Read more

Picking up where we left off

We’ve spent the last two weeks examining the interaction between the metabolic milieu and LDL-related metrics from slightly different angles.

• Part 1 — The Copenhagen Male Study. Stratified by TG and HDL-C.

  • Among those with lower TG and higher HDL-C, 8-year heart disease incidence was similar whether LDL-C was below or above 170 mg/dL — 4.3% vs 5.0%.

• Part 2 — The Quebec Cardiovascular Study. ApoB divided at the median across levels of fasting insulin.

  • Higher ApoB had ~1.8× the odds of heart disease when insulin was lower, vs. ~11× when insulin was higher.

  • Insulin remained associated with heart disease even after accounting for ApoB, suggesting that the association between higher fasting insulin and heart disease was not fully explained by differences in ApoB in their modeling.

Both studies asked about the milieu that surrounds the ApoB. This time, we will take a look inside the ApoB itself.

What’s in your ApoB?

A couple of points worth noting here.

An ApoB level is often thought of as the single best summary of ApoB-containing particle count — every LDL, IDL, VLDL, and remnant particle carries one ApoB molecule, so measuring an ApoB level is essentially a way of counting these particles.

But combining all of the different particles together into one tidy metric doesn’t tell you about the possible distribution of the particles within a given ApoB level.

Two people can carry an identical ApoB level while having very different distributions of particles:

• One dominated by larger LDL particles with few VLDLs or remnants

• Another carrying a larger proportion of VLDL particles and remnants with relatively smaller LDL particles.

When ApoB is used as the summary measure, these particles are effectively counted together irrespective of which type is contributing to the total level.

So, that raises the question this article asks:

Should all ApoB levels be treated equally?

To provide some information to help answer this question, we may need to look beyond a standard lipid panel — and a large UK Biobank analysis with some advanced lipoprotein metrics comes in handy here.

Subscribe now

Enter the UK Biobank

A 2023 UK Biobank (UKB) analysis1 followed ~90,000 participants who were free of cardiovascular disease and not taking statins at baseline. Over a mean follow-up of ~11.5 years, researchers tracked incident coronary heart disease (CHD) — defined as MI, unstable angina, coronary procedures, and coronary death.

Instead of relying on LDL-C from a standard lipid panel and a single total ApoB level, investigators used nuclear magnetic resonance (NMR) spectroscopy — via Nightingale Health — which characterizes the distribution and composition of lipoproteins in great detail, far greater than a standard lipid panel.

The information provided by this lab test allowed the researchers to look within the total ApoB level and ask:

Which ApoB-containing particles carried the strongest association with CHD?

A minor fraction with a major signal

Before we get to that interesting signal, we want to reinforce what this dataset as well as many others often find: ApoB tracked very closely with total LDL particle number.

That is expected — LDL particles generally make up the large majority of ApoB-containing lipoproteins (~90-95%), so conventional lipidology often treats ApoB and LDL particle count as if they are largely interchangeable.

So far, so good — now let’s get to the signal — and it relates to VLDL particle concentration.

• When the researchers accounted for ApoB, VLDL particle concentration stayed significantly associated with CHD (hazard ratio (HR) 1.18 per standard deviation (SD), 99% CI 1.09–1.29), which would be considered unusual in and of itself, given VLDL particles in this dataset only accounted for ~9% of total ApoB.

• But here’s the kicker — when they did the opposite and accounted for VLDL particle concentration, the association between total ApoB and CHD turned non-significant (HR 1.04 per SD, 99% CI 0.96–1.13) — in the authors’ words, “the association of total ApoB level given VLDL concentration attenuated to null”.

Let’s sit with that result for a moment…

VLDL particles are generally a minor fraction of total ApoB — again, ~9% in this particular cohort; LDL made up the vast majority.

Yet these data suggest that minor fraction appeared to carry not only an independent signal from the total ApoB, but when the minor fraction was accounted for with total ApoB, total ApoB no longer carried a significant association with heart disease in their modeling.

These results suggest something we believe may be very relevant:

An ApoB level alone may not tell you enough about what kind of ApoB-containing lipoproteins are present — or, perhaps more importantly, what metabolic state produced them.

A second surprise: what’s inside your LDL?

Before we dive into why the VLDL particle concentration result may possibly make some sense, we want to point out an additional observation — stick with us here because the concepts overlap.

The same dataset also looked inside the LDL particles — at their composition — and likewise, analyzed the association with CHD after accounting for total particle number.

• When total LDL particles were in the model, adding the cholesterol content of said particles had essentially no independent association with CHD (HR ~1.03, not significant).

• However, triglycerides (TG) told a different story. The TG content of LDL — often denoted as LDL-TG — carried a HR of 1.18 per SD (99% CI 1.10–1.26).

In other words, in this model with total LDL particle count, it wasn’t the cholesterol content that tracked with events — it was how triglyceride-enriched they were.

The TG carried specifically inside LDL particles may hold information about the milieu that a standard TG level misses. This LDL-TG finding is quite consistent across a number of other datasets (stay tuned for future articles), which brings us to the question of why?

Why VLDLs? Why TG in LDLs?

Why seeing these patterns combined gives us a powerful clue

Now for a key question — why do ApoB, VLDLs, and LDL-TG tend to rise together?

The short answer is that these metrics tend to rise together because they may often share a similar cause: insulin resistance.

One well-studied link here runs through a small apolipoprotein called ApoC-III.

Within insulin-sensitive physiology, insulin generally suppresses the production of ApoC-III. But when insulin signaling in the liver is impaired — as in insulin resistance — ApoC-III levels commonly increase.

So, what does ApoC-III do and why does this matter?

ApoC-III slows the clearance of TG-rich lipoproteins (like VLDLs) in two ways:

• It inhibits lipoprotein lipase — the enzyme that unloads TG from VLDL — leaving the partially digested end products called remnants.

• It slows the liver’s uptake of the remnants by interfering with the receptors responsible for clearing those particles.

VLDLs and their remnants may hang around in circulation longer, resulting in a rise in VLDL particle concentration, which tracks with elevated TG on a standard lipid panel.

What follows helps explain the LDL-TG finding.

Cholesteryl ester transfer protein (CETP) and LDL-TG

When VLDLs, remnants, and circulating TG are elevated, an enzyme called CETP begins dumping TG into LDL (as well as other lipoproteins) in exchange for cholesterol — a passive process that simply follows with excess TG-rich particles in circulation. That leaves LDL TG-enriched — the LDL-TG our UK Biobank study measured directly.

Another enzyme — hepatic lipase — then strips those TG back out of the LDLs, and what remains is a smaller, denser LDL (sdLDL) particle.

Because each sdLDL carries less cholesterol, it takes more of them to move the same amount of cholesterol. And since ApoB is a count of particles, a shift toward the production of sdLDL can push ApoB higher relative to LDL-C. Higher ApoB relative to LDL-C is what conventional lipidology refers to as discordance (more on this in a future article).

So, the same metabolic environment that raises VLDL and LDL-TG can also push ApoB upward relative to LDL-C.

The takeaway: the VLDL, LDL-TG, and often ApoB signals may be representing a broader dysfunctional metabolic milieu.

Why this loops back

Here’s where we take this full circle.

VLDLs and LDL-TG don’t just rise at random. They tend to climb in insulin resistance — alongside higher TG, lower HDL-C, higher glucose, higher waist circumference, higher blood pressure, higher inflammatory tone, and you guessed it — higher ApoB. This harkens back to collinearity in Part 1 — an elevated ApoB may be part of a broader, remnant-rich, metabolically dysfunctional pattern.

Contrast this with a higher ApoB level consistent of a higher proportion of LDL against a lower proportion of VLDL — a pattern occurring alongside higher HDL-C with lower TG — features more typical of a leaner, insulin-sensitive phenotype, including many lean mass hyper-responders.

Same ApoB number. Different particle distribution. Different metabolic milieu that produced it. And, possibly, different association with heart disease.

But keep in mind that these patterns are only a snapshot of the bloodwork. The metabolic milieu isn’t just the markers you can count at a single moment—it’s the processes that produced those markers in the first place.

Some limitations to consider

As always, there are important caveats:

• ApoB contrast. Across quintiles, ApoB ranged from ~70 to 105 mg/dL — a relatively narrow exposure contrast that may have limited the ability to detect a clear dose-response relationship.

• The top quintile may be borderline. ApoB ticked up at its highest quintile despite the slope being relatively flat — so “attenuated to null” may or may not hold as ApoB continues to climb.

• NMR subclass measurement may have limitations. Concerns have been raised about the precision of specific lipoprotein subclass measurements using this particular assay.

• Observational. This describes associations; it may not establish that any particle class causes events — although still may be helpful for triangulation.

• One dataset. These findings may not be consistent with other datasets and should be replicated to be considered more robust.

Where this leaves us — the end of the start of the evergreen series

Let’s briefly look at these first three studies together.

• Copenhagen: LDL-C’s association changed depending upon TG and HDL-C levels.

• Quebec: ApoB’s association appeared to be stronger alongside higher vs. lower insulin; insulin (and to be fair, ApoB as well) remained independently associated with heart disease.

• UK Biobank: the minor fraction of ApoB — the VLDL particle concentration — carried its own signal, and when modeling VLDL and ApoB together, ApoB lost its association with CHD.

These papers may point to the same idea from different angles: the same lipid number can carry a different association with heart disease depending on the metabolic milieu that produced it.

This reframes the question:

Did ApoB go up?

To:

What changed with it — and what kind of metabolic environment produced it?

This may be the more interesting question here — and it’s the question the rest of this series will continue to explore in future installments.

This is Part 3 of our Metabolic Milieu series. As always, we’d encourage you to read the underlying papers, note the limitations, and resist the urge to oversimplify complex biology.

Read more

Picking up where we left off

In Part 1, we examined the Copenhagen Male Study1 which gave us our first look at some answers for the overarching question behind this series:

Does the metabolic milieu change the association between LDL-related metrics and heart disease?

So what did we find? Among men with lower TG and higher HDL-C — generally considered more metabolically healthy:

• Higher LDL-C (>170 vs ≤170 mg/dL, the cohort median) was associated with a relatively small difference in ischemic heart disease (IHD) incidence

• The absolute event rate stayed fairly low across both LDL-C subgroups — less than half the rate in the higher-TG/lower-HDL-C group.

But Copenhagen certainly has some limitations. In particular, it looked at LDL-C, and conventional lipidology generally considers ApoB the stronger predictor of cardiovascular risk. So, the question we ended our last article on was:

If ApoB is the better marker, does the metabolic milieu still seem to change how strongly it tracks with heart disease — or does ApoB capture the full association with heart disease independent of the environment?

The Quebec Cardiovascular Study2 gives us some insights here — by pairing ApoB with one of the most popular metrics in the metabolic health community: fasting insulin.

Subscribe now

First, a point we touched on earlier

Recall the collinearity point from Part 1. Markers of metabolic dysfunction tend to cluster together, which makes isolating the independent association between any one of them and heart disease quite challenging. ApoB may be no exception — in many populations, higher ApoB may track alongside higher TG, lower HDL-C, higher blood pressure, and impaired glucose handling3.

But ApoB may have an unusual property that sets it apart from that cluster: it doesn't always move with the rest of the group. In some settings the metabolic milieu may be improving even as ApoB rises — as in lean mass hyper-responders, who tend to be leaner with lower TG and higher HDL-C despite having higher ApoB.4

So, if ApoB can rise in a presumably favorable AND unfavorable metabolic environment, then splitting people by a marker of said environment — like fasting insulin — may help us determine whether the same ApoB number has a different association depending on the metabolic milieu it is tracking with — which brings us to Quebec.

Enter the Quebec Cardiovascular Study

The Quebec Cardiovascular Study followed 2,103 men, ages 45 to 76, who were free of IHD at baseline. Over 5 years, 114 individuals had their first event. The analysis used a case-control design: those who developed IHD were matched with a control who likewise stayed event-free, and had similar age, body-mass index, smoking status, and alcohol use.

A couple important points about this paper:

• First, diabetics were excluded; therefore, this is a look at hyperinsulinemia in nondiabetic men — the insulin levels aren’t just standing in for overt diabetes.

• Second, because it’s a case-control study, the results are presented as odds ratios (ORs). There were no absolute event rates denoted for each subgroup — we will do some fun back calculations to estimate this, but these data were not explicitly reported in the study.

Worth noting, the metabolic metric they used for stratification —fasting insulin — is an imperfect marker of insulin resistance, but it’s a reasonable proxy, similar to TG and HDL-C levels.

For the key analysis, the researchers divided the sample into thirds by fasting insulin and split ApoB at its median (119 mg/dL):

• Fasting insulin thirds: <12, 12–15, and >15 μU/mL

• ApoB: below vs. at/above 119 mg/dL

The reference group — assigned an OR of 1.0 — was men in the lowest insulin third and below-median ApoB.

For more on ORs, check out the glossary of terms here.

So, what did they find?

Relative to that reference group:

• Lowest insulin third + higher ApoB: about 1.8× the odds of IHD

• Highest insulin third + lower ApoB: about 3.2× the odds (p=0.04)

• Highest insulin third + higher ApoB: about 11× the odds (p<0.001)

So, what is this saying here?

• Higher ApoB in isolation — in the lowest insulin group— was associated with about 1.8× the odds vs. the reference group.

• But higher insulin in isolation — in the lower ApoB group— was associated with about 3.2× the odds vs. the reference group.

In other words, the metabolic milieu (delineated as fasting insulin) carried a larger apparent signal than the lipid metric (delineated as ApoB) did based on the cut points provided — though, worth noting, the insulin contrast here is top-third vs. the bottom, while ApoB is split at the median so the comparison isn't perfect.

• And when both insulin and ApoB were elevated together, the odds were 11×.

The authors described this as a “synergistic effect”: the strongest association was not higher ApoB on its own, but higher ApoB along with higher fasting insulin (again, the collinearity point).

Now time for some fun calculations 🤓

Relative odds like “1.8×” and “11×” bring up an obvious question: 1.8× what? 11× what? What does this exactly mean?

The study didn’t report absolute event rates for each group — but using some inferences and back calculations, we can actually estimate the absolute event rates to a reasonable degree (see the Appendix for more info).

So, here we go: the full cohort ran about a 5.4% event rate over 5 years. If we look at the six insulin/ApoB groups (assuming ~1/6 in each group) vs. the overall rate and apply the reported ORs, this puts the reference group — lower insulin, lower ApoB — at ~1.1% over 5 years, climbing to ~10 to 12% event rate when both insulin and ApoB were higher.

