This is our independent, journal-club summary of Peter Attia’s conversation with Dave Feldman (Episode 19). We distill what was said, separate solid evidence from opinion, and flag where it gets murky. You can check it against the episode yourself.
The technical hinge is apoB particle number (apolipoprotein B: one copy sits on each artery-clogging lipoprotein particle, so the measurement approximates particle number), not just the cholesterol carried inside those particles; our apoB versus LDL-C primer explains the distinction. For the evidence behind the vascular concern, see APOE4, cholesterol, and heart risk. If this debate resembles your own lab results, the useful next step is the lipid-panel walkthrough, not a conclusion drawn from one episode.
Why this one matters if you carry APOE4
If you carry APOE4 and eat low-carb or keto, there is a good chance you have watched your LDL climb (low-density lipoprotein, whose particles can deposit cholesterol in artery walls) and wondered whether that is dangerous or harmless. This episode is a two-sided argument about exactly that question, and it does not land on “do not worry about it.”
The crux is simple. Is a very high LDL particle count harmful when the rest of your panel looks pristine, with low triglycerides, high HDL (high-density lipoprotein, particles involved in returning cholesterol toward the liver), and low inflammation? That is the question an APOE4 carrier on keto needs answered, because E4 already raises both vascular and brain risk. So even though this episode is about lipids in general and never singles out E4 carriers, the answer matters more for you than for most.
What Feldman claimed and how Attia answered
Dave Feldman is a software engineer turned self-experimenter, not a physician or researcher. He walked through his path from engineering into n=1 experimentation on himself, his experience with low-carb diets, and the idea that grew out of it. His big claim, the “lipid energy model,” is that in some people a high-fat, low-carb diet drives LDL up not because something is going wrong, but because the body is shuttling fat for fuel through more lipoprotein particles. In plain terms: when you run mostly on fat, you ship more fat around the bloodstream, and the carriers that do the shipping show up on a lipid panel as a high particle count.
The central character is the “lean mass hyper-responder,” or LMHR. Feldman defines it by a specific lipid triad: LDL cholesterol (LDL-C, the cholesterol carried inside LDL particles) of 200 or higher, HDL-C (the cholesterol carried inside HDL particles) of 80 or higher, and triglycerides of 70 or lower. He says these people tend to be lean, fit, and low-carb, and that the pattern shows up across genotypes. For our readers, he describes it spanning apoE 3/3s and 2/3s as well as 3/4s and 4/4s.
One detail worth flagging: Feldman himself is not an LMHR. His own panel runs a total cholesterol around 186, LDL-C of 131, HDL of 40, and triglycerides of 80. Attia reads that LDL-C as a little over the 50th percentile and flags the HDL of 40 as well below where he wants to see it, not as ordinary. That is part of why total cholesterol on its own tells you almost nothing.
What makes this episode unusual is the pushback. Before the interview started, Attia laid out a “prebuttal” of the model. This is one of the few times Feldman has been interviewed by someone openly skeptical rather than by a sympathetic low-carb host, and that is exactly what makes it useful.
Attia’s two main objections
Attia’s counterargument rests on two pillars a carrier should understand.
Where does the cholesterol come from (the “mass balance” problem). This is Attia’s sharpest objection. He argues the energy model never accounts for where the extra cholesterol physically comes from. By his framing, the body can only do a few things to raise LDL particle number: make more cholesterol, clear less of it, or move it from pools we cannot easily measure, like cell membranes. His conclusion is that the data make increased production by far the most likely driver, and that Feldman could not balance the books, which is why he says his doubt grew over the conversation. Feldman’s published rebuttal argues the LMHR group would convert less VLDL into LDL than someone with type 2 diabetes, so in his view the particle math does not require runaway production. The principle of mass balance is established biochemistry. The specific conclusion Attia draws from it is his contention, and Feldman disputes it directly. Read this as a live, contested argument, not a settled verdict.
Where you are allowed to look for the answer (the “search population” problem). Feldman has a standing challenge: find a study showing cardiovascular harm in people with high LDL but an excellent triglyceride-to-HDL ratio, while excluding genetic and drug studies. He defends excluding genetic studies with a vivid analogy, comparing it to wanting to know whether six-foot-tall people exist but only being allowed to search a kindergarten class. Attia takes the opposite view: throwing out genetic and drug evidence discards some of the cleanest data we have on LDL’s causal role, which is exactly why he resists those exclusion criteria.
