Ketogenic diet and cholesterol: why “healthy” eating sends your blood fats soaring — and what to do about it

A low-carbohydrate diet improves metabolic markers (triglycerides, HDL, insulin), yet in some lean individuals (the LMHR phenotype) it paradoxically pushes LDL cholesterol to extreme levels. The article examines the metabolic mechanisms behind this phenomenon, the controversial studies tracking atherosclerosis in LMHR subjects, and the conflict between conventional cardiology (LDL = causal risk) and the hypothesis that diet-induced LDL in metabolically healthy people may not be harmful.

Low-carbohydrate diets improve triglycerides, HDL, and insulin sensitivity — and yet in some people they send LDL cholesterol soaring to levels comparable with a severe genetic disorder. How is this possible, and should you be worried?

Imagine you change your diet. You cut out sugar, white bread, pasta. You add eggs, butter, vegetables, fish. After a few months you feel better — more energy, stable blood sugar, a shrinking waistline. Then comes a routine blood test and your doctor raises an eyebrow: total cholesterol 8.5 mmol/l, LDL over 5. The recommendation is unequivocal — statins, or possibly switching the diet back.

But hold on. Your triglycerides dropped by half. Your HDL rose by 30%. Your triglyceride-to-HDL ratio, which predicts heart attacks better than LDL alone, is excellent. Do you have a problem, or are you healthier than before?

This is precisely the question that divides the scientific community into two irreconcilable camps. And the answer — as it turns out — is not black and white.

To understand why a low-carbohydrate diet raises LDL, we have to look into the liver. That is where the metabolic switch occurs that changes the entire logic of fat transport in the blood.

Under normal circumstances the liver burns glucose as its main fuel. Insulin, whose level rises after every carbohydrate-containing meal, governs the whole system — among other things, it maintains so-called LDL receptors on the surface of liver cells. These function like docks: they capture LDL particles from the blood and pull them into the cell, where they are broken down. The more LDL receptors there are, the faster the liver clears LDL cholesterol from the blood.

When you cut carbohydrates, insulin falls. And with it, several things change at once.

First, the number of LDL receptors on the surface of liver cells decreases. Insulin positively regulates their production through the PI3K/Akt/mTORC1 signaling pathway. A study by Ai et al. published in the Journal of Clinical Investigation (2012) demonstrated this mechanism in a mouse model: knocking out the insulin receptor led to a reduction in LDL receptors and increased activity of PCSK9 — an enzyme that actively degrades LDL receptors. The result: the liver captures less LDL from the blood and its level rises. It should be added that the relationship between insulin and PCSK9 is more complex than it appears at first glance — in some models insulin instead increases PCSK9 expression (via a different signaling pathway, SREBP1c). The net effect depends on which pathway predominates in a given metabolic context. (doi:10.1172/JCI61919)

Second, when insulin is low, hormone-sensitive lipase is activated in adipose tissue. Large amounts of free fatty acids are released from fat cells into the bloodstream. The liver takes them up, packages them into triglycerides, and exports them in the form of VLDL particles — a kind of "fat truck." Paradoxically, however, blood triglycerides fall, because peripheral tissues (muscles, the heart) rapidly consume these fats as fuel. VLDL is gradually converted into IDL and finally into LDL. More VLDL therefore means more LDL.

Third, the type of fat in the diet plays a role. Saturated fatty acids (especially C12 to C16 — lauric, myristic, and palmitic acid) increase the content of free cholesterol in liver cells, which inhibits the processing of the transcription factor SREBP-2 and further reduces the production of LDL receptors. The DELTA study (Mustad et al., Journal of Lipid Research, 1997) showed that reducing saturated fat from 15% to 6% of energy intake increased the number of LDL receptors by 10.5%, with a corresponding decline in LDL-C of 11.8%. (PubMed 9101427)

Together, these three mechanisms — fewer LDL receptors, higher VLDL production, and the influence of saturated fats — explain why a ketogenic diet raises LDL in many people.

But not in everyone to the same degree. And this is precisely where the story begins to get interesting.

In 2015, Dave Feldman, a software engineer from Texas, started a ketogenic diet. His LDL cholesterol shot up from normal values to an astronomical 350 mg/dl (over 9 mmol/l). Instead of panicking, he began experimenting. He varied his intake of fat, carbohydrates, and total calories and had his blood drawn every few days. The results were remarkable: he could shift his LDL cholesterol by hundreds of mg/dl within days — simply by adding or removing carbohydrates.

