Ketogenic diet

• Shift the body's main fuel from glucose to fat and ketones.
• Reduce seizures in drug-resistant epilepsy.
• Lower blood sugar, insulin and HbA1c in type 2 diabetes and reduce medication need.
• Cut appetite to make weight loss easier.
• Lower triglycerides and raise HDL cholesterol.

The diet was designed in the 1920s to copy the metabolism of fasting for children with epilepsy [19]. Today the same rule, very low carbohydrate, is used in two very different settings: as supervised medical therapy for epilepsy, and as a self-directed diet for weight loss and blood sugar.

Trials define a very-low-carbohydrate ketogenic diet as no more than 50 g of carbohydrate a day [7]. Supervised diabetes programs often aim lower: the Virta program asked patients to keep total carbohydrate under 30 g a day, with protein around 1.5 g/kg of reference weight [5]. AumniVIDA's keto protocol scores the same pattern: carbohydrate at most 30 g a day, protein 1.2 to 2.0 g/kg, fat at least 60% of energy and at least 20 g of fiber from non-starchy vegetables.

The versions people follow

The Cochrane review of epilepsy trials pooled classic, MCT and modified Atkins versions [2]. The 130 g line comes from the diabetes remission meta-analysis, which treated anything under 130 g a day as low carbohydrate and under 10% of calories as very low [4].

Ketones rise when carbohydrate and insulin fall. The Virta diabetes study set a target of 0.5 to 3.0 mmol/L blood BHB and adjusted carbohydrate and protein to reach it [5]. Phinney and Volek, who designed that program, advise that protein above about 2.0 g/kg can push ketones down, and that ketones stuck below 0.5 mmol/L despite tight carbohydrate limits are a reason to look at protein (Virta Health).

Real-world ketosis is often lighter than the target. In the same program, laboratory BHB at one year averaged 0.31 mmol/L, below the 0.5 mmol/L goal, even though HbA1c and weight improved [5]. Being in strict ketosis every day may matter less for these outcomes than staying low in carbohydrate.

Nutritional ketosis is not ketoacidosis. Diabetic ketoacidosis is a medical emergency that happens when insulin is missing. People with working insulin rarely get there on diet alone. The exception to know about is the SGLT2 inhibitor class of diabetes drugs: case reports describe ketoacidosis with near-normal blood glucose in people who combined these drugs with a ketogenic diet, in one case within a week [20].

Urine strips and breath meters are cheaper but less direct. If a number matters to you, a blood meter is the one trials use.

This is the original use of the diet and the one with the clearest evidence. The UK trial randomized 145 children who had failed at least two drugs. After 3 months, average seizure frequency was 62% of baseline on the diet versus 137% in controls [1].

The Cochrane review pooled 13 randomized trials with 932 participants. Compared with usual care, ketogenic diets made seizure freedom about 3 times more likely and halving seizures about 6 times more likely in children. Up to 55% of children on a classic 4:1 diet were seizure free at 3 months. Evidence in adults was thinner, and the reviewers rated certainty low to very low because families and doctors always knew who was on the diet [2].

The International Ketogenic Diet Study Group guideline covers who should try it, how to start, which supplements to give, how to handle side effects and when to stop [3]. Families can find hospital programs through the Charlie Foundation.

The common side effects in trials were constipation, vomiting, low energy and hunger [1]. This is a medical therapy, not a diet to start for a child at home.

Two meta-analyses of randomized trials found a small edge: 0.91 kg more weight loss than low-fat diets at 12 months or more [7], and 2.17 kg more at 6 months or more [8]. The DIETFITS trial of 609 adults found no meaningful difference between a healthy low-carbohydrate and a healthy low-fat diet at 12 months (6.0 kg vs 5.3 kg lost), and neither genes nor insulin response predicted who did better [9].

The metabolic advantage debate

The carbohydrate-insulin model holds that high-glycemic diets raise insulin, which locks calories into fat cells and drives hunger, so cutting carbohydrate should help weight loss beyond its calories [21]. Metabolic ward studies have tested this directly:

• At equal calories, switching men to a ketogenic diet did not speed fat loss. Energy expenditure rose by about 57 to 151 kcal a day, near the limit of what can be measured [10].
• When people could eat freely, they ate 689 kcal a day less on a plant-based low-fat diet than on an animal-based ketogenic diet, the opposite of what the model predicts [11].

The National Lipid Association concluded that very-low-carbohydrate diets are not superior for weight loss but may help with appetite control [22]. For a person who finds keto easy to follow and feels less hungry, it can work. It is not a shortcut around energy balance.

A BMJ meta-analysis of 23 randomized trials found that 57% of people on low-carbohydrate diets reached an HbA1c under 6.5% at 6 months, versus 31% on control diets. Remission without diabetes medication was less common, benefits faded by 12 months, and very-low-carbohydrate diets did not beat less strict ones [4].

The Virta program, a non-randomized study funded by the company, combined nutritional ketosis with remote medical supervision. At one year, HbA1c fell from 7.6% to 6.3%, weight fell 13.8 kg, and insulin was reduced or stopped in 94% of those using it. Triglycerides fell 24%, HDL rose 18%, hsCRP fell 39%, and LDL cholesterol rose 10% while ApoB did not change [5].

Stanford's Keto-Med trial gave the same people 12 weeks of a well-formulated keto diet and 12 weeks of a Mediterranean diet. HbA1c improved equally on both. Keto lowered triglycerides more but raised LDL 10%, lowered fiber and several nutrients, and most people drifted back toward Mediterranean eating afterward [6].