For our purposes, we would probably want to know the difference between higher and lower ApoB within the lower insulin subgroup — this is the split that distinguishes leaner, more insulin sensitive people between higher and lower LDL-related metrics.

So, moving from lower to higher ApoB corresponds to about 1.8× the odds so in absolute terms, this equates to an estimated event rate changing from ~1% to ~2% over 5 years. Juxtaposed to when insulin was high, that same higher ApoB tracked with a far larger absolute event rate at ~10 to 12%.

The estimated weight of higher ApoB looked more modest in a favorable metabolic setting and carried a much higher OR in an unfavorable one — similar to the pattern we saw with LDL-C in Copenhagen.

A few points on insulin and ApoB

In conventional lipidology, ApoB is sometimes described as capturing much of the insulin resistance-mediated association with heart disease — however, Quebec addressed this assertion directly with their modeling. Insulin’s OR was about 1.7 per standard deviation, and it moved to about 1.6 after adjustment for ApoB (as well as LDL-C, TG, and HDL-C).

In plain English: the metabolic environment was predictive of events even when ApoB was accounted for in the modeling. ApoB did not fully encapsulate the insulin resistance-mediated association; insulin carried some association that ApoB didn’t appear to fully capture.

(Of note, ApoB also survived the reverse adjustment too — holding an odds ratio of about 1.9 with insulin in the model. Both appeared to stand on their own, but together, they carried the highest odds of all.)

Likewise, lower insulin was defined as <12 μU/mL — lower than the other groups, but not a widely accepted cut-point for insulin sensitivity.

What would happen if insulin were stratified at say <5 μU/mL? If insulin were very low, might the association between higher ApoB and cardiovascular disease change?

The same question may apply to ApoB as well — the study split it near its median (~119 mg/dL), but a wider contrast, say ApoB >160 vs. <80 mg/dL, might be different.

Some limitations to consider

Quebec draws an interesting picture, but it certainly comes with its caveats (like every study):

• The analysis was fairly small. Though the cohort was large overall, the insulin × ApoB comparisons consisted of 91 cases and 105 controls. Smaller number of events may increase the uncertainty around the estimates.

• The interaction wasn't statistically significant. The formal test for a multiplicative ApoB and insulin interaction didn't reach statistical significance (p=0.2). This essentially means the modeling didn't establish a true “synergistic” effect between the two metrics. However, to be fair, the analysis may have been underpowered to detect the interaction.

• Absolute rates weren’t reported by subgroup. To reinforce, our figures above are estimates.

• Men only, and observational. Like Copenhagen, this was men only and almost entirely those of French-Canadian descent — likewise, the study was observational, so it can describe associations, but methodologically, has challenges with establishing explicit causation.

Where this leaves us

Copenhagen used TG/HDL-C and LDL-C stratification. Quebec used fasting insulin and ApoB stratification. Both point to a similar idea around the series: the metabolic milieu around LDL-C or ApoB may determine how strongly they associate with heart disease — even if neither study offers completely definitive answers here. Likewise, insulin appeared to associate with events even when accounting for ApoB and other lipid metrics in the modeling — ApoB did not appear to capture all of the association.

Next week, we will move to the UK and dive into their wonderful biobank of treasures, taking a peek at a large study investigating advanced lipoprotein metrics and their association with heart disease.

As a little teaser, do note, a single ApoB value is derived from LDL particles, VLDL particles, IDL, and remnants — and those don’t always necessarily mean the same thing. They can reflect different metabolic milieus and different lipoprotein trafficking dynamics.

So perhaps the next question isn't only what milieu surrounds the ApoB, but which particles make up that ApoB — and what that tells us about the environment in which the higher ApoB emerges?

We’ll pick it up there in Part 3.

Share

Leave a comment

Appendix: How the absolute event rates were estimated

• Overall, there were 114 events among 2,103 men, or 5.42% over 5 years.

• Figure 1 reports 6 ORs for the ApoB/insulin groups: 1.0, 1.8, 3.0, 9.7, 3.2, and 11.0.

• We treated the ORs as rough risk ratios, which may be reasonable here given the relatively low overall event rate.

• Insulin was divided into tertiles and ApoB at the median, so each of the 6 cells should contain ~1/6 of the cohort if the 2 variables are not strongly associated. Their reported correlation was fairly weak (r 0.16), so we think this is a reasonable approximation.

• The 6 ORs sum to 29.7, giving an ave OR of 4.95.

• Since the overall 5-year event rate was 5.42%, the implied event rate in the reference group is:

  • 5.42% ÷ 4.95 ≈ 1.1%

• We then multiplied that ~1.1% reference rate by each reported OR to estimate the absolute event rate for each group.

These are the full reconstructed estimates below:

Some checks on the estimate:

• Internal consistency: The 6 estimated event rates ave to about 5.45%, which is very close to the 5.42% event rate reported for the full cohort.

• External consistency: We also compared the estimates with WOSCOPS (citation from Copenhagen in Part 1), which reported ~5.3% 5-year event rate in men with isolated hypercholesterolemia and ~14.1% in those with metabolic syndrome. Our estimates are in the same general range. The higher insulin/ApoB group is at ~10–12%, the lower insulin groups are closer to ~1–3.5%.

• Ranges: OR as a risk ratio becomes less accurate as event rates go up, so we also estimated using the standard OR to RR conversion equation: RR = OR / [(1 − reference risk) + (reference risk × OR)]. Using the ~1.1% reference rate lowers the estimate from ~12.1% to ~10.9%. — we think ~10–12% is a fairer estimate than giving a single number.

Read more

Two people, same ApoB

Let’s take a moment to picture two individuals with the exact same ApoB level.

The first carries some extra weight around their middle, has higher triglycerides (TG) and lower HDL-C, consistently elevated blood pressure readings, and blood sugar creeping up year after year.

The second is lean and physically active, with lower TG and higher HDL-C, no signs of blood pressure issues, optimal fasting glucose, and has adopted a lower carbohydrate diet for years.

They have the same number on the lab report — but the metabolic milieu that gave rise to it may be very different.

Within conventional lipidology, LDL-C and ApoB are generally regarded as part of the causal pathway for atherosclerosis — and ApoB-containing lipoproteins are often discussed as though their association with cardiovascular disease is the same in every scenario: more particles = more risk, and that risk is almost always considered unacceptably high.

But at the end of the day, ApoB may not always arise the same way for the same reasons. So, the question we pose here is…

Does this context even matter? When the same number shows up inside two nearly opposite metabolic environments, does it carry the same association with heart disease in both?

ApoB may rise for different reasons: starting with insulin resistance

One of the most common settings in which ApoB rises is insulin resistance, which tends to cluster with other traits associated with chronic disease: abdominal obesity, higher blood pressure, higher TG, lower HDL-C, and elevated blood sugar12.

In insulin resistance, the body is often handling more incoming energy than its tissues such as muscle or fat can readily use or store. Insulin signaling becomes less effective at holding fatty acids inside fat cells, so lipolysis increases and more fatty acids spill into the bloodstream and travel to the liver.

Some accumulate in the liver itself — contributing to conditions like fatty liver disease, which may further blunt insulin signaling — while others are repackaged into TG-rich VLDL particles and sent back into circulation.

But the tissues receiving that fatty fuel are often already flush with energy, and thus these metabolic conditions can alter how efficiently these particles are processed. As VLDL particles unload only part of their TG payload, they leave behind remnants that may remain in circulation.

Combined with changes in lipoprotein remodeling that can further impair their clearance by the liver, the result is a larger circulating pool of ApoB-containing particles emerging in the context of an environment consistent with metabolic dysfunction.

Note: if any of the terminology here is unfamiliar, our Glossary of Terms here is a good place to start.

Subscribe now

ApoB may rise for different reasons: moving to lean mass hyper-responders

In some lean people eating a very-low-carbohydrate diet, the proposed physiology may result in a much different path to a given ApoB level. Under the Lipid Energy Model (LEM)3, greater reliance on fat for fuel creates increased peripheral demand for fatty acids.

The liver exports TG in VLDL, those TG are rapidly unloaded and utilized for energy, and the VLDL particles are progressively remodeled into LDL particles. Along the way, surface components shed from the VLDLs may be mopped up by HDL, in turn, increasing HDL-C as TG levels fall.

For more details, see our free LMHR article here.

So, in one setting, ApoB climbs alongside an oversupply of energy and impaired handling of fuel. On the other hand, the LEM posits increased trafficking and use of fat to meet much-needed peripheral energy demand — perhaps adaptive physiology with more ApoB left in the process.

The question this series explores

If the number can arise from such different circumstances, this poses an obvious question:

Does the association between elevated LDL-C or ApoB and cardiovascular disease look the same when the surrounding metabolic milieu looks different?

Plenty of papers hint at answering this question. We will continue to publish in this series and compile papers that let us look at LDL-C and ApoB through markers associated with metabolic health such as TG/HDL-C levels, fasting insulin, some advanced lipoprotein metrics, and more.

We’ll start with well-known metrics: the TG and HDL-C levels.

The Copenhagen Male Study: our starting point

The Copenhagen Male Study4 followed 2,906 men, ages 53 to 74, who were free of ischemic heart disease (IHD) at baseline. Over 8 years, 229 of them had a first IHD event.

Participants were divided into three groups based on their fasting TG and HDL-C:

• Low TG–high HDL-C (TG ≤97 mg/dL and HDL-C ≥57 mg/dL) — a profile typically consistent with insulin sensitivity

• High TG–low HDL-C (TG ≥142 mg/dL and HDL-C ≤46 mg/dL) — a profile typically consistent with insulin resistance

• Intermediate — everyone else

Of note, TG and HDL-C are not always perfect reflections of metabolic health, but they often give useful clues about the current state of a person’s milieu.

And in this particular study, IHD incidence differed significantly across the TG and HDL-C groups.

Men with high TG and low HDL-C had an 8-year IHD incidence of approximately 12.2%, compared with 4.5% among men with low TG and high HDL-C—a ~2.7-fold difference (see Figure 3).

This higher incidence of events with higher TG and lower HDL-C is commonly observed in many data sets, but the more interesting question here is:

What happened to LDL-C in the lower TG and higher HDL-C group?

Higher LDL-C, but similar event rates

Among the men with low TG and high HDL-C, the researchers split participants by LDL-C above and below 170 mg/dL, approximately the median. Over 8 years:

• LDL-C ≤170 mg/dL: 15 events in 347 men — about 4.3%

• LDL-C >170 mg/dL: 9 events in 181 men — about 5.0%

That is an absolute difference of ~0.7% over 8 years.

Put differently, if you followed 1,000 men with low TG/HDL-C for 8 years:

• About 957 out of 1000 men would be expected to remain event-free in the lower LDL-C group.

• About 950 out of 1000 men would be expected to remain event-free in the higher LDL-C group.

That is a difference of about 7 extra events per 1,000 men over 8 years (see Figure 5).

At first glance that looks like a small difference.

But to be fair, there are important limitations to consider.

1. The LDL-C contrast here was relatively modest — below and above 170 mg/dL. We don’t know what the picture would look like with a wider spread, say LDL-C below 100 mg/dL versus above 200 mg/dL or analyzed as a continuous variable.

2. The subgroups also had relatively few events, limiting statistical power and widening the uncertainty around the estimates.

3. This was only men, aged 53 to 74, and nearly all of European descent — so the findings may not be generalizable to other populations.

Thus, a more careful summary may be this:

In this particular subgroup of men with low TG and high HDL-C, the absolute difference in ischemic heart disease incidence between lower and higher LDL-C was generally small.

And it wasn’t only LDL-C

The Copenhagen data also showed something else interesting. Across other well-established characteristics that associate with heart disease like smoking and lower physical activity, a similar pattern was present: men with low TG and high HDL-C generally had lower IHD incidence within those strata as well. The association with hypertension was similar but pointed to something important that we touched on earlier.

Markers of metabolic dysfunction tend to cluster together— a property called collinearity. Of all the men with hypertension in the cohort, less than 14% had lower TG and higher HDL-C — and, of note, this group accounted for less than 5% of all IHD events among hypertensive men.

The insulin-sensitive lipid pattern and high blood pressure were less likely to occur together, whereas higher TG and lower HDL-C tended to associate with higher BMI, diabetes, lower physical activity (and high blood pressure too).

This collinearity is exactly why isolating the independent contribution of any single metric/characteristic like LDL-C or ApoB can be quite challenging.

We are not attempting to draw the conclusion that smoking, hypertension, physical activity or LDL-C are completely irrelevant when TG are lower, and HDL-C is higher, rather the incidence associated with any one characteristic may be different depending on the metabolic milieu in which it operates.

Where this leaves us

The Copenhagen Male Study may be a useful opening because it’s simple and observational — which can be helpful for hypothesis generation. But it leaves a few gaps. In particular, it looked at LDL-C, not ApoB. And conventional lipidology generally considers ApoB the stronger marker for cardiovascular disease.

So naturally, the next question is whether the same phenomenon survives when we look specifically at ApoB.

If ApoB is the better marker, does the metabolic milieu still seem to change how strongly it tracks with heart disease — or does ApoB capture the full association with heart disease independent of the environment?

That’s where the Quebec Cardiovascular Study kicks in, by pairing ApoB with fasting insulin. We’ll pick it up there in next week’s newsletter.

Share

Leave a comment

In Part 1, we looked at a 2023 analysis from the Western Denmark Heart Registry involving more than 12,000 patients with a coronary artery calcium (CAC) score of zero followed for 4.3 years1.

The question was straightforward:

With the authors stratifying by CAC, does LDL-C still associate with cardiovascular events?

What they found:

LDL-C was associated with events when calcified plaque was present, but not when CAC = 0.

In the CAC = 0 group, smoking, diabetes, and lower HDL-C still tracked with events; LDL-C did not. MESA reported a similar pattern over more than 16 years of follow-up.

In that analysis, the “Power of Zero” held strong. See our Cholesterol vs. Calcium article for more details on the power of zero.

But a newer analysis from the same registry has now reported a different result.

A larger cohort with longer follow-up

A newer 2025 analysis2 followed 23,777 patients with CAC = 0 for a median of 7.1 years. The investigators examined how LDL-C associated with two separate outcomes:

• Non-calcified coronary plaque (i.e. soft plaque) on coronary CT angiography (CCTA)

• Future coronary events, including myocardial infarction (MI) and a broader coronary heart disease (CHD) outcome

The broader CHD outcome included MI or coronary revascularization.

The median age was 54, and 61% of participants were women—details that may become relevant later.

Just like the last WDHR study, this was also a symptomatic referral cohort. These were not people who requested imaging out of curiosity. They had symptoms potentially suggestive of coronary disease—such as chest pain, pressure or tightness—and were referred for imaging.

