The Study the Keto Community Uses to Say LDL Doesn’t Matter. Then the Follow-Up Got Retracted.

I help run codes in the emergency department.

I have done CPR on people whose hearts stopped before anyone understood what went wrong. I have felt ribs crack beneath my hands. I have packaged people up and rushed them to the cath lab when we could get a rhythm back, and I have watched families fall apart in the corner of the room when we couldn’t. That is where I work.

I am not a weekend warrior typing hot takes from a laptop. I am writing this on my lunch break from a job where I help keep people alive, because the science here matters to me, and because a year ago I found out I was far closer to being the patient on that table than I ever understood.

That is why I care about this study. Not because I dislike keto. Not because I want to win an argument online. I care because I believed a version of the story it was used to tell: that rising LDL was not a real concern as long as my metabolic markers were clean.

I had the clean markers. I had a clean coronary CT angiogram at 36. I had the reassuring podcasts and the confidence that came with them.

Eight years later, I had a severe, flow-limiting blockage in my right coronary artery.

So after my stent, I went back to the papers. Not the blog posts. Not the podcast clips. Not the threads from people who had never run a code or watched an artery win. The actual papers.

The Bottom Line

A 2024 study of 80 people on keto with extremely high LDL found no more total coronary plaque than matched controls. It was used everywhere to argue that high LDL is harmless if you are lean and metabolically healthy. The study could not actually support that claim. It was cross-sectional, it never reported the soft-plaque breakdown it had the technology to measure, and it could not test whether high ApoB matters because everyone in it already had high ApoB.

The 2025 follow-up did measure soft plaque over a year, and it grew. That follow-up was then retracted in 2026, at the request of the authors and the editors, after a methodological problem surfaced.

Here is what I am not saying: I am not saying keto is the villain, and I am not claiming one diet explains my own blockage. What I am saying is narrower and better supported. There is no good evidence that extreme LDL and ApoB elevations on a ketogenic diet are safe just because someone is lean, active, and metabolically healthy. The causal link between ApoB and atherosclerosis is one of the most established findings in cardiovascular medicine, and one underpowered study with a retracted follow-up does not move it.

Vocabulary that matters

A few terms show up throughout. Worth knowing before we go further.

• ApoB (apolipoprotein B): one ApoB molecule rides on each atherogenic particle, so ApoB is essentially a count of the cholesterol-carrying particles most able to lodge in an artery wall. Often a better risk marker than LDL-C.
• LMHR (lean mass hyper-responder): the phenotype at the center of this debate. Lean, active, metabolically healthy people whose LDL shoots up dramatically on a low-carb or ketogenic diet.
• Soft plaque (non-calcified plaque): the live, lipid-rich plaque that can rupture without warning. Measured here as non-calcified plaque volume (NCPV).
• PAV (percent atheroma volume): the share of the artery wall taken up by plaque. A summary number that can move differently than raw soft-plaque volume.
• CCTA (coronary CT angiography): the scan that can actually see plaque inside the artery wall and tell soft plaque apart from calcified plaque.

How I Got Here

I did not come to keto through influencer culture at first. I came through the back door, in 2017, while studying mitochondrial theories of cancer and the Warburg effect. It was a year after a completely clean coronary CT angiogram at 36, and the metabolic argument sounded genuinely compelling. Restrict glucose. Shift fuel use. Improve insulin sensitivity. Lower inflammation. It sounded mechanistic, not magical.

At the same time I was being pulled deeper into the biohacking world because of my background. I had spent 15 years as a CEO in the health and wellness industry. I knew the products, the claims, the conferences, the personalities, and the way a plausible mechanism can turn into a movement. The low-carb, ketogenic, and eventually carnivore-light message fit neatly into all of it.

So I needed to know whether the science I had built my diet around actually said what I thought it said. Because while I was trusting those conclusions, my right coronary artery was quietly closing. My hsCRP was 0.45, no sign of systemic inflammation. No insulin resistance by labs or a continuous glucose monitor. No Lp(a) issue. By every metric the keto and carnivore community uses to wave away an LDL of 160, I should have been the safe case.

