is a strong research signal with clear boundaries. It is not a promise, not a cure and not a reason to dismiss other dietary approaches. It is a carefully measured result from a small, demanding experiment that has opened a more precise question: how much of metabolic improvement comes from weight loss, and how much comes from what the liver is asked to process?


Corrections and updates

Nexuswild welcomes factual corrections. Email contact@nexuswild.com with evidence and the article URL.