So what did the study find?

Non-calcified plaque was present in ~11% of participants—about 1 in 10.

For every 38.7 mg/dL (1 mmol/L) higher LDL-C, non-calcified plaque was:

• 21% more likely across the full cohort

• 39% more likely among participants age 45 or younger

• 22% more likely among those age 46 to 60

• 11% more likely among those older than 60

Higher LDL-C was also associated with future coronary events.

For every 38.7 mg/dL higher LDL-C, a CHD event was 28% more likely overall and 37% more likely among participants age 45 or younger.

In plain English, people with symptoms of heart disease and CAC = 0 who had higher LDL-C had more soft plaque and more heart disease-related events. Both associations appeared strongest in younger participants.

Taken at face value, that's an interesting finding. With a larger cohort and longer follow-up, this analysis detected something that the earlier Western Denmark study did not.

The question (and the reason this article is worth writing) is not whether the association exists at all. It’s what this association means for an individual and whether these data support the clinical conclusion that followed.

The authors wrote:

“These findings are valuable for clinical practice, as they suggest that lowering LDL-C in younger individuals with hypercholesterolaemia should be considered independent of existing CAC or not to reduce atherosclerosis progression and lower long-term CHD risk.”

Let’s pause here for a moment.

The conclusion goes somewhat beyond what the study itself showed. Because conventional lipidology favors lowering LDL-C early and keeping it low over time, these findings could understandably be used to argue for broader treatment of younger people, including those with CAC = 0.

Likewise, the authors did not specifically define “hypercholesterolaemia” using a single LDL-C threshold, so it may be unclear to whom the conclusion should be applied.

Therefore, a question conventional doctors could ask:

Based on these data, is it reasonable to conclude that virtually every younger person should consider strategies to lower LDL-C, regardless of CAC score?

What may be considered concerning is how quickly a modest observational association can become a broad treatment message.

Before drawing this conclusion, one could consider asking a few questions such as what did the study actually measure, how large was the absolute difference, what may have caused the events, and would lowering LDL-C have prevented them?

We think it is worth looking carefully at what the study demonstrated—and just as importantly, what it did not.

We will start by zooming out from the relative statistics and looking at the absolute outcomes to put these results into perspective.

This is where the paid subscription picks up — if you want to follow the thread all the way through, consider subscribing!

Subscribe now

Read more

Who is Darius Sharpe, and 24 years seeing the sickest patients [1:10]

• Darius Sharpe is an ER nurse, a former paramedic, a father of a nine-month-old, and has been in emergency medicine for 24 years. He is candid from the start about what he is not: not a researcher, not a holder of advanced degrees, not someone who has published a study. What he has is more than two decades of real-world exposure to the sickest patients in civilian life — first as an EMT at 18, then a paramedic for 13 years, and now as an ER nurse at a large California hospital system for eight years. He works in the environment that receives everything: heart attacks, strokes, stabbings, shootings, pediatric codes, and the full spectrum of metabolic disease that drives the majority of 911 calls.

From fire explorer to EMT at 18: why emergency medicine, not firefighting [3:04]

• Darius started as a fire explorer in high school — a career-oriented branch of the Boy Scouts that trains with the fire department — and completed the program intending to become a firefighter-paramedic. Instead, after starting as an EMT at 18 and working in 911 immediately, he found himself far more drawn to the healthcare and medicine side than to firefighting. Part of the reason was the brutal culture of the fire service at the time: the Kelly schedule meant some firefighters were working 24-hour shifts back to back, and mandatory overtime could push people to 72- or 96-hour stretches away from their families.

• He also describes a cultural mismatch — a “very bro attitude” that he did not see himself in long-term. After working as a paramedic for several years and receiving standing offers from fire captains to join whenever he was ready, he made a different call: nursing. California ER nursing offered better hourly wages, more clinical depth, predictable 8- or 12-hour shifts, and the ability to go home every night. He spent 13 years as a paramedic, including time working flex units in parking lots waiting for calls, before finishing nursing school while still on the ambulance. After a mandatory year at a different hospital (an hour commute each way) required before returning to his current ER, he has now been at that hospital for seven years.

Heart attacks do not look like the movies: atypical symptoms and why men especially minimize them [12:07]

• Darius flags one of the most important public health messages he can deliver from inside an ER: the classic Hollywood heart attack — clutching the chest, left arm pain, pale and sweaty, collapsing — may represent as little as 50% of actual MI presentations. He describes a recent patient: male, 30s, morbidly obese, presenting only with left arm and left neck pain and nothing else. No chest pain. No shortness of breath. As a nurse with 24 years of pattern recognition, Darius triaged him immediately and got an EKG. It showed a massive inferior MI. The patient had none of the classic symptoms.

• Men in particular minimize symptoms. The dynamic is exactly what you would expect: they do not want to look like they are making a big deal out of nothing, they do not want to be embarrassed, they reach for alternative explanations. “My wife sent me here because she was tired of me complaining.” Darius’s message is direct: embarrassment kills. Every second counts. People have died not just from the event itself but from delaying getting to the hospital because they talked themselves out of it. The phrase from inside medicine is “time is tissue” — the longer a blockage persists, the more tissue downstream of it dies. You cannot get it back.

“Time is tissue”: stroke symptoms, clot-busters, and the six-hour window you cannot miss [16:10]

• Dave and Darius work through the urgency of stroke recognition in detail. Stroke symptoms to know and act on immediately: sudden onset weakness or numbness in the left arm or left side of the face; facial droop (the smile test — if one side of the face droops when the person tries to smile, that is a stroke signal); sudden inability to find or form words (aphasia); and sudden onset splitting headache or severe dizziness with no obvious trigger. Any of these symptoms requires going to the hospital now, not in three days when it seems like it might be sticking around.

• The reason the window matters mechanically: there is a clot-busting medication called tenecteplase that can break up the blockage and restore blood flow, but it can only be given within six hours of the onset of symptoms. After that, the risk of the medication (it is a systemic anticoagulant and carries a 5–10% risk of causing a brain bleed) outweighs any benefit because the tissue damage is already done. If a patient arrives at the ER 20 hours after symptom onset, this option is gone. Darius’s protocol once stroke is suspected: vitals, neurologist consult, neurological exam to assess severity, CT scan to rule out a brain bleed (since both ischemic and hemorrhagic stroke can produce similar symptoms but have opposite treatments), and if the patient is a candidate and the scan is clear, tenecteplase is given under continuous monitoring for neurological improvement or deterioration. A 5–10% bleed complication rate is a big number in absolute terms — which is exactly why the timing, severity assessment, and pre-medication CT are all non-negotiable steps.

The copy-paste patient: why an 18-year-old on an ambulance saw the metabolic disease pattern before anyone taught him to [24:47]

• Even at 18, working on the ambulance fresh out of EMT school, Darius noticed a pattern. The vast majority of 911 calls were not trauma — they were medical: chest pain, abdominal pain, shortness of breath, dizziness. And the patients were almost interchangeable. They all had hypertension, diabetes, COPD, obesity, and heart disease. They were all on lisinopril, Lipitor, metoprolol, metformin, hydrochlorothiazide. You could copy and paste the medication list from one patient to the next. He remembers asking his paramedic partners as a teenager: why don’t we prescribe a diet for these patients so they don’t have to keep taking these medications and keep coming to the hospital? He never got a good answer. Twenty-four years later, he is still waiting for one.

What actually burns you out in emergency medicine: it is not the dead five-year-old [28:06]

• Darius describes one of the most clarifying things about working in emergency medicine: the situations that are hardest to witness — the pediatric codes, the traumatic deaths, the moments of profound human suffering — are not what burns people out. Those are what they signed up for. He recalls a night shift at a small hospital, five nurses, a dad running in with a limp five-year-old, all hands on deck for an hour of resuscitation, ultimately futile. The sound of a mother screaming about her dead child, he says, never leaves you. Every one of those cases lives in him.

• After it was over, several of them walked outside into the rain. No one said anything. They stood there for a minute and then went back in. The first patient room he walked into: “Where’s my sandwich? I asked for that an hour ago.” That, he says, is what makes you want to quit. Not the tragedy you are there to face — the treatment by people who have no idea what you just went through, because the curtain was closed and they have been sitting in their room for an hour and they are hungry. He is careful not to fully fault those patients either: they do not know. But the structural problem — that a nurse who just ran a code on a child is then immediately back to managing four other patients, all of whose needs built up during that hour, while one of them is yelling about a sandwich — is where the real burnout lives in emergency medicine.

All four grandparents dead by 21, all diabetic: why genetics was always on his radar [40:24]

• Darius grew up with awareness of his genetic risk in a way that most people in their 20s do not. By the time he was 21, all four of his biological grandparents were dead. Every single one had diabetes. His maternal grandmother had type 1 diabetes and died at 62 of a massive heart attack on Christmas morning. The others all had strokes and heart disease. He also discovered during nursing school microbiology that he carries the ApoE 3/4 genotype, adding a possible Alzheimer’s risk on top of the cardiovascular picture. His response: I better be really on my health.

• This shaped a mindset that ran in parallel to his diet evolution: even in his early 20s, when he was still eating whatever he wanted on the ambulance and taking Jamba Juice runs between shifts, he was pulling back on overall intake and watching for weight gain in a way his coworkers were not. He watched people he worked with at 22 and 23 start gaining weight as they hit their late 20s and made a deliberate decision not to let that happen to him. He also, without anyone having told him to do it, was reading ingredient labels from the moment he first started shopping for himself. He thought that was just what you did. He was surprised to discover later that almost nobody did that.

LDL of 179 at age 24 — eating all the carbs, no FH, no awareness that this was unusual [54:37]

• Darius had labs drawn at 24 and his LDL came back at 179 mg/dL. He was eating a standard high-carbohydrate diet at the time — no low-carb anything. His doctor messaged him and said he wanted to talk. Darius thought little of it and never followed up. He did not know at the time that an LDL of 179 at 24, in someone otherwise apparently healthy and with no diagnosed FH, is unusual. He has since confirmed through genetic testing that he does not carry FH mutations. His mother has had high LDL her entire life; a doctor started her on a low-dose statin about five years ago and she is now 72, mobile and active, never having had a cardiovascular event, though she does have macular degeneration that has progressively worsened.

• The LDL pattern over time, as tracked in his personal spreadsheet: 179 at age 24 (eating all carbs). By age 30, he thinks it was around 220, with his triglyceride/HDL ratio having worsened slightly (HDL dropping into the 60s, triglycerides into the 90s). By the time he was in nursing school and had found CholesterolCode, it was over 350 — the lab just flagged it as “greater than 350” because it did not go higher.

Nursing school glucose check: 126 at a table of classmates all in the 70s and 80s [1:01:14]

• At 32, in nursing school, Darius’s class went to a facility to practice using glucometers on themselves before their clinical rotation. Everyone sat around the table and checked their own blood sugar. The readings came back: 75, 92, 83. Darius checked his: 126. No one else at the table was anywhere near that number. He had had a smoothie on the way in — whole milk, natural peanut butter, no-sugar-added, protein powder, frozen strawberries, blueberries, and banana. Objectively reasonable-sounding by conventional standards. No added sugar. Just fruit.

• His reaction was, notably, not alarm. It was more of a “huh, that’s odd.” He was planning a wedding, working two jobs, and in nursing school simultaneously. He noted it and moved on. What makes this a meaningful data point in retrospect is not just the number itself but the context: he was not eating something obviously junk. He was eating what most dietitians would call a healthy breakfast. And his blood sugar at a table full of peers of similar age reading entirely normal numbers was 126.

The Toll House cookie dough semester: eating his feelings during nursing school’s final stretch [1:04:45]

• After his engagement fell apart in the third semester of nursing school, Darius went through a period he describes plainly as eating his feelings. He went to Costco and bought the five-pound bucket of Toll House chocolate chip cookie dough. He ate it over roughly three to four weeks — sometimes baking it, sometimes eating it straight from the tub. He was not exercising. He was not tracking anything. He gained weight up to 192 lbs (he is 5’11” and normally runs around 178–180). He describes knowing exactly what he was doing and not caring: “You’re not eating well. You’re not exercising. You’re going to gain a little weight. It’s fine. Once you’re done with this, you’ll get back on track.”

• He connects this to what Dave’s wife Sharon calls “eating your feelings” — a more socially acceptable form of self-soothing than alcohol or drugs, actively encouraged by a food environment that surrounds you with the message that you deserve it. He is self-aware enough to name exactly what he was doing in the moment. It did not stop him from doing it. Once nursing school ended, he did what he had told himself he would do and got back on track.

Genius Foods, Max Lugavere, and the gravitational pull toward low carb without trying [1:08:22]

• In early 2018, finished with nursing school and looking for his first nursing job, Darius came across a Facebook comment recommending Max Lugavere’s book Genius Foods. He listened to it on audio and describes it as the first time someone laid out a clear, evidence-referenced case for why diet quality mattered beyond just calories. Lugavere was not extreme — he was not prescribing keto or carnivore. He was saying: eat grass-fed beef, eat broccoli, use olive oil, watch the carbs generally, get away from packaged food. Straightforward whole-foods guidance.

• But when Darius implemented it, he found himself drifting toward something around 100 grams of net carbs per day without anyone telling him to hit a specific target. The weight dropped effortlessly. He also notes he had been practicing occasional 24-hour fasts since his mid-20s after reading about it in a Men’s Health article — well before intermittent fasting became a widespread topic — which gave him some baseline familiarity with not eating. He was not tracking closely, still eating fast food occasionally when out, but his home diet had become meaningfully lower in carbohydrates and higher in whole-food quality.

The In-N-Out experiment, night one: blood sugar of 256 an hour after a burger, fries, and root beer [1:12:26]

• In 2019, Darius had been eating generally low-carb at home but was still loose when eating out. He got off a long hospital shift, felt like In-N-Out, and treated himself: a double patty burger with bun, fries, ketchup, and a root beer. About 10 minutes into the hour-long drive home he felt awful — a head rush, just felt terrible. He had a glucometer by then. He got home an hour after finishing the meal and checked. 256 mg/dL.

• Dave notes that this is exactly the data point where context transforms the number: Darius was a paramedic and now a nurse who had spent years seeing patients in the ER with glucose readings in the 200s and routinely chastising them for not managing their condition. “Dude, what did you eat? Your glucose is in the 200s. You’re clearly not taking care of yourself.” And here he was, staring at 256 from a single standard American meal that almost everyone eats regularly. The morning fasting glucose the next day: high 80s. Normal. It was not a diabetes diagnosis — it was a postprandial spike. But it was a profound one.