I was not the safe case. I was one bad day away from a heart attack at 44.

The 2024 KETO Trial: What It Actually Showed

In August 2024, the KETO trial ¹ made waves. Eighty people on ketogenic diets with a mean LDL-C of 272 mg/dL, more than double the usual recommended level, showed no more total coronary plaque than matched controls with a mean LDL-C of 123. The keto community celebrated. Case closed. LDL does not matter.

Not so fast.

The study was cross-sectional. It compared two groups at one moment in time. It did not follow anyone for 10 or 20 years, it did not count heart attacks, and it did not prove that decades of high ApoB exposure are benign in lean, insulin-sensitive people. And most importantly for my story, it did not give readers the clean soft-plaque answer they thought it gave them.

Why Soft Plaque Is the Whole Game

The 2024 study reported total plaque scores and coronary calcium. It did not give readers the full breakdown of the plaque phenotype, and that distinction matters more than almost anything else here.

Calcified plaque is usually older, more organized disease. It is not “good,” and it is not harmless. It is still evidence that atherosclerosis has been happening. But the plaque that scares me most as an ER nurse is the softer, non-calcified, lower-attenuation plaque that can build silently in the artery wall and rupture without warning. That is the plaque that turns a podcast debate into a family standing in the corner of a resuscitation room, and yes, it happens to people in their 40s and 50s, as I see in the emergency department and nearly experienced myself.

The 2024 trial had the imaging technology to look at it. CCTA can tell soft plaque apart from calcified plaque. But the primary analysis did not report the full soft-versus-calcified breakdown in the way readers needed, which is a notable omission for data they already had.

That omission is not a cosmetic detail. In the SCOT-HEART analysis, low-attenuation non-calcified plaque on CCTA was the strongest predictor of fatal or nonfatal heart attack, outperforming standard risk scores and even challenging the primacy of stenosis severity alone. ¹⁰

This is the study I had in the back of my mind when my LDL hit 160 and I did not blink. It was also there when I felt gnawing chest tightness and mid-back discomfort that I wrote off as anxiety, disc issues, or years of heavy lifting. When my doctor suggested I consider a statin, I politely declined. In my mind, mild-to-moderate LDL elevation was acceptable, because every podcast I listened to told me that metabolically healthy people did not need to worry about LDL the way everyone else did.

That belief did not survive contact with my angiogram.

The Follow-Up That Flipped the Narrative

In 2025, the same research group published longitudinal data on 100 participants. ² This time they used AI-guided quantitative plaque analysis, and this time the soft plaque was measured directly. The signal was impossible to miss.

Non-calcified plaque volume increased by a median of 18.9 mm³ in a single year. In their published reply to criticism, the authors stated this represented a 42.8% relative increase. ⁴

In one year. In lean, metabolically healthy people. In the exact phenotype being described online as protected.

There was also a deeper design problem. There was no proper longitudinal control group receiving the same serial CCTA scans, so the study could not tell us how that progression compared with matched people who did not have keto-induced extreme LDL and ApoB elevations. Without that comparison, you cannot fully contextualize how alarming the progression is, because the study was not built to let you. That is not a footnote. That is a limitation that should have been front and center.

That soft plaque was building in my arteries at a rate I could not feel, could not catch with standard screening, and did not know to look for, because the study that should have raised the alarm was busy telling a different story.

Then It Got Retracted

Here is where most people sharing this research on social media have not caught up.

The 2025 paper was retracted in 2026. ³ Not corrected. Not amended. Retracted. And it is worth being precise about how that happened, because it is more interesting than the usual version. The retraction was made at the request of both the authors and the editors. After publication, a methodological problem with how the imaging had been analyzed came to light, and all parties agreed the errors were too great to fix with a correction. To their credit, the authors raised the concern and requested the retraction themselves.