Feldman realized he was not alone. He founded the Citizen Science Foundation and began collecting data from thousands of people on low-carbohydrate diets. A pattern emerged, which he called the Lean Mass Hyper-Responder (LMHR) — a lean individual with a strong lipid response to carbohydrate restriction.

The definition is specific: LDL-C ≥ 200 mg/dl (≥ 5.2 mmol/l), HDL-C ≥ 80 mg/dl (≥ 2.1 mmol/l), and triglycerides ≤ 70 mg/dl (≤ 0.8 mmol/l). This combination is the metabolic mirror image of what cardiologists consider a dangerous lipid profile — high triglycerides, low HDL, and small dense LDL.

In the cohort study by Norwitz et al. (Current Developments in Nutrition, 2022), 18% of 548 individuals on a low-carbohydrate diet met the full LMHR criteria. The mean LDL-C in this group was 320 mg/dl (8.3 mmol/l) — whereas before the diet these people had a mean LDL of around 133 mg/dl (3.4 mmol/l), an entirely normal value. In the general population (the NHANES database, 70,310 people), only three individuals met the criteria. The phenomenon is therefore almost exclusively associated with the state of ketosis. (fulltext)

The typical LMHR is lean (mean BMI 22), metabolically healthy (HbA1c below 6%, insulin below 3 µIU/ml, high-sensitivity CRP below 2 mg/l), and often physically active. And the whole effect is fully reversible — all it takes is to start eating carbohydrates again.

How reversible? A remarkable experiment showed this. Norwitz and Cromwell (Metabolites, 2024) published a case report in which the first author himself (Norwitz, a researcher at Harvard Medical School and at the same time an LMHR) consumed twelve Oreo cookies a day (an extra 100 g of carbohydrates) for 16 days. His LDL-C fell from 384 to 111 mg/dl — a 71% drop. For comparison: rosuvastatin at a dose of 20 mg lowered LDL in the same individual by only 32.5% over six weeks. The cookies thus "treated" high cholesterol twice as effectively as a strong statin. It must, however, be emphasized that this is a case report of a single person — the result demonstrates a mechanism, not a universally valid conclusion. (doi:10.3390/metabo14010073)

A meta-analysis of 41 randomized controlled trials (Soto-Mota et al., American Journal of Clinical Nutrition, 2024) yielded a key finding: a rise in LDL-C on a low-carbohydrate diet appeared statistically significantly only in participants with a BMI below 25. People with grade II obesity, by contrast, lowered their LDL-C on such a diet. BMI was, moreover, more than five times stronger a predictor of an LDL rise than saturated fat intake. (fulltext)

An explanation is offered by the Lipid Energy Model (LEM), which Feldman's team published as a hypothesis article (Norwitz, Feldman et al., Metabolites, 2022). The model views lipoproteins as a system of energy delivery, not merely cholesterol transport. The logic is as follows: a lean body has small fat stores and a limited capacity of adipose tissue to absorb free fatty acids. When carbohydrates are restricted, peripheral tissues must obtain energy almost exclusively from fats — and VLDL particles are the "trucks" that distribute this energy. The less of one's own fat a person has, the more "trucks" they need. More VLDL means more LDL and higher LDL-C in the blood. (PMC9147253)

The model elegantly explains why obese people on a keto diet do not lower their LDL — they have large fat stores that serve as a reservoir, so they do not need such intensive lipoprotein transport. Supporting evidence was provided by the study of Cooper et al. (Frontiers in Endocrinology, 2023), in which BMI and free triiodothyronine — not saturated fat — predicted changes in LDL-C in lean women on a keto diet. (fulltext)

Proponents of the ketogenic diet argue that the type of LDL particles is more important than their total quantity. And they are partly right — the keto diet demonstrably shifts the LDL profile toward large, light particles (pattern A).

A meta-analysis of 38 randomized trials (Lechner et al., American Journal of Clinical Nutrition, 2022) demonstrated an increase in the average size of LDL particles and a decrease in their number on low-carbohydrate diets. (ScienceDirect)

Small dense LDL particles (pattern B) are indeed more dangerous: they penetrate the arterial wall more easily, circulate in the blood longer, and oxidize more readily. The ARIC study (11,419 participants, eleven years of follow-up) confirmed that small dense LDL predicted cardiac events, whereas large light LDL did not. (PMC3999643)

But here is a catch that the keto community often overlooks.