Four In-N-Out experiments, progressively removing the carbs: from 256 to 105 [1:18:49]

• What followed over the next several weeks was a self-designed elimination experiment, repeated while keeping the post-work timing and the hour-long drive home constant (effectively a natural one-hour postprandial measurement window). Each iteration removed one source of carbohydrate:

• Experiment 1 (baseline): double patty burger with bun, fries, ketchup, root beer. Blood sugar after 1 hour: 256 mg/dL.

• Experiment 2 (remove soda): same burger with bun, fries, ketchup, water. Blood sugar: 205 mg/dL. Better, but still clearly elevated.

• Experiment 3 (remove bun): protein style lettuce-wrap burger, two patties, fries, ketchup, water. The sauce remained throughout all experiments (the thousand-island-style spread — containing sugar — was never removed). Blood sugar: 156 mg/dL.

• Experiment 4 (remove fries): two lettuce-wrapped burgers with two patties each, sauce, water. No fries. Blood sugar: 105 mg/dL.

• The arc from 256 to 105 by removing the sugar drink, the bread bun, and the fried potatoes from a standard American fast food meal — without removing the beef, the fat, or even the condiment sauce — became a foundational piece of Darius’s understanding. As he put it: “This is probably the way you should be eating, Darius.”

Share

Subscribe now

Joining the LMHR Facebook group in 2019: one of the original members [1:21:45]

• After the In-N-Out experiments, Darius went deep into metabolic health YouTube and eventually found CholesterolCode and Dave’s work. He joined the Lean Mass Hyper-Responder Facebook group in 2019 — the group had been founded only in the summer of 2018, making him one of its earliest members. His first post was a shirtless photo (the group is named after a lipid phenotype, and many of the original members posted their labs and physique together). He clearly met the phenotype: the triad of LDL above 200 (his was over 350), HDL above 80, and triglycerides below 70, occurring together in a lean, metabolically healthy, fat-adapted individual.

• Dave notes that Darius later became one of the most prominent and consistent voices in the group, and that when selecting someone to speak at the first COSI (Citizen Science Initiative) conference, part of what made Darius stand out was his evenhanded approach: he was clearly part of the LMHR community but was not dismissive of the legitimate questions around what very high LDL means for individual cardiovascular risk. He represented the thoughtful version of the community rather than the dogmatic one.

The ER during 2020 to 2022: what metabolic illness looked like in the context of COVID [1:29:06]

• Darius is careful about how much he says on this topic given YouTube’s content policy (two previous episodes were removed for reasons he never fully understood), but his clinical observation is direct: the patients who got seriously ill were the metabolically sick ones. Not by age necessarily — a 70-year-old with no metabolic problems did better than a 45-year-old with obesity and diabetes. He watched this pattern repeat case after case. Two communities he treated were particularly severely affected: the Hmong community and Marshallese populations in his area, both of which carry extremely high rates of obesity and diabetes. Dark humor in the ER: when a Marshallese patient came in, they knew it was going to be bad.

• He describes one case that stayed with him: a family where three members had already died and three were hospitalized simultaneously. One morbidly obese patient refused intubation despite oxygen saturations in the 70s on BiPAP, having heard warnings from community members not to let the hospital intubate. His sister was in the same ER as his patient. Darius wheeled the sister’s bed across the ER so she could talk to her brother before he was intubated. He did not survive the week. Dave and Darius share their frustration at the missed opportunity: that period had the strongest real-world metabolic health signal many of them had ever seen, with disease severity tracking almost perfectly with metabolic status, and public health messaging never named it as something people could actively address.

• A secondary consequence Dave raises: people who needed cancer screenings, who had symptoms they should have followed up on, who were due for regular monitoring — millions of them stayed away from hospitals out of fear and missed early detection windows we will never get back.

Getting more fit during COVID: outdoor workouts, paddleboard, and the hospital as a ghost town [1:41:35]

• While many people gained the “COVID 19” pounds when gyms closed, Darius went the other way. He found a park overlooking a marshland near his home and started working out there. He bought a paddleboard. He was doing more outdoor exercise than he had been doing before. The hospital, at least initially, was nearly empty — people were terrified of hospitals and stopped coming in even when they needed to. There was no overtime. He was working regular shifts and had extra time and energy to exercise. By the time things returned to normal and the hospital filled back up with the wave of serious illness, he was in better shape than when it started.

First CGM in 2023: a dinner roll sends blood sugar to 220 [1:43:25]

• Darius got his first continuous glucose monitor in 2023, after attending Low Carb Denver when conferences started returning. One of the first experiments: at a friend’s barbecue, he ate a single dinner roll by itself and watched his CGM. His glucose went to 220. One dinner roll. Not a meal. Not a soda and a burger. A single bread roll. This is where the CGM became more than a curiosity and started becoming a tool for understanding his individual glucose physiology in real time, independent of any particular dietary framework.

• He describes a period of repeated discovery as he tested different foods. His general baseline: eating loosely low-carb at home, more permissive when eating out, never strict. Even so, his CGM showed him how carbohydrate-sensitive he was at an individual level. Not only did bread and sugar products spike him dramatically, but the same food cooked differently would produce different results — raw broccoli barely moved his glucose, lightly steamed produced a small bump, well-boiled produced a meaningful rise. The cooking method alters how available the carbohydrates are for digestion.

The keto week vs. high-carb week OGTT experiment: 250 going in, 190 coming out [1:51:10]

• Darius ran a structured self-experiment using Own Your Labs for the oral glucose tolerance tests and his CGM throughout. He spent one week eating strict keto (under 25 grams net carbs), then one week eating high-carb (roughly 60% of calories from carbohydrates), with a 75-gram oral glucose tolerance test at the end of each week conducted under the same conditions.

• At the end of the keto week, the OGTT produced a glucose spike to approximately 250 mg/dL, with a subsequent dip into the low 60s (reactive hypoglycemia on the CGM, but asymptomatic — he felt nothing because he had adequate ketone substrate available for the brain). Dave notes this is entirely expected: after a week of strict keto, the body has downregulated insulin-dependent glucose disposal pathways and the sudden 75-gram bolus produces an exaggerated glycemic response, a well-documented phenomenon called physiologic insulin resistance. It does not mean he has broken something; it means his metabolism has genuinely shifted its substrate preference.

• After the OGTT he drove across the parking lot to his favorite Indian restaurant and ate a full meal. His glucose spiked to approximately 200 on top of the OGTT recovery. Then came the high-carb week. He ate roughly 60% of calories from carbohydrates but actively tried to mitigate spikes — protein and fiber first, carbs last, post-meal walks. Even with mitigation, he was regularly seeing readings in the 150s and 180s throughout the week, with baseline glucose running 10–15 points higher than his keto week baseline. At the end of the high-carb week, the OGTT produced a peak of approximately 190 — better than 250, but still too high by his standard.

• His takeaway: during the keto week, his CGM trace was nearly flat with only one glucose excursion (from an exercise-induced glycogen dump during a timed effort). During the high-carb week, regular spikes over 150, elevated baseline, and feeling subjectively worse — while eating equal calories. He does not think that eating more carbs to lower the OGTT spike (the concept of “carb loading before a glucose tolerance test” to improve the result) represents a meaningful health improvement when the overall glucose picture was dramatically worse throughout that week.

Net carbs versus total carbs: why Dave is moving away from the net carbs framework [1:59:38]

• Dave raises a definitional point he has been rethinking: the net carb convention (subtracting fiber from total carbohydrates) may be giving people too much permission to eat foods whose glycemic impact is higher than the net carb count suggests. Two reasons he is moving toward total carbs. First, fiber is not necessarily metabolically inert — even insoluble fiber, once it reaches the hindgut in bulk form, can activate GLP-1 and other incretin responses and produce measurable insulin effects. Second, and more practically as demonstrated by Darius’s broccoli experiment: the same food cooked differently has very different glucose impacts, and the net carb label on a package does not tell you how the fiber in that food will behave given your specific preparation.

• His working proposal: low-carb could be defined as 120 grams or less of total carbs (which would roughly correspond to under 100 grams net for most diets); keto as under 30 grams total (which he considers the threshold at which most people who are not exercising heavily will see measurable ketones). He also raises the “keto label” problem: family members who tried keto using products labeled as keto-friendly saw their glucose and A1C worsen, partly because these products exploit the net carb loophole with resistant starch and sugar alcohols that have a real glucose impact in sensitive individuals.

Defining low carb, keto, and how poor study definitions make research on these diets nearly meaningless [2:00:42]

• Darius and Dave align on frustration with how loosely these terms are used in the research literature. Darius’s working definitions: low carb as under roughly 20% of total calories from carbohydrates; keto as producing measurable ketones, which for him personally means his blood ketones generally sit between 0.1 and 0.5 mmol/L almost continuously while eating low-carb, though he rarely crosses above 0.5 unless fasting or eating a large fat bolus. Dave pushes back on using ketone levels as the primary marker of how “in ketosis” someone is, for the same reason you would not use blood glucose levels as the primary marker of how much glucose the cell is actually burning — the circulating substrate level tells you about what is in transit, not what is being utilized. Respiratory exchange ratio (RER), the actual ratio of carbon dioxide produced to oxygen consumed at the cellular level, is a much better measure of whether fat is being burned, but it requires specialized testing.

• The literature problem, illustrated with an example Dave raises: he has seen studies claim to test a “low-carb diet” at 46% of calories from carbohydrates. He checked in Chronometer: a diet of nothing but pepperoni pizza has a carbohydrate ratio of around 42%. Lower than the study’s “low carb” group. The fundamental question a study should ask before labeling a dietary intervention: would the community that actually adopts and advocates for that diet recognize what you’re calling it as their diet? The answer, for most “low-carb” research using 40-plus percent carbohydrate ratios, is no.

The calories vs. carbohydrate insulin model debate: both sides are partially right, and the online version of this argument is exhausting [2:19:37]

• Darius goes on a short rant that Dave finds entirely reasonable: the social media war between “calories in, calories out is all that matters” and “insulin is everything and you can eat as many calories as you want as long as they’re keto” is stupid. He has run the experiments. If he eats too many Keto Bricks, he gains weight. Calories matter. This is not controversial. At the same time, for the people insisting hormones have nothing to do with weight: go tell the women in your life that their hormones do not affect their weight and they just need to eat less. Do not be holding any sharp objects when you say it. Insulin matters. Bodybuilders inject exogenous insulin to gain mass while also eating excess calories. Both things matter simultaneously.

• Dave steelmans the nuanced carbohydrate-insulin model position: it is not that energy balance is wrong, it is that energy balance is tautological as a model — you cannot falsify it because “overeating” is defined by whether weight was gained, which makes it circular. The carbohydrate-insulin model proponents’ real claim is that hormonal environment drives appetite, satiety, and how much spontaneous eating occurs, which makes the caloric intake downstream of the hormonal state rather than the primary input. The best current evidence for this being a real and meaningful mechanism: GLP-1 agonists demonstrably reduce appetite through a hormonal mechanism. Hormones drive how much people eat. Dave is careful to stay nuanced: some people on GLP-1 genuinely use it as a bridge to better habits; others use it to continue the same habits with smaller quantities of the same food, which he sees in his family as a better-than-nothing but not ideal outcome.

Hospital food: where one hospital system is doing better than the running joke suggests [2:28:44]

• Dave puts a pin in hospital food earlier in the conversation and Darius delivers on it. His view: postprandial glucose is the most important period for metabolic health and tissue healing, and hospitals almost never measure it or account for it in dietary decisions. A patient recovering from a MI who gets a stack of pancakes or a soda or juice boxes is having their postprandial glucose driven to levels that actively impair tissue healing and increase clotting risk — exactly what you do not want in the acute post-MI period. The data on this is clear: glucose variability and elevation in the hospital acute care setting predicts worse outcomes.

• That said, Darius gives credit to the hospital system he currently works at, which he says actually serves reasonably good food: bacon and eggs in the morning, grass-fed meatloaf with broccoli at dinner, salads at lunch, meals labeled with carbohydrate counts so a nurse can pull two trays, remove the starch from one and double up on the protein on the other, and give the modified tray to a diabetic patient trying to control their glucose. He calls out whoever pushed for that within the system, because it is meaningfully better than what he has seen elsewhere.

The juice problem: diabetic patients being given the exact substance used to treat hypoglycemia [2:35:20]

• Darius’s specific clinical frustration, the one that makes him “want to strangle people”: walking into a diabetic patient’s room and finding four juice boxes sitting next to them, their glucose at 350, because a tech or another nurse brought them over. Juice is the substance ER nurses give patients who are hypoglycemic to raise their glucose fast. If you give that same substance to a patient with normal glucose, it raises it. If you give it to a patient whose glucose is already elevated, it raises it further. This should not require explaining to healthcare workers. And yet it happens constantly.

• He recounts a patient who had been discharged from the hospital the previous day for hyperglycemia and came back the next day with glucose in the 400s. He asked what the patient had consumed since discharge: some chicken and salad, four large glasses of juice, and a gallon of milk. The wife: “But it’s 100% juice.” She was not wrong that it was juice. She was taught by the hospital that juice was something appropriate to give to a diabetic patient — because the hospital was giving it to him during his stay. The system modeled the behavior. When the family repeated it at home, it sent the patient back to the ER. Darius and a couple of other nurses are putting together a formal project to propose eliminating juice from their hospital, keeping only a minimal supply for genuine hypoglycemia management. He wants to present it live, drinking a glass of juice at the beginning of a presentation with his CGM on and showing what happens to his glucose over the next 30 minutes in real time.

Bedside patient education with CGM data: real-time glucose as the most powerful teaching tool [2:44:26]

• Darius describes why he still loves his job despite the frustrations: he gets to encounter patients at the moment of maximum receptivity. When someone comes into the ER in DKA, they know something went badly wrong. He can sit with them, pull up his CGM on his phone, show them what his glucose looks like when he eats junk versus when he eats well, and give them an efficient, non-preachy version of the metabolic health message: eat meat, fish, eggs, green vegetables, a little whole fruit, some dairy, some avocados and nuts, and stay away from the things that light up your glucose. He does not need to go into lipid hypotheses or LMHR phenotypes or ApoB debates. The CGM data, shown in real time to a patient who is in the middle of experiencing the consequence of not managing their glucose, does more than any lecture could.

• His specific recommendation to patients who ask what to do next: get a CGM. Stelo is his current go-to recommendation because it does not require a prescription and is accessible over the counter. His framing to family members who resist dietary change: “Don’t convince me. Convince your blood work.” Let the device show you what your body does with different foods, in real time, in your specific physiology. That individualized real-time feedback is, in his view, the single most powerful tool available to someone who genuinely wants to understand their own metabolic health.