That honesty is real, and I will give it full weight. But the practical reality stands: the paper that produced the headline “Plaque Predicts Plaque, ApoB Does Not” is now marked as unreliable by the journal, and it is still being shared as if it were not. The claim escaped into the bloodstream long before the retraction caught up to it.

The Attempted Rehabilitation

In January 2026, Budoff and colleagues posted a preprint on medRxiv, a reanalysis of the same cohort using different plaque-quantification methods. ⁵ This is the team returning to the same data with different analytical software.

I want to be precise about what this is. A medRxiv preprint is a manuscript posted publicly before peer review. It has not been vetted by independent reviewers, and it has not been accepted by a journal. It is a draft. And there is a deeper issue: going back to the same dataset with different tools after the original analysis was retracted is a post hoc reanalysis. It may eventually hold up, and it may not. But the bar for credibility here should be higher, not lower, and that bar is peer review, which this preprint has not cleared.

The “ApoB Doesn’t Predict Plaque” Problem

The 2025 study’s marquee claim, that ApoB does not predict plaque progression, is the one most likely to get someone in trouble. Here is what actually happened.

Soft plaque grew in nearly everyone in the study. Median non-calcified plaque volume went up 18.9 mm³. PAV went up 0.8%. The plaque increased. That part is not in dispute.

What the authors found is that within their cohort, the people with higher ApoB did not progress faster than those with lower ApoB. From that, they concluded ApoB does not drive plaque in this population.

The problem? Every single person in the study already had massively elevated ApoB. The median was 178 mg/dL, with an interquartile range of 149 to 214. For context, a desirable ApoB is generally put around 90 mg/dL or lower, and the ESC/EAS guidelines push the target down to 65 to 80 mg/dL for higher-risk patients. ⁹ So the “lower ApoB” group in this study was still running levels most lipidologists would flag as high risk.

This is a textbook restriction-of-range problem. If you took 100 people who all smoke two packs a day and tracked them for a year, you probably would not find a clean dose-response relationship between cigarettes and lung damage either. Not because cigarettes are safe, but because there is no meaningful variation in the exposure. Everyone is already deep in the danger zone.

To actually prove ApoB does not drive plaque, you would need a control group with normal ApoB getting the same scans over the same period. They do not have that. What they have is 100 people with uniformly high ApoB all progressing at roughly similar rates, which is exactly what you would expect if ApoB is causal and everyone is above the threshold that matters.

And this is not a fringe position. The causal relationship between ApoB-containing lipoproteins and atherosclerosis is one of the most established in cardiovascular medicine, confirmed by Mendelian randomization and decades of evidence across study designs. ⁶ A 2019 Mendelian randomization study of more than 654,000 participants found that coronary heart disease risk tracks ApoB particle number, not LDL-C or triglycerides independently. ⁷ And in September 2024, the National Lipid Association published an expert consensus confirming ApoB’s superiority over LDL-C for risk assessment. ⁸ When LDL-C and ApoB disagree, risk tracks with ApoB. That is the consensus of the field, not one person’s opinion.

One underpowered study with a retracted follow-up does not overturn that. It does not even scratch it.

There Is Also a Survivorship Problem

To be in this study, you had to be alive, still on keto, and willing to participate. Anyone who developed cardiovascular disease and stopped the diet was less likely to show up in the data. Anyone who died definitely did not.

I know what that looks like from the other side of it. I have helped run codes on people who never got the chance to become a follow-up data point. They are not sharing their n=1 online. They are not in the keto success stories. They are not in the cohort. Some of them are in the cemetery. That is what survivorship bias looks like when it stops being a statistics term.

One Transparency Note

It is also worth being transparent about who produced this work. At least one author is affiliated with an advocacy foundation in the low-carbohydrate space, and several of the authors are public proponents of the diet being studied, some with paid subscription platforms built around that message. That creates a potential financial conflict of interest.

To be clear, a conflict of interest does not prove bias, and it does not make the data wrong. Plenty of conflicted research holds up fine, and in this case the authors themselves requested the retraction once the problem came to light. A conflict is simply a transparency flag, and a reason to read the evidence with more care, not less, especially when a headline conclusion happens to line up with the authors’ public position.