Why supervision matters. Blood sugar can drop quickly when carbohydrate falls, so insulin and sulfonylureas often need cutting at the start [5]. SGLT2 inhibitors combined with keto have caused ketoacidosis [20]. See the HbA1c and fasting insulin hubs for what these tests show.

Keto changes blood lipids in two directions at once. Meta-analyses show lower triglycerides and higher HDL, alongside higher LDL cholesterol [7] [8].

Averages hide the spread. In a Swedish feeding trial, every one of 17 healthy normal-weight women saw LDL rise on a 4-week ketogenic diet, by about 70 mg/dL (1.82 mmol/L)1.82 mmol/L (about 70 mg/dL) on average, with ApoB rising too [12]. In people with diabetes or obesity the rise is usually smaller: 10% in both the Virta study and Keto-Med [5] [6].

Does it matter for heart disease? No randomized trial has followed keto dieters to heart attacks. In the UK Biobank, people whose food records matched a low-carbohydrate high-fat pattern had higher LDL and ApoB and about twice the rate of major cardiovascular events over 11.8 years (HR 2.18), though diet was self-reported and observational [13]. Low-carbohydrate eating built on animal fat and protein was linked to higher mortality in cohort data, while plant-based versions were linked to lower mortality [18].

The practical levers within keto are fat type and fiber: swapping butter and fatty meat for olive oil, nuts, avocado and fish, and adding non-starchy vegetables. For how to read your results, see the LDL cholesterol, ApoB and triglycerides hubs.

The term comes from a 2022 analysis of 548 low-carbohydrate eaters. Lower BMI and a lower triglyceride to HDL ratio predicted bigger LDL rises. The authors defined the phenotype as LDL of 200 mg/dL or more, HDL of 80 mg/dL or more and triglycerides of 70 mg/dL or less, and found 100 people who met it. Modest carbohydrate reintroduction lowered LDL sharply in their case series [23]. One case report describes LDL rising from 95 to 545 mg/dL on a ketogenic diet low in saturated fat [24].

What the plaque studies show

• Cross-sectional: 80 people with diet-induced LDL of 190 mg/dL or more (mean 272 mg/dL, for a mean 4.7 years) had no more coronary plaque on CT than matched controls with a mean LDL of 123 mg/dL [14].
• Longitudinal: a 2025 paper reported that ApoB did not predict 1-year plaque progression in 100 such people. It was retracted in 2026 at the request of the authors and editors because of methodological errors [25].
• Case series: German clinicians described 3 patients with LDL of 364 to 682 mg/dL on keto who were referred for suspected familial hypercholesterolemia. LDL fell after modest carbohydrate reintroduction, and the authors warn that severe diet-induced ApoB elevation should not be assumed benign [15].

What is known and not known. It is well established that a subset of lean people see very large LDL rises on keto. It is not known whether that LDL carries the same risk as LDL from other causes, because no outcome trial exists. Mainstream lipidology treats ApoB-containing particles as causal at any level. Anyone in this range has a decision to make with a clinician, and the ApoB and lipoprotein(a) hubs explain the tests involved.

Mental health. Stanford researchers gave a ketogenic diet to 23 people with schizophrenia or bipolar disorder who also had metabolic problems. After 4 months, none met metabolic syndrome criteria and clinical severity scores improved 31% on average. There was no control group, so placebo, attention and weight loss could explain part of it [16].

Inflammation. In the Virta program, hsCRP fell 39% at one year, alongside large weight loss [5]. Separating the diet from the weight loss is not possible in that design. See the hs-CRP hub.

Longevity. Mice started on a ketogenic diet in middle age lived longer and stayed fitter than controls [17]. Mouse lifespan results often fail to carry over to people. In human cohorts, both very low and very high carbohydrate intakes were linked to higher mortality, with the lowest risk at 50 to 55% of energy from carbohydrate, and animal-based low-carbohydrate patterns linked to higher risk [18]. These are observational and cannot prove cause, but they give no support to keto as a longevity diet.

How keto compares with related protocols

• Carnivore diet: removes all plants, usually ketogenic, with far less evidence.
• High-protein diet: shares the appetite benefit without needing ketosis.
• Mediterranean diet: matched keto on HbA1c in Keto-Med with lower LDL [6].
• Intermittent fasting: another route to periodic ketosis.

• Treating keto as permission for unlimited butter, bacon and processed meat, which pushes LDL higher.
• Skipping vegetables and fiber, which leads to constipation.
• Eating so little protein that muscle is lost, or so much that ketones stay low.
• Starting on insulin or SGLT2 drugs without a medication plan.
• Chasing high ketone numbers when the carbohydrate cap is what drives results.
• Never rechecking LDL and ApoB after starting.

• You take an SGLT2 inhibitor (such as empagliflozin or dapagliflozin): the combination can cause ketoacidosis with normal glucose.
• You use insulin or sulfonylureas, unless a clinician adjusts doses, because of hypoglycemia risk.
• You have type 1 diabetes, pancreatitis, liver failure, a fat metabolism disorder or porphyria.
• You have familial hypercholesterolemia, high lipoprotein(a) or known heart disease, without close lipid monitoring.
• You are pregnant or breastfeeding, have kidney disease or a history of kidney stones, or have an eating disorder.
• For a child with epilepsy, only under a specialist ketogenic diet team.

• Fatigue, headache and dizziness in the first week or two, often eased by fluids and salt.
• Constipation, vomiting or diarrhea.
• Rises in LDL cholesterol and ApoB, sometimes to levels that mimic familial hypercholesterolemia.
• Lower intake of fiber, thiamin, vitamins B6, C, D and E compared with a Mediterranean diet.
• Hunger and low energy, reported in the first months of the epilepsy trial.