Leave a comment

A recent video works through the KETO-CTA preprint we posted in January and lays out six “major problems,” building to the conclusion that the manuscript is “fatally flawed, statistically indefensible, and actively misleading.”

In this article:

1. We’ll discuss what a preprint is and why it matters

2. How AI has changed reviewing papers and how that applies here

3. We’ll evaluate the merits for each of the six items John outlines

4. We'll pull these together into a summary Risk Matrix

5. We’ll compare John’s standards against studies he’s reviewed more favorably, particularly the study he chose as a benchmark in his previous video on our research

6. I’ll give two respectful challenges

7. I will then give my final thoughts

Preprints are expected to have rough edges (and that’s part of the point)

Preprints are posted before formal peer review and should be understood as provisional. They may contain errors, omissions, incomplete reporting, or areas needing clarification.

Comparisons of preprints with their subsequently published versions suggest that reporting often improves and that additional data or analytical content may be added during revision, although most published preprints retain their central conclusions (Brierley et al., 2022). That does not make any individual error unimportant; it means only that identifying and correcting such issues is an expected part of the process.

By sharing work at an early stage, authors invite the broader scientific community to scrutinize findings, spot issues, and offer input that can strengthen the research long before it reaches journal reviewers. This is why I’m emphasizing the timeline for those unfamiliar with what preprints are.

Our research is controversial, but that’s all the more reason to maximize transparency for even the toughest of critics to weigh in — even before formal review. The catch is timing: if a critique arrives months after posting (in this case, John’s is arriving six months later), the manuscript is usually already in the formal peer review process — as ours is now — which is exactly when it’s appropriate to work the revisions through the journal rather than post a running series of public versions.

That said, I’m going to try and thread the needle of responding to what degree I can, but not disrupt the peer-review process we’re in now.

AI has changed everything, including how we review papers

Thanks to AI, we’re far better at spotting errors in papers than ever before.

Give a modern AI a published paper and some patience, and it will often flag mislabeled statistics, numbers that don’t match between text and tables, incorrect figure axes, and/or contradictory disclosures, often catching things human reviewers miss. Many of these minor issues were once quietly spotted by a few, but then ignored as harmless “easter eggs.” That assumption no longer holds—such problems are now detectable at scale on both new and old papers.

Errors are still errors, large or small. However, the amount of scrutiny applied determines what surfaces. Heavier checking on one paper than another can make the first look worse, not because it necessarily is, but because it was examined more closely — and this capability is multiplied dramatically with AI. But fair comparisons require the same level of scrutiny on both.

It’s worth noting research authors are interested in using AI, but there’s a reason many can’t simply hand the whole job to AI up front: institutions and journals set policies — still shifting month to month — on how much AI is permitted and for what, and much of this work involves sensitive patient data that understandably can’t be fed into these networks at all. So the tooling that catches these errors after the fact often isn’t the tooling an author was cleared to use while writing. (As an engineer, I have so much to say on how I feel about this — which I’ll be doing in my book.)

To be clear about my own use: I used these tools to pressure-test this response — and NATURE-CT — and checked every factual claim and quotation in this article against its source. That’s a separate exercise from the preprint’s original analysis.

John’s Six “Major Problems”

John outlines six items he considers “major problems.” I want to steel-man John’s positions as much as possible, including giving him credit where it is due.

However, I would likewise hope John would agree the standards he's calling out should apply consistently, particularly for studies he’s reviewed favorably. I’ll have more to say on that later.

How to size a problem

Okay, but before addressing the six issues John discusses, how can we quantify how “major” — or minor — a problem really is?

Let’s focus on one small example — one of John’s own from this video.

Early on, he makes a fair point: a number of participants in our study actually regressed; their plaque went down:

“Also, despite the problems, this study is interesting, and the dataset could be valuable, and we actually do see some participants with plaque regression. However, it’s only over 1 year, and we don’t know how this will play out long-term.”

John then animates a yellow band on our chart to highlight the regressors. This is the screenshot taken from his video:

But the band is in the wrong place. Regression means a change below zero — below the dashed line. The top of his yellow band sits about two units lower, so it encloses 7 of the 15 people who regressed and leaves 8 out.

Is this a serious error? Is it worth making a big deal about? I don’t think so — and I would imagine most others would agree.

Stating “the band leaves out over half of the regressors in our study!” is technically accurate, but (1) almost no one watching would have come away with a different understanding because of it — his stated point was that some participants regressed, and some did; and (2) correcting the band leaves that statement where it was, since he says “some participants with plaque regression” rather than specify it was 15 (in the QAngio analysis).

These two considerations — probability of impact and severity of impact (and variations of them) — are often plotted as the axes of something known as a Risk Matrix. It’s a great way to size up how big or small a problem is.

Probability of Impact is the chance the error actually changed what a reader came away with — not whether it’s visible, but whether it moved anyone’s understanding.

Severity of Impact is how much the finding itself shifts once the error is corrected.

An error sitting on page one can have a low probability of impact if no one is misled by it, and an error that misleads everyone who encounters it can still be minor if correcting it leaves the conclusion where it was.

Some simple examples at each extreme:

• Low probability, low severity — a figure caption that misstates the follow-up interval while the title, abstract, and Methods all state it correctly, so almost no one is misled — and correcting it leaves every conclusion in the paper where it was.

• Low probability, high severity — an error in a secondary analysis that few readers will ever reach, but which reverses that analysis’s direction once corrected.

• High probability, low severity — a summary statistic misstated in the abstract, which nearly every reader absorbs and carries away wrong, while every conclusion drawn from it still stands once it’s corrected.

• High probability, high severity — a headline finding built on a third-party analysis that doesn’t hold up: everyone who reads the paper relies on it, and correcting it overturns the paper’s main claim. (We’ll come back to this example.)

So to return to John’s misplaced yellow band: it’s a real error — but a small one. Its probability of impact is very low — a viewer who caught the band and a viewer who didn’t both come away with the same thing, which is that some participants regressed — and its severity is minor, since it has only a little weight-bearing on his statement as illustrated. So it would be placed on the bottom row, in the second square starting from the left — Probability of Impact: Very Low, Severity of Impact: Minor.

John’s Problem 1 — “Misdefined LDL-C ‘Exposure’ variable”

“A cumulative exposure should represent LDL-C over time, essentially area under the time curve… So calling follow-up LDL-C exposure makes it sound much more rigorous and more informative than it actually is.”

Per my note above, and not to take anything away from John catching it independently: this has already been raised in peer review several months ago. A reviewer flagged the same point — that baseline-plus-change shouldn’t be labeled as cumulative exposure — and on review, we agreed and had already addressed it in an initial revision.

Notably, we had already run models on each variant — baseline, follow-up, the change, and the two-visit average — all agreeing with each other.

I’d rate this item low on probability of impact, minor on severity.

Low on probability not because the label is buried — it appears throughout — but because nothing in the analysis depends on reading it as a lifetime measure. The posted formula reduces to follow-up LDL-C, so a reader who takes the name at face value and one who re-derives it land on the same number. Not on the same idea of what “exposure” means, though. John is right about that, and it’s why the label is already changing in revision.

Minor on severity because the variable we actually fitted doesn’t change, and because the absence of association with plaque holds across every variant we tested.

John’s Problem 2 — “Fabricated IQR Metric”

There are two separate claims here.

“They calculated the median plus or minus half of the IQR, which forces the interval to be symmetric around the median… making the middle 50% of the plaque progression look lower than it actually is.”

This first one is correct: the figure caption says the whiskers are the interquartile range, and they are actually the median plus or minus half the IQR. That is a labeling error in the preprint.

And he later states…

“This median plus or minus half of the IQR is not a standard metric. Nobody uses this metric. It does not exist. It is not taught anywhere… this metric has to be custom programmed.”

This second claim — that the quantity does not exist and is taught nowhere — is not correct.

The half-width he’s describing — half the IQR — is the semi-interquartile range, also called the quartile deviation: a standard, named statistic. (Centering it on the median to draw the whiskers isn’t the same as plotting Q1–Q3 — I’ll come to that — but the quantity itself is real and named.) This is a measure of spread defined in foundational statistics texts (Yule & Kendall, An Introduction to the Theory of Statistics; Spiegel & Stephens, Schaum's Outline of Statistics, §4.4).

• Yule & Kendall, An Introduction to the Theory of Statistics — defines the semi‑interquartile range / quartile deviation as ½(Q₃−Q₁) (verified, pp. 147–148).

• Spiegel & Stephens, Schaum’s Outline of Statistics, §4.4 — same definition, standard modern teaching reference (verified, 6th ed.).

• There are also many modern online resources when simply searching “semi‑interquartile range / quartile deviation”. (Below is via Statistics How To)

Moreover, it’s worth noting that we are extraordinarily transparent with the data, with every point represented on the graph. In other words, the data itself already conveys the spread quite effectively — to the point where both kinds of whiskers are much less valuable, visually.

Now to the scoring. To be precise: the whisker half-width is the semi-interquartile range — a named statistic, so “it doesn’t exist” is wrong. What’s wrong is the label. Median ± SIQR isn’t the interquartile range, and on right-skewed data the two come apart: our lower whisker sits at about −2.2 while the 25th percentile is about +1.0. A reader who takes “IQR” at face value would put a quarter of the cohort below −2.2, and that isn’t where the data sit.

So there are two ways to square it, and only one of them is a correction. We can label the whisker what it actually is. Or we can plot Q1–Q3 instead — the more typical convention, and what we’re considering for the revision. The second isn’t repairing a bad statistic; it’s switching to a more expected one.

Either way the finding doesn’t move, because we plot every individual point — the real distribution is visible regardless of the whisker. A reader looking at that figure still sees the spread the data actually have, whether or not they read the whisker as an IQR. I’d rate it very low on probability of impact, negligible on severity.

John’s Problem 3 — “Missing Results and Explanations”

“They do not report the ApoB models anywhere...”

Correct. The ApoB model coefficients are not reported in the supplementary tables. ApoB is named in the conclusion, so that model belongs there, and we're adding it. This one is John’s catch, and credit to him for it.

This I consider the highest of the items John identifies on the risk matrix, with a moderate probability of impact and a minor severity. I rank its probability highest of the five because ApoB is named in the conclusion as a headline result, and a reader who goes to the supplementary tables looking for that model and doesn’t find it is left genuinely uncertain what it showed. That’s a real gap in what a reader can come away with, in a way the subtler labeling issues elsewhere are not.

But the severity stays minor because the problem is that the model wasn’t displayed, not that it was wrong or absent: the fix is to report the models fully — coefficients and confidence intervals, ApoB included — which we’re doing in revision; shown that way, that model tells the same story as the other lipid measures — no association with plaque change. The fix fills in a table — it completes what a reader can check, without moving the finding.

John’s Problem 4 — “Incoherent Equivalence Testing”

This is the most technical of the six, and I’m going to try to be careful and precise. That said, I’m ultimately deferring to our statistician on this one to expand on in a later piece (he’s currently unavailable due to prior commitments).

Let’s start with what John gets right. He raises two things here. The one I can settle now is the reporting inconsistency; the other — whether a single equivalence bound can carry across models whose slopes are in different units — is a real question, and it belongs in the statistician’s write-up rather than in a summary from me.

The supplemental analysis includes an equivalence test — a “two one-sided tests,” or TOST, procedure — and the way it specified its confidence setting in the Methods doesn’t follow the usual convention, and it doesn’t properly match the Discussion, which describes it as a 90% interval. That’s a real inconsistency, and we’ll be correcting this in the revision, with credit to John for catching it.

But here’s what matters for the finding: the equivalence test was never what the finding rests on. The manuscript introduces it with the words “to further support” the result — it’s a backstop, not the foundation. The result itself comes from ordinary linear regression, it holds across two independently-operated analysis platforms applied to the same scans, and it’s backed by Bayes factors of roughly ten to one in favor of no association. None of that depends on the equivalence test.

So the fair question underneath John’s point is: do the conclusions change with the correctly specified settings? At the package default (ci = 0.95, which yields a 90% equivalence interval, versus the 80% the preprint produced) — no, they don’t. Do they change with stricter settings (ci = 0.99 → 98%)? On the exposure variables the conclusion is actually about — ApoB exposure and total LDL-C exposure — the answer holds at both.

Ironically, there is one place the stricter setting changes a verdict. It isn’t on the exposure variables; it’s on a secondary model of the raw change in LDL-C, where the estimate is slightly negative — more LDL-C change tracking with marginally less plaque, not more. In the models and reruns described here, at no setting does a lipid measure turn into evidence associating higher lipids with more plaque.

In other words, the conclusions of the paper remain the same.

On the risk matrix, then: the correction is real, so it isn’t nothing — but the probability of impact is very low and the severity is minor. Very low probability because the finding never leaned on the equivalence test: a reader who took the TOST as written came away with the same conclusion as one reading it corrected. Minor severity because the fix leaves a supporting analysis supporting. Very low probability of impact, minor severity.

John’s Problem 5 — “The ‘Reassuring’ Power Justification”

“They use a post hoc or after the fact power calculation… A post hoc power calculation cannot turn we failed to detect an effect into there was no meaningful effect.”

It’s a narrow point, but one I’d like to grant — running a power calculation after the fact can’t convert “we didn’t detect an effect” into “there is no effect.” I think this is good feedback on the specific wording nuance.

But the null here doesn’t rest on power. It rests on two things the power calculation has nothing to do with: the point estimates for the lipid–plaque association sit near zero on both imaging platforms, and the Bayes factors run about ten to one in favor of the null over an effect. Under the specified prior, those Bayes factors favor the null by about ten to one — evidence for no association under that model, not proof that any effect is absent.

Or to state it a bit more plainly: remove the post-hoc power calculation entirely and the result is unchanged.

I’d rate this low on probability of impact and minor on severity — the power calculation is a supporting sentence, not a load-bearing one, so a reader who accepted it and a reader who struck it come away with the same finding, and taking it out changes the paper’s wording, not its finding.

John’s Problem 6 — “A Signal-Weakening Pipeline Built on Fragile Models”

“The models are crude, univariable simple linear regressions… they do not even adjust for basic confounders like age and sex.”

This one is different in kind from the other five. It isn’t a single finding I can check and correct — it’s thirteen separate statements about the study, stacked into one conclusion: that the design, the measurements, and the models each drain signal, and that what’s left is noise rather than a result.

The other twelve

Most of the thirteen name something factual. We did recruit through social media. There is no control group. The sample is a hundred people. The interval is one year. The plaque data are right-skewed. I dispute none of that.