Every LDL particle — regardless of size — carries on its surface exactly one molecule of apolipoprotein B (ApoB). And ApoB is what allows the particle to penetrate the arterial wall and initiate the atherosclerotic process. The systematic review by Sehayek, Sniderman et al. (Journal of Clinical Lipidology, 2025; 593,354 participants, 15 discordance analyses) demonstrated that ApoB is a better predictor of cardiovascular events than LDL-C in nine of nine analyzed studies. Studies from the UK Biobank, the Women's Health Study, and the Framingham Heart Study did not demonstrate that large LDL particles were less atherogenic on a per-particle-number basis.

In other words: what matters is the number of "trucks" (ApoB), not their size. And individuals with an LMHR profile have a median ApoB of around 178 mg/dl — substantially above the recommended limits (below 80–130 mg/dl depending on risk level).

The largest imaging study to date in an LMHR population comes from Feldman's team in collaboration with cardiologist Matt Budoff of the Lundquist Institute.

The cross-sectional part (Budoff et al., JACC Advances, 2024) compared 80 LMHR individuals (mean LDL-C 272 mg/dl, mean time on keto 4.7 years) with 80 matched controls (LDL-C 123 mg/dl). The result: no statistically significant difference in coronary plaque. The median coronary artery calcium (CAC) was 0 in both groups. (doi:10.1016/j.jacadv.2024.101109)

Feldman's team interpreted this as evidence that a diet-induced rise in LDL does not predict atherosclerosis. The study generated media coverage and strengthened the position of keto-diet advocates.

The longitudinal continuation (Soto-Mota et al., JACC Advances, 2025) followed 100 individuals over one year using serial coronary CT angiography. At baseline, 57% had zero CAC. The authors stated that neither ApoB nor LDL-C exposure predicted plaque progression. (doi:10.1016/j.jacadv.2025.101686)

But critics noticed something else. The median increase in non-calcified plaque volume was 18.9 mm³ per year — a value 2.5 times higher than the authors had predicted in the study protocol. Cardiologist John Mandrola pointed out in a commentary for Medscape (2025) that the same group had measured a progression of only 4.9 mm³ per year in healthy controls in the NATURE-CT study. LMHR individuals therefore progressed roughly four times faster. Brad Stanfield compared the progression with type 2 diabetics from Budoff's own earlier cohort (~17.4 mm³/year) and stated that the LMHR group was in a comparable position.

Critics further point to the absence of a control group, the limited range of ApoB (all participants had high values, so a correlation with progression could not emerge), and follow-up too short to capture clinical events. In their response (JACC Advances, 2025), the authors acknowledged an inadequate description of the primary outcome measure in the text and stated that their results are "compatible with a causal role of apolipoprotein B in atherosclerosis."

A remarkable admission from a team that had set out to prove the opposite.

Important context: The journal JACC Advances subsequently issued a so-called "Expression of Concern" regarding the longitudinal part of the study — a formal statement of editorial doubts, relating in particular to the inadequate presentation of the pre-registered primary outcome measure and the absence of a control group. The study was not retracted, but readers should interpret its conclusions with this reservation.

In 2023, the American Heart Association (AHA) rated the ketogenic diet with a score of 31 out of 100 — the lowest of all ten dietary patterns evaluated (Gardner et al., Circulation, 2023). The European Society of Cardiology (ESC/EAS), in its guidelines, set a target of LDL-C below 1.4 mmol/l for high-risk patients and explicitly stated that there is no lower threshold below which lowering LDL ceases to be beneficial.

The strongest argument of conventional cardiology is Mendelian randomization studies — a method that uses genetic variants as a "natural experiment." The consensus statement of the European Atherosclerosis Society (Ference et al., European Heart Journal, 2017) summarized more than 200 prospective studies, Mendelian randomization studies, and randomized trials (more than two million participants, over 150,000 cardiovascular events) with an unequivocal conclusion: LDL causally causes atherosclerosis. Every mechanism of lowering LDL — statins, ezetimibe, PCSK9 inhibitors, and genetic variants — reduces risk in proportion to the absolute decline in LDL-C. (PMC11439750)

Ference et al. (2012, JACC) additionally showed that lifelong genetically lower LDL is associated with a threefold greater reduction in coronary disease risk per unit of LDL-C reduction compared with later initiation of statin therapy. This suggests that the key factor is cumulative exposure — how many years and at what levels your arteries are "bathing" in LDL particles.