The Calibrated Claim Audit

Two claims are really on trial in this post. Here is where each one stands.

Claim	Mechanism	Evidence	My read
High LDL/ApoB on keto is harmless if you are a lean “hyper-responder”	Unestablished. No pathway explains why an ApoB particle would behave differently here	Weak. One cross-sectional study, a retracted follow-up, restriction of range, survivorship bias	Not supported by the evidence that exists
ApoB-containing particles cause atherosclerosis	Very strong	Very strong. Mendelian randomization (654,000+), RCTs, imaging, NLA and ESC/EAS consensus	High confidence

Commercial distortion risk: Moderate to high. The LMHR claim is advanced largely by advocates with affiliated foundations and, in some cases, paid platforms in the low-carb space. That does not make the claim wrong, but it changes how carefully I read the evidence.

Let Me Be Clear

I am not here to tell everyone to abandon ketogenic diets. Keto is a tool. It has legitimate evidence in epilepsy, it helps some people lose weight, and it may have specific uses in certain metabolic and neurologic contexts. Context matters.

But “context matters” is not the same thing as “LDL does not matter.”

The problem is not keto. The problem is using metabolic health as a permission slip to ignore ApoB. The problem is telling lean, active, insulin-sensitive people that an LDL of 160, 200, or 270 is background noise because their triglycerides look good and their glucose monitor is boring.

That claim is not harmless. I believed a version of it. I had the clean labs, the clean glucose data, the low inflammation, and a clean coronary scan at 36. Eight years later, I had a severe, flow-limiting blockage in my right coronary artery. A clean scan at 36 did not buy me lifetime immunity from what eight more years of biology could do.

What I’d Do If I Were You

If your LDL is elevated and rising, and you are leaning on one underpowered study whose follow-up was retracted to tell you everything is fine, you owe it to yourself to dig deeper.

Ask your clinician about ApoB testing. Ask whether your particle burden fits the story you have been told. Ask whether a calcium score is enough for your situation, or whether a CCTA with plaque characterization would answer a different question. And do not wait for crushing chest pain. I did not have crushing chest pain. I had nagging chest tightness, mid-back discomfort, and less pep in my workouts. That was enough for my artery to already be in trouble.

If your doctor is telling you your LDL deserves attention, think very carefully before using a podcast or a retracted study as the reason to ignore them.

What Happened to Me

After my stent, the interventional cardiologist who treated me came to see me in recovery. His message was blunt: get your LDL as low as you can. I took him seriously. My LDL is now dramatically lower, and a follow-up CTCA last month showed an open stent and regression of soft plaque.

Did I become vegan? No. Do I still eat steak? Yes. But I gave up keto, I gave up carnivore-light, and I stopped treating LDL like a harmless side effect of being metabolically healthy. Right now my plumbing is not getting worse. It is clearing.

What Would Change My Mind

I am angry about this, but I am not closed-minded. I hold this position with conviction, and I hold it loosely. Three pieces of evidence would update it:

• Long-term hard-outcome data on the LMHR phenotype. Decade-scale prospective data showing equivalent or lower cardiovascular event rates in lean mass hyper-responders versus matched controls. Not one-year imaging. Not cross-sectional snapshots. Real endpoints over 10 to 20 years. The hypothesis has not met this bar.
• A mechanism for why ApoB behaves differently in this phenotype. Not “their metabolism is different.” A specific molecular pathway showing why an ApoB particle in an LMHR person fails to lodge in the artery wall and trigger the cascade it triggers in everyone else. The lipid energy model explains why the LDL is high. It does not explain why that LDL would not do what high LDL always does.
• Mendelian randomization showing LDL-lowering variants do not reduce risk in this phenotype. The cleanest possible test. It has not been done.

If any of those emerged, I would update, publicly, on this site. The path to changing my position is open. It just runs through evidence I have not seen yet.