But naming a feature of a study isn’t yet an argument about its result — for that, the feature has to be shown to push the association toward zero, and to push it hard enough to erase an effect of the size the prior literature predicts.

John claims the first — that these features drain signal. But he doesn't establish it, and doesn't reach the second.

Take the sample size. A cohort around our size over a year is ordinary for longitudinal imaging studies — a description, not a distinction. The question is whether it’s enough for the effect being looked for. The prior literature puts the LDL-C–plaque-change correlation between 0.4 and 0.6. The sample size itself was set in advance — 100 participants, assuming 15% dropout, against the plaque-change variability we expected. The figure people quote back at me, that ~80 participants would suffice for a correlation as low as 0.3, is labelled in the paper as not part of that original calculation, and I’d rather say so than let it read as prospective — it’s the same after-the-fact reasoning I just granted John under Problem 5. What it does establish is arithmetic, and arithmetic doesn’t care when it was run: at the effect size the prior literature predicts, a cohort this size isn’t underpowered. Being told the sample is “small” is not the same as being shown it was too small for the effect anyone claims is there.

Or take the recruitment. Social media recruitment produces an unusual sample, and ours is unusual — but it’s unusual in the exposure we selected on, not in the outcome we measured. Choosing participants by their lipid phenotype doesn’t, by itself, bend the direction of the relationship between lipids and plaque, though it does narrow the range we can observe it over. Choosing them by their plaque would.

The same gap shows up in the claim that our “LDL-C range is restricted to high and extremely high.” Our baseline LDL-C runs from 49 mg/dL to 591, with a standard deviation of 84.7 mg/dL — to date I haven’t found another longitudinal imaging study with a wider spread. And our Discussion already made the point that the observed values span a wide range. A distribution with a floor of 49 isn’t restricted to the “high” end.

The univariable modeling

The one item I do want to unpack — because it isn’t covered in the previous five, and because I think it has the most relevance of anything in the list — is the univariable modeling, and the absence of an age- and sex-adjusted analysis.

Reporting crude, univariable associations is a standard, accepted approach for a descriptive, hypothesis-generating analysis — not a causal test, and not a claim that lipids don’t matter. It’s also worth being precise about what an unadjusted model actually risks. It’s a strong reason to distrust a positive finding, because a confounder can manufacture an association that isn’t there.

This analysis reports an absence, and for a missing adjustment to have produced a false absence, a confounder would have to be actively hiding a real relationship — moving with LDL-C in one direction and with plaque in the other, hard enough to cancel it out. Age is the obvious candidate, and age tracks positively with plaque progression, so for age to be masking a lipid effect it would have to run negatively against LDL-C in this cohort. Nobody has proposed that.

And we don’t have to argue it, because we ran it. Adding age and sex, then baseline plaque, leaves the lipid coefficient non-significant (p ≈ 0.6–1.0). The crude and adjusted analyses agree: in this cohort, lipids didn’t track with plaque. Those numbers are going into the revision so nobody has to take my word for it.

That’s also why Problem 6 isn’t on the matrix. The grid scores errors, large or small — how likely one is to have changed what a reader came away with, and how much the finding moves once it’s corrected. A modeling choice you’d argue differently isn’t an error, and there’s no correction to plot — no revised number to set beside the original. I’ve kept it on the legend with the reason attached rather than quietly dropping it, because the note itself is worth taking.

I’m glad to take it to the team, as with the previous items. If we adopt it, it’ll be because we concluded it was an improvement on its own merits.

Putting the Risk Matrix Together

Here's where five of the six land, scored the same way — Problem 6 is the exception. (I’ve also included John’s own yellow-band error as the *, so the grid isn’t only scoring other people’s mistakes.)

Five of the six sit low and to the left — little changed when corrected, and few readers would have come away with a different understanding on account of them (the ApoB-table gap is the one likeliest to have left a reader genuinely unsure). Problem 6, as noted, is a standard rather than a correction — and a standard is exactly the kind of claim you can check for even application, which is what the rest of this article does.

John cites the retraction, not the Cleerly anomalies behind it

The one marker in the serious corner — top-right — isn’t one of John’s six. It’s what our team concluded was a serious problem in the third-party Cleerly analysis behind our earlier paper — serious enough that we asked the journal to retract it.

Those anomalies aren’t part of the preprint, but they're relevant here because John revisits that retraction — which came at our own request, over the reliability of the third-party plaque analysis the paper was built on.

John gives the “what” — that it was retracted — without the “who” (we requested it) or the “why.” (In the earlier video he noted this somewhat; but here he just leaves it at “retracted.”) And without the why, most people reasonably assume a retraction happened to the authors rather than by them. Ours ran the other way — we asked for it — and here’s the why, from my announcement the day we made it (from the transcript, lightly normalized; emphasis mine):

…we’ve had multiple participants from the KETO-CTA study independently resubmit their heart scans to Cleerly. The results showed notable differences from the original Cleerly analysis. That said, these individual submissions were broadly consistent with the other independent analyses within our study. Moreover, after publication of the paper, we learned the data provided to Cleerly was not fully blinded. Despite our repeated requests, including offers to cover costs, Cleerly declined to perform a fully blinded reanalysis. It’s worth emphasizing, as with any other longitudinal study, it has always been the expectation of the research team that all analyses would be fully blinded. This isn’t just standard, it’s vital for the integrity of the study itself. Given we could no longer stand behind the Cleerly-specific portion of the April 7th paper, we formally asked JACC: Advances to withdraw this paper, citing this Cleerly dataset and [Cleerly's] refusal to perform any quality assurance to ensure its accuracy. On January 12, 2026, the journal confirmed a full retraction.

That’s a big part of the why, in our own words. I’ve said from the start I won’t read motives, and I’m not going to here. And in fairness, John does engage the retraction — the unblinding, and the cross-platform question — but to the best of my knowledge, he hasn’t commented directly on the participant-level resubmission anomalies shown below.

In the “It Got Worse” episode, John is candid about this: he states he isn’t adjudicating the dispute. Early on: “I just don’t want to get into the whole debate about that, because it’s not something that I have any knowledge about.” Near the end, after a long stretch on the unblinding question: “who knows what the hell is going on.” What I haven't seen him engage is the specific reliability evidence itself — whether Cleerly's numbers hold up when the same scans are re-submitted.

It’s possible John simply hasn’t seen the resubmission data — though it was part of our retraction announcement video (above), shown participant by participant. To remove any doubt, I’ll put it right here:

Each pair compares the original Cleerly study analysis (red) against the same participant’s re-submitted scan, re-analyzed by Cleerly (blue). For several people the numbers don’t just differ in size — they flip direction: P8 goes from +32.0 mm³ to −47.9; P7 from +47.6 to −9.6.

This is where I’ll be the most firm in this piece: the pattern above should interest every data scientist looking at this story — supporter or critic alike. The same scans, resubmitted to the same platform, materially different answers. That’s the reliability question at the center of the retraction, and it’s the one most in a data scientist’s lane.

A data-integrity problem in a third-party dataset isn’t vague to investigate — it has a specific toolkit in data science itself.

When a platform’s numbers are in question, the checks are reproducibility ones: does it agree with itself on a resubmission, and with independent platforms reading the same images? Our study is almost purpose-built for that test — three separate platforms quantified the same scans, and when participants resubmitted, the Cleerly numbers moved materially — and, in our own comparison, stayed broadly consistent with the other platforms’ existing analyses.

There is more in the original Cleerly dataset than I can cover here, and some of it can be reviewed directly from the open dataset linked below. That's a piece of its own.

(For reference to readers, our dataset for the previous paper has been available on Citizen Science Foundation (CSF) since May 2025 and can be downloaded here without restriction. Anyone can download and explore that visit-level imaging dataset themselves — typically with the help of AI. The re-submission summary is shown above; the full per-subject cross-platform tables behind our retraction account aren't public yet — but I’ll have more to share on that soon.)

Have John's Standards Been Consistent?

“I’m not anti-keto. I am anti bad statistical methods and bad statistical analyses.”

I’ll take him at his word on that, and I'm not going to speculate about anyone's motives — I'd ask the same in return. But notice what the phrase does.

“Bad statistics” isn’t a matter of taste, the way one might prefer one chart style to another — it’s a verdict, and John is asserting he can deliver it as a data scientist. He doesn’t call these “things I’d have done differently.” He calls them major problems, violations, a paper that is statistically indefensible.

That’s the posture of an expert who is enforcing standards. It comes with one obligation: a standard is only a standard if it gives the same answer regardless of whose work it lands on. So let’s check that — not against my standards, but against his, in the order he laid them out.

Is there consistency on criteria?

Last November, John released a video (See Video 3 in timeline, above) on our previous paper laying out the statistical criteria he argued it had to meet — adjust for confounders like age and sex, check that model assumptions hold, follow specific reporting practices, keep causal language in check. That's a distinct list, and it can be applied to any other study point by point, which is the fair test of whether these are standards or preferences.

However, in his very next video on our research, John selected a new study as a reference group to argue our study looks worse by comparison, and he chose one: NATURE-CT. John choosing this specific study matters. Whatever first put the comparison in the air, he’s the one who selected this study and vouched for it — “the best we got for now, I think.”

This is important because the paper documents almost none of these practices, and John raised almost none of them from the video before. It runs no regression models at all — it makes no covariate adjustment for age or sex, and the paper carries no data-availability statement and no code. To his credit, he does note some things that make the study less comparable — that it’s retrospective, that it’s “not a perfect comparison,” that he “wouldn’t take it as like a 1-to-1 perfect comparison” — and he even flags that the plaque-change table omits means. I want to be fair about that: he gave caveats.

Here’s the distinction, and I want to be precise about it: a purely descriptive paper doesn’t owe a regression model — so I’m not faulting NATURE-CT for not adjusting. But John doesn’t use it descriptively. He uses its numbers to conclude the keto group has a “much greater amount of people who have a rapid plaque progression” — a comparison. And the moment he makes that comparison, the very standard he made central against us — adjust for age and sex, or it’s “omitted variable bias” — applies to it. His cross-cohort comparison uses no multivariable adjustment beyond restricting on baseline CAC, across groups that differ in scan interval and referral. The standard John applies that governed our models doesn't govern the comparison he built from NATURE-CT's.

He was then asked about this directly in the comments under that video — “for fairness” — whether the NATURE-CT study had any statistical errors.

John’s reply:

“Generally the paper is just reporting the data with conventional summary statistics, without statistical modeling. I don’t see any major problems with it.”

He pointed to a mislabeled “median” that should have read “mean,” and percentages “stated as 0.16%, but should be 16%” (a 100× difference) — filing both as “much smaller” reporting errors.

He filed those as “smaller.” But “no major problems” is a strong claim — so I did exactly what I described at the start: I ran NATURE-CT through the same kind of AI auditing, and verified the ones I cite here by hand against the paper. The two errors he named turned out not to be the only ones of their kind; a careful read surfaces more of the same class of reporting inconsistency than the one or two he characterized as trivial.

Importantly — I’m deliberately not cataloguing them exhaustively here as it’s simply more productive (and of course, better etiquette) to bring notice of these issues to the corresponding author directly and privately, particularly when our two studies are both, in part, in partnership with the same lab (the Lundquist Institute). I’ve sent the detailed observations directly to the paper’s corresponding author, Dr. Ronald P. Karlsberg, to help with any corrections they choose to make.

The only items I’ll need to be specific on with NATURE-CT will be covered in the checklist graphic below (“Do John’s standards travel?”) to allow specific and verifiable comparisons.

None of this makes NATURE-CT “a complete shambles” — that phrase was used on our preprint, and the issues are the ordinary, individually correctable kind. That’s exactly the point: they’re the same class of reporting slip John has cast as “smaller” on NATURE-CT and called “actively misleading” on ours. “No major problems with it” simply doesn’t survive the same lens he applied to our paper.

And “without statistical modeling” is offered here as entirely fine — “I don’t see any major problems with it” — even as the same modeling standard stays firmly on for our preprint. The point isn’t that NATURE-CT should have modeled; it’s that the standard is only being fired in one direction.

The same asymmetry runs through the other points he pressed hardest on. NATURE-CT’s paper carries no data-availability statement and no code — where he published his own reanalysis code specifically to contrast with us. It reports no confidence interval on its headline progression rate — a close cousin of the “serious reporting omission” he charged us with. And the model-assumption checks he ran on our data himself — the ones he called decisive, where “no statistician could look at that and say okay, there’s no problem with this model” — have no counterpart here: NATURE-CT’s methods prescribe paired t-tests for its continuous outcomes, yet the paper shows no paired-difference diagnostics and no normality check at all.

And the broader critique he leveled at our previous paper — stating conclusions without showing the analysis behind them, and running with no pre-specified plan — lands here too: NATURE-CT reports its progression findings as descriptive medians and IQRs with no p-values in Table 2 — even though its Methods state paired t-tests. It isn’t exempt as “just descriptive,” either — it runs its own paired t-tests and McNemar, so these reporting standards apply to it directly.

The checklist below shows where each of his own standards lands on it.

In short, the best way to evaluate John’s standards is to take his own checklist and see whether it lands evenly on the study he chose — here, NATURE-CT. The graphic above does exactly that, item by item, and the same checklist can be applied to other studies he’s reviewed in previous episodes.

And my larger point is that this kind of auditing is now something anyone can do with these rapidly advancing AI models — the whole reason I flagged it at the start. I ran it on our own preprint too, not only on NATURE-CT, and it flags things in our paper as readily as in anyone's. That's precisely the point: this is the new normal for every quantitative paper, ours included, and the fair response is to hold them all to it evenly rather than one of them selectively.

Is there consistency on the evidence bar?

The sharpest version of this isn’t what he asked of our models — it’s what he asked of his own.

He pressed age/sex adjustment on our work and kept pressing it, including in the same episode where he built his NATURE-CT comparison. The standard didn’t switch off; it just never pointed at the benchmark he chose.

His own low-CAC comparison is the clearest case. He spotted the confounder himself — “plaque predicts plaque,” so a group starting with less plaque may progress less, which makes it “not exactly a fair comparison for keto” — and he named the fix: restrict the keto group to those with CAC under 100. He didn’t have our follow-up plaque data, so he estimated what that subgroup would show. From that estimate he concluded it had “dramatically more rapid plaque progression than the nature study in all 3 methods.”

There’s the asymmetry in a sentence: what our models weren’t allowed to conclude from data we measured, his estimate was allowed to conclude from data he didn’t have. That’s the inconsistency, and it’s what the word “standard” is supposed to rule out — a standard gives the same answer, whichever paper it’s pointed at.

Is there consistency on causal language?