The UK Biobank study (Yang et al., JAMA Network Open, 2024; 347,797 participants) confirmed a positive dose-dependent relationship between genetically predicted ApoB and LDL-C and cardiovascular events, all-cause mortality, and cardiovascular mortality — without any threshold. (PMC10799266)

The causal argument, in a nutshell, runs as follows: it does not matter why your LDL is high — whether genetically, through diet, or otherwise. Exposure to atherogenic particles damages the arterial wall. The absence of evidence of harm in LMHR in the short term does not constitute evidence of safety.

It would be convenient to close the whole story with a simple "LDL = bad, lower it." But the science is more complicated.

First, the study by Virta Health (Bhanpuri et al., Cardiovascular Diabetology, 2018) in patients with type 2 diabetes in nutritional ketosis showed no adverse changes in carotid intima-media thickness (CIMT) after one and two years — and this despite a 9.9% rise in LDL-C. The composition of the LDL meanwhile shifted toward pattern A (large light particles). (fulltext)

Second, Mendelian randomization has its limitations. It examines lifelong genetically determined exposure — which is not the same as a few years of elevated LDL against the background of an otherwise excellent metabolic profile (low triglycerides, high HDL, low insulin, low high-sensitivity CRP). The LMHR population is metabolically so different from the populations in statin studies that directly transferring the risk is methodologically questionable.

Third, even Peter Attia — a prominent advocate of aggressive ApoB lowering — acknowledges that in insulin-sensitive individuals (which LMHRs probably are) the dynamics of lipoprotein metabolism may be different. He himself, however, uses three lipid-lowering drugs to keep his ApoB below 50 mg/dl and recommends caution.

Fourth, a key variable that no one has yet measured in the LMHR population is the residence time of LDL particles in the circulation. If LMHRs have high LDL-C because they are exporting many large particles, but these "turn over" quickly (deliver energy and return to the liver), the actual atherogenic exposure may be lower than the ApoB number alone would suggest. This hypothesis, however, has no direct evidence as yet.

What remains valid even after accounting for these nuances: LMHRs have ApoB values comparable to familial hypercholesterolemia, the KETO-CTA study demonstrated plaque progression at a rate comparable to diabetics, and long-term data on clinical events (heart attacks, strokes) for this population do not exist. In medicine, uncertainty generally calls for the precautionary principle.

The standard lipid panel (total cholesterol, LDL-C, HDL-C, triglycerides) is insufficient for a person on a ketogenic diet. The recommendations of the National Lipid Association of the USA (Kirkpatrick et al., Journal of Clinical Lipidology, 2019) and of Norwitz and Loh (Frontiers in Medicine, 2020) explicitly recommend expanded testing.

The basic panel (at the start of the diet and then every 3–6 months) includes a standard fasting lipid profile, ApoB (the NLA 2024 consensus designates it the superior risk marker), Lp(a) once (a genetically determined independent risk factor; normal value below 75 nmol/l), high-sensitivity CRP, fasting insulin and glucose, and HbA1c. (Kirkpatrick: fulltext, Norwitz & Loh: PMC7174731, NLA 2024: PMC11734832)

The expanded panel when LDL-C is elevated: an NMR lipoprotein profile (measures both the number and size of LDL particles, target LDL-P below 1,000 nmol/l). A coronary calcium score (CAC) in people over 40 — CAC = 0 means very low risk, CAC 1–99 intermediate, CAC above 300 is equivalent to evidence of disease. (PMC10635695, PMC11462328)

Genetic testing (a familial hypercholesterolemia panel: LDLR, ApoB, PCSK9 mutations) is warranted in the case of a dramatic rise in LDL-C (above 10 mmol/l) to rule out an inherited disorder of cholesterol metabolism.

A useful orientation marker is the triglyceride-to-HDL-C ratio — a value below 0.9 (in mmol/l) indicates insulin sensitivity and a predominance of large LDL particles (PubMed 11092292). It cannot, however, replace direct measurement of ApoB.

The science of LDL on a ketogenic diet is at a point where two conceptual frameworks clash over the same data.

The Lipid Energy Model correctly identified that the keto-diet-induced rise in LDL is a metabolically distinct phenomenon from the classic blood lipid disorder. It is tied to a low proportion of body fat, is fully reversible, and is accompanied by an otherwise excellent metabolic profile.

On the other side stands the powerful evidence base from Mendelian randomization studies, statin trials, and PCSK9 inhibitor studies, unequivocally demonstrating that ApoB is a causal agent of atherosclerosis — and no metabolic context variable has been demonstrated to be a factor that would overturn this causality.