The Final Signal

• What the claim gets right: total plaque and calcium did not differ in the cross-sectional snapshot, and the LMHR phenotype is a real, interesting metabolic pattern.
• What it gets wrong: it treats a cross-sectional study with no soft-plaque breakdown, a retracted follow-up, and severe restriction of range as proof that high ApoB is safe. It is not.
• What I changed: I stopped treating a rising LDL as background noise, got my ApoB and LDL down hard, and my soft plaque is now regressing.
• What would change my mind: the three specific tests above. None has been met.
• What’s next: if no standard test caught my blockage, which test actually would have? That is the next post.

A Note on Convergence

In April 2026, Peter Attia and Tom Dayspring published a much deeper lipidology version of this same critique (“There is no safe gamble with high LDL cholesterol”). Their angle is technical. Mine is personal and clinical. They are asking whether the lean mass hyper-responder hypothesis has the evidence to overturn the ApoB-causality consensus. I am asking what happens when a real person believes that hypothesis before the evidence is ready. Different angles, same practical conclusion. That is how a field’s consensus actually moves, and we are watching it happen.

What’s Next

Here is the part that still haunts me. My 12-lead ECG was clean. My echocardiogram was normal. My Holter monitor was unremarkable. My labs looked reassuring. My wearable data looked great. Every standard cardiac test said I was fine. I still needed a stent.

So next week I am going to walk through exactly why standard cardiac screening missed a severe, flow-limiting blockage, and what I would ask for now if I were sitting across from a primary care clinician with new chest tightness, mid-back discomfort, or that hard-to-describe sense that my workouts had lost their usual pep. That question might save someone’s life.

---

References

1. Budoff MJ, et al. “Carbohydrate Restriction-Induced Elevations in LDL-Cholesterol and Atherosclerosis: The KETO Trial.” JACC Advances. 2024;3(8):101109. PMID: 39372369 · doi:10.1016/j.jacadv.2024.101109
2. Soto-Mota A, Norwitz NG, et al. “Longitudinal Data From the KETO-CTA Study: Plaque Predicts Plaque, ApoB Does Not.” JACC Advances. 2025;4(7):101686. PMID: 40192608 [RETRACTED]
3. Retraction: “Longitudinal Data From the KETO-CTA Study: Plaque Predicts Plaque, ApoB Does Not.” JACC Advances. 2026;5(5):102824. PMID: 42206798 · doi:10.1016/j.jacadv.2026.102824
4. Soto-Mota A, et al. “Reply: The Keto CTA Study.” JACC Advances. 2025;4(7):101862. PMID: 40450909
5. Budoff MJ, et al. “The Impact of Sustained LDL-C Elevation on Plaque Changes.” medRxiv. Posted January 2026. [PREPRINT, NOT PEER-REVIEWED]
6. Ference BA, et al. “Low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies.” European Heart Journal. 2017;38(32):2459-2472. PMID: 28444290
7. Ference BA, et al. “Association of Triglyceride-Lowering LPL Variants and LDL-C-Lowering LDLR Variants With Risk of Coronary Heart Disease.” JAMA. 2019;321(4):364-373. PMID: 30694319
8. Soffer DE, et al. “Role of apolipoprotein B in the clinical management of cardiovascular risk in adults: An Expert Clinical Consensus from the National Lipid Association.” J Clin Lipidol. 2024;18(5):e647-e663. PMID: 39256087
9. Mach F, et al. “2019 ESC/EAS Guidelines for the management of dyslipidaemias: lipid modification to reduce cardiovascular risk.” European Heart Journal. 2020;41(1):111-188. PMID: 31504418
10. Williams MC, et al. “Low-Attenuation Noncalcified Plaque on Coronary Computed Tomography Angiography Predicts Myocardial Infarction: Results From the Multicenter SCOT-HEART Trial.” Circulation. 2020;141(18):1452-1462. PMID: 32174130 · doi:10.1161/CIRCULATIONAHA.119.044720