There’s a second thread worth pulling here, on causal language specifically — because it’s a standard John has been especially firm about. Throughout, he’s held that our data can’t support causal or comparative claims: the study is descriptive, underpowered, a single year long.

I partially agree, and we say much the same in our own Limitations. But watch what happens over the course of the “It got worse…” video. Early on, discussing the keto plaque metrics, he says they should be read as “just descriptive of this group” and — his words — “not generalizable or causal or predictive.”

Then, roughly forty minutes later, he builds his closing comparison on them anyway. His words: “You see 2 groups that are relatively comparable in baseline metrics and even, you know, the keto group other than the LDL-C is generally healthier. But then you see much greater amount of people who have a rapid plaque progression in the keto group.” Set the tone aside and look only at the structure of that argument — two groups he calls relatively comparable, one difference he foregrounds, therefore a conclusion about progression. That is the causal-flavored, matched-comparison form of reasoning he had just said these metrics could not carry, and the same kind of language he insists our paper isn’t entitled to.

To his credit, he qualified the comparison repeatedly — “not a perfect comparison,” “I wouldn’t take it as like a 1-to-1 perfect comparison,” “not exactly a fair comparison for keto,” and that the differences “could cut both ways.” I want that on the record. My concern isn’t that he skipped the caveats — it’s that the strength of the closing claim outruns them: the qualifications describe a comparison you can’t lean on, and the finale leans on it anyway.

So I’ll close this the way I’ve tried to hold the whole piece — with a question rather than a verdict. If you’d known that the standard used to condemn our paper — adjust for basic confounders like age and sex, which he has pressed steadily from his first video on our work through his most recent — is one he didn't apply when he built his comparison on the study he chose to measure us against, would it change how you weigh that conclusion? The ask is simple and fair — hold his conclusions to the same test he holds ours to: check it, and see whether it’s applied evenly.

Two respectful challenges

I’ll end with two asks — of John, and of anyone doing this kind of work, myself included.

The first is about consistency. AI has made finding flaws cheap. Point a capable model at any quantitative paper and it will surface something — I’ve used it here, on our own preprint as much as on NATURE-CT. But that shifts where the judgment sits. When the finding is easy, the weight falls on a different decision: which papers you point it at, and how hard you press once you’re there. A bar applied at full strength to one paper and half strength to the next isn’t a standard; it’s a preference wearing a standard’s vocabulary. I’m not asking for a gentler bar. I’m asking that whatever bar is set for our work be the one carried to the next study, whatever it finds.

The second is about proportion. Errors aren’t all the same size. Some move a paper’s conclusion and some don’t, and that difference matters more than the count of them. It’s why five of the six went on a matrix instead of being simply conceded or disputed one by one: a mislabeled whisker and a headline finding built on a third-party analysis that doesn’t hold up are not the same kind of problem, and treating them as interchangeable tells a reader nothing about which one to care about. So the second ask is that the sizing travel too. If a class of error is fatal in one paper, it’s fatal in the next paper that has it — and if it’s minor there, it was minor here.

Final thoughts…

I want to close on a positive note.

Yes, I’ve had a lot to call out with regard to what John's analysis has chosen to focus on, not focus on, and the consistency of how these criticisms are applied.

That said, I truly value the added time and care behind the critique. And I’ll again emphasize John has spotted items worth addressing in the present manuscript, even if they are low on probability and severity of impact — they are still worth consideration and correction, and I’m very thankful for the feedback.

The Western Denmark Heart Registry (WDHR) is a large clinical database that tracks patients undergoing cardiac testing across Western Denmark. Because it sits inside a unified healthcare system, it can follow people consistently over time and capture real-world outcomes, including coronary artery calcium (CAC) scores and cardiovascular events. It’s an excellent dataset that is well-suited for evaluating heart disease event rates in real-world populations.

A recent analysis published in Circulation1 included more than 23,000 middle-aged symptomatic patients followed for over four years, examining how LDL-C, CAC score, and cardiovascular events related to one another.

What they found may challenge some common assumptions about LDL-C.

The surprising finding

For the ~53% of participants (n=~12,000) with a CAC score of zero, LDL-C (even at very high levels by most conventional standards) was not associated with an increased incidence of heart attack or broader cardiovascular events.

Share

So what did the study actually find?

The cohort was 23,132 symptomatic adults, median age 57, followed for a median of 4.3 years.

Across the full group, LDL-C was modestly associated with cardiovascular events when analyzed as a continuous variable.

Take a look at the figure below.

This figure shows that the incidence of heart attack and cardiovascular disease increases by 28% and 14%, respectively, for every 38.7mg/dl higher level of LDL-C.

But let’s add calcium scoring to the picture

When the researchers stratified participants by calcium score, the picture changed.

Keep in mind, a CAC score isn’t just another risk factor. It’s evidence that calcified plaque is already present in the coronary arteries. In practical terms, CAC helps separate people with detectable coronary plaque from those without it.

And in this study, it appeared to make a big difference.

• CAC > 0 (47% of participants): incidence of heart attack and overall cardiovascular disease increased by 30% and 14%, respectively, per 38.7 mg/dL rise in LDL-C

• CAC = 0 (53% of participants): no significant association between LDL-C and heart attack or overall cardiovascular disease incidence

In other words, LDL-C was associated with events only when plaque was already present.

The authors put it plainly…

“Our results challenge the general assumption that LDL-C provides consistent and universal increases in risk throughout different patient populations.”

So what could be happening here to explain this?

At this point, it may be fair to ask…

Is this just because the CAC = 0 group had lower LDL-C to begin with?

And if LDL-C did not associate with heart disease in the CAC = 0 group, what did?

That's where the paid subscription picks up. If you're finding this useful and want to follow the thread all the way through, please consider subscribing! We will tease Part 2 of this newsletter at the end of the post.

Subscribe now

Read more

Read more

This glossary covers the full range of terms denoted in TFP_: from fundamental lipoprotein structure to complex parts of Dave’s research.

This is a living reference. We will try to link back to this post in newsletters so if you are new or unfamiliar with some terminology, this may help bolster your learning. We will also attempt to update each term with references in case you want a deeper dive on your deep dive.

Terms are organized by topic, not alphabetically. We hope you enjoy!

Subscribe now

1. Core Lipoprotein Biology

Structure & Particles

Lipoprotein

Typically, a spherical particle that transports lipids (fat-soluble) through the aqueous bloodstream (water-based). Each particle has a hydrophobic core of cholesteryl esters and triglycerides, wrapped in a phospholipid monolayer covered with proteins called apolipoproteins. Think of it as a purpose-built fat-soluble cargo pod for a water-based delivery network.

Cholesteryl Ester (CE)

The storage form of cholesterol. Free cholesterol is esterified (a fatty acid is attached) by the enzyme LCAT in plasma or by ACAT inside cells, making it hydrophobic and suitable for packing into the lipoprotein core. The majority of cholesterol in LDL and HDL is in this esterified form.

Triglyceride (TG)

A glycerol molecule with three fatty acid chains attached. The primary form of dietary and stored fat. In the circulation, triglycerides are packaged into TG-rich lipoproteins — chylomicrons (from food) and VLDL (from the liver) — and delivered to tissues via LPL-mediated hydrolysis. Elevated TG can be a sign of insulin resistance.

Check out one of Dave’s older articles from cholesterolcode.com: The article proposes a “triglyceride carryover” effect, where elevated triglycerides in the morning may be from incomplete fasting related to the prior meal rather than true metabolic dysfunction. It shows that triglycerides consistently normalize after ~12+ hours of fasting, suggesting shorter fasts can result in a false positive for hypertriglyceridemia and potentially lead to misinterpretation of lipid results.

Phospholipids

Amphipathic molecules (one water-loving head, two fat-loving tails) that form the outer shell of every lipoprotein particle. This structural role makes them essential to lipoprotein integrity. Phosphatidylcholine is the most abundant type. Phospholipids are also substrates for several key enzymes including LCAT and Lp-PLA2.

Free (Unesterified) Cholesterol

Cholesterol in its unmodified form, present on lipoprotein surfaces and in cell membranes. Unlike cholesteryl esters, free cholesterol is not stored in lipoprotein cores — it sits in the surface layer. Its proportion relative to esterified cholesterol affects membrane fluidity and lipoprotein function.

Apolipoproteins

Proteins embedded in or associated with lipoprotein surfaces. They perform two essential functions: structural (holding the particle together) and functional (serving as enzyme cofactors, receptor ligands, and metabolic signals).

The major apolipoproteins:

• ApoB-100 — Structural protein of all liver-derived lipoproteins (VLDL, IDL, LDL). One ApoB-100 per particle — the direct measure of “particle number”.

• ApoB-48 — Intestinal isoform of ApoB. Structural protein of chylomicrons. Truncated version lacking the LDL receptor-binding domain.

• ApoA-I — Major structural protein of HDL. Activates LCAT and drives ABCA1-mediated cholesterol efflux from macrophages. Primary functional protein of reverse cholesterol transport.

• ApoE — Ligand for LDL receptor and LRP1-mediated clearance of remnant particles. Three isoforms (E2, E3, E4) with dramatically different metabolic and neurological implications.

• ApoC-II — Essential activating cofactor for lipoprotein lipase (LPL). Without ApoC-II, LPL cannot efficiently hydrolyze TG in chylomicrons and VLDL.

• ApoC-III — Inhibitor of LPL and of hepatic uptake of remnant particles. High ApoC-III is associated with elevated TG and increased remnant cholesterol.

One of the most important conceptual shifts in modern lipidology: apolipoproteins are not passive structural scaffolding…they help the lipoproteins go where they need to go the moment they enter circulation.

HDL-C (HDL Cholesterol)

The cholesterol content of high-density lipoprotein particles. Measured as part of the standard lipid panel. Sometimes called ‘good cholesterol’ because higher levels of HDL-C are inversely associated with cardiovascular risk.

LDL-C (LDL Cholesterol)

The cholesterol content of low-density lipoprotein particles. The most widely used lipid treatment target in clinical guidelines worldwide. Usually calculated via the Friedewald equation (TC − HDL-C − TG/5) which may grossly underestimate LDL-C when TG are higher.

LDL-C measures the cargo, not the number of ships. In certain situations, LDL-C by itself may not give the best indication of how many LDL particles there are. For example, someone with a lower LDL-C may have a higher LDL particle count than anticipated if those LDL particles are smaller, more lipid poor, particles.

NEFA (Non-Esterified Fatty Acids / Free Fatty Acids)

Fatty acids circulating in the bloodstream bound to albumin, not packaged in lipoproteins. Released from adipose tissue during lipolysis (fasting, exercise, stress) or from TG hydrolysis by LPL. The primary fuel substrate during fasting and sustained aerobic exercise.

Elevated fasting NEFA in the context of elevated insulin is a marker of insulin resistance — adipose tissue lipolysis is normally suppressed by insulin. High NEFA may also drive hepatic VLDL production and contribute to ectopic fat deposition in this context.

Major Lipoprotein Classes

Chylomicrons

The largest lipoprotein (~75–1200 nm), assembled in intestinal enterocytes (intestinal lining cells) from dietary fat and secreted into the lymph before reaching the bloodstream. Contain ApoB-48, ApoC-II, ApoC-III, and ApoE. Deliver dietary TG to peripheral tissues via LPL-mediated hydrolysis. What remains after TG is stripped is the chylomicron remnant.

Chylomicrons themselves are typically considered too large to enter the arterial wall while their remnants (partially digested byproducts) are not.

VLDL (Very Low-Density Lipoprotein)

The liver’s endogenous TG-transport particle. Contains ApoB-100. Secreted in large quantities in states of insulin resistance, high carbohydrate intake, or excess free fatty acid flux to the liver. Progressive LPL-mediated lipolysis converts VLDL → IDL → LDL.

Excess VLDL secretion paired with poor VLDL turnover is the upstream driver of the most common atherogenic dyslipidemia pattern: high TG, low HDL-C, and small dense LDL.

IDL (Intermediate-Density Lipoprotein)

The transitional particle formed when VLDL is partially lipolyzed. Carries ApoB-100 and ApoE. Either cleared by the LDL receptor (via ApoE binding) or further processed by hepatic lipase into LDL. Elevated levels are typically considered a risk for heart disease, but IDL isn’t measured on standard lipid panels. Levels are measured indirectly via non-HDL-C, ApoB, and remnant cholesterol calculations.

LDL (Low-Density Lipoprotein)

The end product of VLDL lipolysis. Cholesterol-enriched, ApoB-100-bearing. Typically considered by many lipidologists to be the particle most strongly associated with atherosclerosis. Some research indicates it enters the subendothelial space via transcytosis at a rate proportional to circulating particle concentration.

HDL (High-Density Lipoprotein)

The smallest and densest lipoprotein. Accepts cholesterol from peripheral tissues and macrophages via ABCA1 and ABCG1 transporters, matures via LCAT, and delivers cholesteryl esters to the liver via SR-B1. This cycle — from tissue to particle to liver — is reverse cholesterol transport (RCT).

HDL is often considered to be anti-atherogenic primarily through cholesterol efflux and RCT.

Lp(a) (Lipoprotein(a))

An LDL-like particle with an additional protein — Apo(a) — covalently bonded to ApoB-100 via a disulfide bridge. Apo(a) has structural homology to plasminogen, and as such many in lipidology consider Lp(a) to have both atherogenic properties (subintimal retention, foam cell induction, vascular calcification) and prothrombotic properties (impaired fibrinolysis). Some research also indicates Lp(a) may also play a role in wound healing, and in the immune system. Lp(a) level is typically assumed to be ~90% genetically determined by the LPA gene. However, other influences on Lp(a) level have also been observed. For example, Lp(a) is also an Acute Phase Reactant (APR) and thus may be influenced by inflammatory signaling, and changes from diet-related factors have also been observed (Vitamin C, carnitine, low carbohydrate diets).

Particle Subtypes & Patterns

Small Dense LDL (sdLDL) / Pattern B

A subpopulation of LDL particles smaller than ~25.5 nm and denser than the predominant large buoyant LDL. Associated with lower LDL receptor affinity (longer plasma residence time), and more susceptibility to oxidative modification. Many lipidologists consider sdLDL particles to be more atherogenic for these reasons, and because some evidence suggests they may be more likely to be retained by the arterial wall. Pattern B denotes predominance of sdLDL.

Pattern B is has been strongly associated with insulin resistance, elevated TG, and low HDL-C. It starts to emerge more frequently when TGs exceed 95 mg/dl. Some lipidologists use a LDL-C/ApoB ratio as a quick screen: a low ratio (<1.2) is associated with small, cholesterol-poor particles even when LDL-C is low.