It is telling that even Feldman's own team acknowledged the causal role of LDL and, in a published statement (Journal of Clinical Lipidology, 2022), recommended that individuals with an LMHR profile actively work with a physician on lipid lowering. (fulltext)

The practical approach for a person on a low-carbohydrate diet? Measure ApoB, not just LDL-C. Lp(a) once. Over the age of 40, consider a coronary calcium score. With ApoB above 130 mg/dl or the presence of plaque on imaging, discuss either dietary adjustment — adding carbohydrates, reducing saturated fat, or possibly switching to a moderate low-carbohydrate diet (50–100 g of carbohydrates per day) — or drug treatment.

Oreo cookies can lower LDL twice as effectively as statins. But that does not mean they are healthier. It means that the question "is my LDL on keto dangerous?" has no answer in a single number on a sheet of paper from the lab. It requires ApoB, knowledge of the context — and an honest admission of what we still do not know.

This article is an analytical review of the current scientific literature. It does not replace consultation with a physician. Data and sources verified as of March 2026.

Mechanisms

Ai et al. (2012): Regulation of hepatic LDL receptors by mTORC1 and PCSK9. Journal of Clinical Investigation. doi:10.1172/JCI61919

Mustad et al. (1997): Reducing saturated fat intake is associated with increased levels of LDL receptors. Journal of Lipid Research. PubMed 9101427

Soto-Mota et al. (2024): Increased LDL-cholesterol on a low-carbohydrate diet in adults with normal but not high body weight. American Journal of Clinical Nutrition. fulltext

The LMHR phenomenon

Norwitz et al. (2022): Elevated LDL-C with a carbohydrate-restricted diet: evidence for a "Lean Mass Hyper-Responder" phenotype. Current Developments in Nutrition. fulltext

Norwitz, Feldman et al. (2022): The Lipid Energy Model. Metabolites. PMC9147253

Norwitz & Cromwell (2024): Oreo Cookie Treatment vs. statin therapy in a LMHR (case report, n = 1). Metabolites. doi:10.3390/metabo14010073

Cooper et al. (2023): Case Report — LMHR phenotype in low saturated fat context. Frontiers in Endocrinology. fulltext

Imaging studies

Budoff et al. (2024): KETO Trial — carbohydrate restriction-induced elevations in LDL-C and atherosclerosis. JACC Advances. doi:10.1016/j.jacadv.2024.101109

Soto-Mota et al. (2025): Longitudinal KETO-CTA — plaque predicts plaque, ApoB does not. JACC Advances. ⚠️ Expression of Concern issued. doi:10.1016/j.jacadv.2025.101686

Bhanpuri et al. (2018): CVD risk factor responses to nutritional ketosis at 1 year. Cardiovascular Diabetology. fulltext

Mainstream cardiology

Ference et al. (2017): EAS Consensus — LDL causally linked to ASCVD. European Heart Journal. PMC11439750

Yang et al. (2024): Dose-response associations of lipid traits with CAD and mortality. JAMA Network Open. PMC10799266

Lechner et al. (2022): Effect of carbohydrate-restricted interventions on LDL particle size. American Journal of Clinical Nutrition. ScienceDirect

Sehayek, Sniderman et al. (2025): ApoB, LDL-C, and non-HDL-C as markers of cardiovascular risk — systematic review, 593,354 participants. Journal of Clinical Lipidology. PubMed 40681368

Diagnostics

Kirkpatrick et al. (2019): NLA scientific statement on low-carbohydrate diets. Journal of Clinical Lipidology. fulltext

Norwitz & Loh (2020): A standard lipid panel is insufficient for ketogenic diet patients. Frontiers in Medicine. PMC7174731

NLA Expert Consensus (2024): Role of ApoB in cardiovascular risk management. PMC11734832

Gardner et al. (2023): Popular Dietary Patterns — alignment with AHA 2021 Dietary Guidance. Circulation. fulltext

Production transparency

The concept, structure, and editorial line of the article are the work of the author, who prepared the content outline, established the key theses, and directed the entire creative process. Generative AI (Claude Opus 4.6, Anthropic) was used as a tool for research, fact-checking, and fleshing out the author's draft.

The author verified the key findings and approved the final wording. No part of the text was published without conscious authorial control. Factual data were verified against the publicly available sources cited in the text.

This procedure complies with the transparency principles of EU Regulation 2024/1689 (AI Act). #poweredByAI