Here it helps to separate the established consensus from either man’s opinion. Large Mendelian randomization studies have not found a threshold below which lowering apoB or LDL-C stops cutting cardiovascular risk. That is the basis for the “lower is better” view, and it is well established. The conventional mechanism is also well established: apoB-containing lipoproteins get retained in the artery wall and seed atherosclerosis. In the episode, Attia’s appeal is broader, leaning on decades of lipid literature and asserting LDL’s causal role. The threshold data above is the wider body of evidence that view rests on, and it is also the evidence Feldman is arguing against.
Feldman’s triad, and the outcome data this debate did not have
A few figures are worth holding onto, with the caveat that most come from Feldman’s own observations of self-selected people, not from a controlled study.
- The triad he uses as a definition: LDL-C at or above 200, HDL-C at or above 80, triglycerides at or below 70. He described calling on a conference speaker and testing it on the spot, reporting her LDL-C as 189, HDL-C as 80, and triglycerides as 70. Notice the tension: an LDL-C of 189 is below his own 200 threshold, so by the strict definition that panel does not actually meet the triad, even though he offered it as a live example.
- The high end of what he sees: Feldman reports this group shows the highest LDL he encounters, alongside very low triglycerides and very high HDL, sometimes well above 100. Treat that as his clinical impression from people who sought him out, not as population data.
- Attia’s larger informal sample: Attia says he has seen the hyper-response across what he calls “10s if not 100s of people” in recent years, while noting he does not have the response himself. His read is that the response is variable, and the open question is what else is driving it.
Notice what is missing: hard outcome data. No event rates, no relative or absolute risk reductions tied to this phenotype, and no plaque or outcome data for the LMHR pattern in this conversation. Imaging work on this group, including the KETO-CTA study led by cardiologist Matthew Budoff with Feldman among the co-authors, came later and sits outside this debate.
Fair warning
A few things a carrier should keep in mind before adopting any of this as a personal strategy.
- Conflict of interest and audience. Feldman runs cholesterolcode.com and is a public advocate for the low-carb community. The model he is defending is also the one that most reassures the audience most invested in it. Attia’s position is the more clinically conservative one, and he is candid that this is not abstract for him, because he has real patients with high LDL he must counsel.
- Where the claims outrun the evidence. The idea that a high LDL particle count is safe in the LMHR context is, at most, a hypothesis under investigation. It is not established. The mass-balance question Attia raised went unresolved, and the request to exclude genetic and drug studies removes the very lines of evidence that most cleanly isolate LDL’s causal role. None of this proves Feldman wrong, but it means “my triglycerides and HDL are great, so my high LDL is fine” is not a settled conclusion.
- The APOE4 wrinkle. Feldman’s claim that the pattern spans 3/4 and 4/4 carriers is interesting, but it cuts both ways. The next figures are established external evidence, not numbers from this episode. Across pooled analyses, the APOE4 allele carries roughly a 22 percent higher relative risk of coronary disease for 3/4 heterozygotes and around 45 percent higher for 4/4 homozygotes versus the neutral 3/3 genotype. These are well-replicated relative figures, and the absolute impact depends on your baseline risk: for someone whose 10-year risk is low the absolute difference is modest, while for someone already at high risk it is larger. Whether a 4/4 carrier with an LDL of 200 on keto is safe is simply not settled, and that carrier sits in a different risk context than the model’s more optimistic framing suggests.
A soaring LDL on low-carb is still a particle problem
This is one of the most honest, two-sided cholesterol debates in the podcast world, and the skeptic is in the room, which is exactly what makes it worth your time. But do not walk away thinking high LDL on keto has been shown to be safe, especially as an APOE4 carrier. Feldman offers a provocative, internally consistent hypothesis. Attia offers the established biochemistry of mass balance and the bulk of outcome evidence, both of which still favor caution. If your LDL or apoB has climbed substantially on a low-carb diet, the responsible move is to measure apoB, consider a coronary calcium score, know your genotype, and have a real conversation with a physician about your individual risk, rather than betting your arteries on a model its own creator describes as still being tested.
This is our distillation rather than the guest’s exact words. See the episode.