Large Buoyant LDL / Pattern A

The larger, less dense LDL subclass — diameter above ~25.5 nm. Associated with better LDL receptor affinity and lower oxidizability compared to sdLDL. Pattern A denotes predominance of large buoyant particles.

HDL2 vs HDL3

HDL exists as a spectrum of subclasses. HDL2 is larger and more buoyant, enriched in cholesteryl esters, and reflects more mature HDL. HDL3 is smaller and denser, reflecting nascent or remodeled HDL. Hepatic lipase converts HDL2 → HDL3; LCAT activity drives the reverse.

Remnant Particles

Chylomicron remnants and VLDL remnants (including IDL) formed after LPL-mediated TG hydrolysis. Enriched in cholesteryl esters, ApoE, and ApoC-III.

Functional Lipid Fractions & Metrics

Remnant Cholesterol (RC)

Cholesterol carried in VLDL, IDL, and chylomicron remnants. Calculated as TC − HDL-C − LDL-C. A direct measure of the cholesterol burden from TG-rich lipoprotein remnants. Not a separate lab test on standard panels — it is derived from existing values.

LDL-TG / eLDL-TG (LDL Triglyceride Content)

The triglyceride content within LDL particles. LDL normally carries very little TG, but in insulin-resistant states with high CETP activity, TG is exchanged into LDL in return for cholesteryl esters — producing TG-enriched, cholesteryl ester-depleted LDL. eLDL-TG is an estimated version using surrogate calculations. It has been proposed that LDL-TG may be an early sign of one’s fat cells becoming overstuffed and insulin resistant (sometimes also called a Personal Fat Threshold).

Non-HDL-C

Total cholesterol minus HDL-C. Captures cholesterol in all apoB-containing lipoprotein classes. Requires no fasting and no extra calculations beyond standard panel results.ApoB Concentration

The direct plasma concentration of apolipoprotein B — one molecule per particle. Measured by immunoassay. Captures LDL, VLDL, IDL, Lp(a), and chylomicrons in a single number. In mainstream lipidology, considered the most accurate single measure of cardiovascular “particle burden”.

ApoA-I Concentration

The plasma concentration of apolipoprotein A-I, measured by immunoassay. The primary structural and functional protein of HDL. Often considered to be one of the more accurate proxies for HDL particle number and potentially, by extension, reverse cholesterol transport capacity.

The ApoB/ApoA-I ratio captures both possible “atherogenic” burden (ApoB) and antiatherogenic capacity (ApoA-I) in a single number and is among the lipid ratios most strongly associated with CV risk in large population studies.

2. Lipid Transport, Enzymes & Regulatory Proteins

Lipid Metabolism Enzymes

LPL (Lipoprotein Lipase)

Anchored to capillary endothelium in muscle, adipose, and cardiac tissue. Hydrolyzes TG in circulating chylomicrons and VLDL, releasing fatty acids for local uptake and energy use. Activated by ApoC-II and ApoA-V; inhibited by ApoC-III and ANGPTL3/4/8.

LPL is the gatekeeper of plasma TG clearance.

Hepatic Lipase (HL)

A liver-expressed lipase that remodels IDL into LDL (by removing residual TG) and converts larger HDL2 into smaller HDL3 (by hydrolyzing HDL phospholipids and TG). High hepatic lipase activity is associated with small dense LDL and smaller HDL particles.

Hepatic lipase activity is increased in insulin-resistant states — one mechanism by which metabolic syndrome drives Pattern B dyslipidemia and lowers HDL2.

Endothelial Lipase (EL)

A lipase expressed on vascular endothelium with preferential activity for phospholipids in HDL. Hydrolyzes HDL phospholipids, reducing HDL particle size and accelerating HDL catabolism. Upregulated by inflammatory cytokines.

Endothelial lipase is a key link between systemic inflammation and low HDL-C — it partly explains why inflammatory conditions (sepsis, rheumatoid arthritis, acute illness) cause rapid HDL-C drops independent of other lipid changes.

CETP (Cholesteryl Ester Transfer Protein)

Shuttles cholesteryl esters from HDL to VLDL/LDL in exchange for triglycerides. Bridges the HDL and atherogenic lipoprotein metabolic pools. Elevated CETP activity simultaneously lowers HDL-C and enriches VLDL with cholesteryl esters.

LCAT (Lecithin-Cholesterol Acyltransferase)

Esterifies free cholesterol on HDL, converting nascent discoidal HDL into mature spherical HDL. Activated by ApoA-I. A critical step in HDL maturation and reverse cholesterol transport efficiency.

ACAT2 (Acyl-CoA:Cholesterol Acyltransferase 2)

Expressed in intestinal enterocytes and hepatocytes. Esterifies free cholesterol for packaging into chylomicrons (intestine) and VLDL (liver). Enables dietary and newly synthesized cholesterol to be incorporated into lipoproteins for transport.

HMG-CoA Reductase

The rate-limiting enzyme in the mevalonate pathway — converts HMG-CoA to mevalonate, the precursor to cholesterol and isoprenoids. Expressed primarily in the liver. Upregulated by SREBP-2 when intracellular cholesterol falls. The direct molecular target of statins.

Lathosterol and desmosterol, measurable on sterol panels, reflect HMG-CoA reductase activity. A ‘hypersynthesizer’ with elevated lathosterol has upregulated flux through this pathway.

PLTP (Phospholipid Transfer Protein)

Transfers phospholipids between lipoprotein particles — particularly from TG-rich lipoproteins to HDL during lipolysis. PLTP activity modulates HDL particle size and composition and influences the overall lipoprotein remodeling landscape.

PLTP is elevated in insulin-resistant states and may contribute to HDL dysfunction and the production of small, dysfunctional HDL particles independent of CETP activity.

Apolipoproteins (Regulatory & Structural)

Note: ApoB-100, ApoB-48, ApoA-I, ApoE are covered in depth in Section 1. The entries below focus on the regulatory apolipoproteins and Apo(a).

ApoC-II

The obligate activating cofactor for LPL. Without ApoC-II on the chylomicron or VLDL surface, LPL cannot efficiently hydrolyze the particle’s TG core. Deficiency causes a rare but severe hypertriglyceridemia with pancreatitis risk.

ApoC-II must be transferred from HDL to chylomicrons and VLDL for LPL activation to occur — a neat example of how HDL acts not just as a cholesterol carrier but as a TG metabolism coordinator.

ApoC-III

Considered by some lipidologists to be one of the most clinically important regulatory apolipoproteins. Inhibits LPL directly, inhibits hepatic uptake of TG-rich lipoprotein remnants, and may promote hepatic VLDL secretion. Elevated ApoC-III is consistently associated with elevated TG and increased remnant particle burden.

ApoC-III on LDL particles may also be an independent predictor of CV risk.

Apo(a)

The defining structural protein of Lp(a). Covalently bonded to ApoB-100. Contains multiple kringle repeat domains with strong structural homology to plasminogen’s kringle domains. The number of kringle IV type 2 repeats (encoded by the LPA gene) inversely determines Lp(a) particle size and is typically considered to be the primary determinant of plasma Lp(a) concentration.

Shorter Apo(a) isoforms (fewer kringle repeats) are produced and secreted more efficiently, leading to higher circulating Lp(a) levels.

Receptors & Transporters

LDL Receptor (LDLr)

The primary hepatic receptor for LDL and remnant particle clearance. Binds ApoB-100 (on LDL) and ApoE (on remnants and IDL), internalizes them via endocytosis, releases cargo in the lysosome, and recycles to the cell surface. LDL receptors can also be expressed by other nucleated cells, like endothelial cells and macrophages. Expression is transcriptionally regulated by SREBP-2 and post-translationally by PCSK9.

The most common cause of FH an LDLR receptor defect.

LRP (LDL Receptor-Related Protein / LRP1)

A large multi-ligand endocytic receptor on hepatocytes and macrophages. Binds ApoE-containing remnant particles (chylomicron remnants, IDL) for clearance. Acts as a backup to the LDL receptor for remnant clearance.

LRP1 is particularly important in the postprandial state when large numbers of chylomicron remnants must be cleared from circulation. Its activity may help explain why remnant cholesterol clearance varies considerably between individuals.

SR-B1 (Scavenger Receptor Class B Type 1)

The hepatic receptor for selective cholesteryl ester uptake from HDL — without internalizing the whole particle. HDL docks, offloads its CE cargo, and returns to circulation lipid-depleted. The terminal step of reverse cholesterol transport.

SR-B1 deficiency in mice produces high HDL-C and simultaneously increased atherosclerosis — possibly because despite the higher HDL-C, the HDL particles are not able to function normally.

LOX-1 (Lectin-Like Oxidized LDL Receptor-1)

A scavenger receptor expressed on endothelial cells and macrophages that mediates uptake of oxidized LDL. Unlike the LDL receptor, LOX-1 uptake is not feedback-regulated by cellular cholesterol, allowing uncontrolled intracellular cholesterol accumulation.

LOX-1 expression is upregulated by inflammatory cytokines, angiotensin II, and shear stress — potentially explaining why hypertension and inflammation may accelerate foam cell formation and atherosclerosis independently of plasma LDL-C levels.

NPC1L1 (Niemann-Pick C1-Like 1)

Expressed in intestinal enterocyte brush borders and liver canaliculi. Mediates cholesterol absorption from the intestinal lumen into enterocytes. The molecular target of ezetimibe.

Hyperabsorbers — identifiable by elevated sitosterol and campesterol on a sterol panel — have constitutively high NPC1L1 activity.

ABCG5 / ABCG8

A heterodimeric transporter pair in intestinal enterocytes and hepatocytes. Pumps plant sterols (sitosterol, campesterol) and excess cholesterol back into the intestinal lumen or bile, preventing systemic accumulation.

Loss-of-function mutations cause sitosterolemia — dramatically elevated plant sterols, xanthomas, and premature atherosclerosis. Often misdiagnosed as FH because plant sterols cross-react in some cholesterol assays.

ABCA1 (ATP-Binding Cassette Transporter A1)

Effluxes cholesterol and phospholipids from cell membranes to lipid-free ApoA-I, forming nascent discoidal HDL. The first obligate step in reverse cholesterol transport. Loss-of-function causes Tangier disease (near-absent HDL, cholesterol accumulation in tissues).

Macrophage ABCA1-mediated cholesterol efflux capacity — a functional measure of how much cholesterol can be exported per unit time — has shown stronger associations with CV risk than HDL-C concentration in some prospective studies.

ABCG1 (ATP-Binding Cassette Transporter G1)

Works downstream of ABCA1. Effluxes cholesterol from cells to mature, lipidated HDL particles. ABCA1 and ABCG1 operate in tandem to complete macrophage cholesterol efflux.

While ABCA1 handles the first step (nascent HDL formation), ABCG1 handles the second (cholesterol loading into mature HDL). Both are transcriptionally regulated by LXR — a key nuclear receptor activated by cholesterol excess.

Regulatory Pathways & Proteins

PCSK9

A hepatocyte-secreted serine protease that binds LDL receptors after LDL internalization and routes the LDLr-PCSK9 complex to lysosomal degradation rather than receptor recycling. Net result: fewer LDL receptors on the hepatocyte surface, less LDL clearance, higher circulating LDL-C.

PCSK9 gain-of-function mutations are a cause of FH.

SREBP-2 (Sterol Regulatory Element-Binding Protein 2)

A transcription factor that acts as the master regulator of cholesterol synthesis and uptake. When intracellular cholesterol falls, SREBP-2 is processed and translocates to the nucleus, upregulating HMG-CoA reductase (synthesis) and LDL receptor (uptake). This is the primary mechanism by which statins lower LDL-C.

SREBP-1c (Sterol Regulatory Element-Binding Protein 1c)

A transcription factor that is the master regulator of fatty acid and triglyceride synthesis. Activated by insulin and LXR. Drives expression of genes for de novo lipogenesis (DNL) — the conversion of carbohydrates to fat in the liver.

SREBP-1c is a key mechanistic link between high carbohydrate intake, hyperinsulinemia, hepatic fat synthesis, and VLDL overproduction. This pathway is central to understanding how refined carbohydrates may drive hypertriglyceridemia. Necrotic core growth is driven by both lipid supply (LDL, remnants) and impaired efferocytosis. Reducing circulating lipids and reducing inflammation are both considered to be mechanistically relevant to plaque stabilization.

FoxO1 (Forkhead Box O1)

A transcription factor that regulates hepatic gluconeogenesis and is suppressed by insulin signaling. Also contributes to regulation of ApoC-III and VLDL production. In insulin resistance, FoxO1 remains active when it should be suppressed.

FoxO1 provides a mechanistic link between insulin resistance and elevated ApoC-III: impaired insulin suppression of FoxO1 allows continued ApoC-III expression, which inhibits LPL and raises plasma TG.

FXR (Farnesoid X Receptor)

A nuclear receptor activated by bile acids. Acts as the body’s primary bile acid sensor. When activated, FXR suppresses bile acid synthesis, upregulates bile acid transporters, and reduces hepatic TG production (via suppression of SREBP-1c).

FGF19 / FGF21

FGF19 is an intestinal hormone secreted in response to FXR activation by bile acids. It acts on the liver to suppress further bile acid synthesis and inhibits hepatic glucose and lipid production. FGF21 is a liver-derived metabolic hormone activated by PPARα and metabolic stress. It regulates fatty acid oxidation, glucose uptake, and energy expenditure.

ANGPTL3 / 4 / 8 (Angiopoietin-Like Proteins)

A family of circulating proteins that inhibit LPL activity in a coordinated, tissue-specific manner, regulating where fatty acids from TG hydrolysis are delivered depending on fed/fasted state. ANGPTL3 also inhibits endothelial lipase. ANGPTL3 loss-of-function produces a unique phenotype: simultaneously low LDL-C, low TG, and low HDL-C — with markedly reduced ASCVD.

LXR (Liver X Receptor)

A nuclear receptor activated by oxysterols (cholesterol metabolites), functioning as an intracellular cholesterol sensor. When activated, LXR upregulates ABCA1, ABCG1, ABCG5/G8, and SREBP-1c. Net effect: promotes cholesterol efflux and export while also increasing fatty acid synthesis.

PPARα (Peroxisome Proliferator-Activated Receptor Alpha)

A nuclear receptor expressed primarily in liver and muscle. Activated by fatty acids and fibrates. Upregulates LPL expression, downregulates ApoC-III, increases fatty acid oxidation, and reduces hepatic VLDL production.

PPARα is also activated by fasting and ketogenic diets — increased fatty acid availability during carbohydrate restriction upregulates PPARα target genes, increasing fatty acid oxidation and reducing TG. This is part of the metabolic shift underlying low-carb-induced TG reduction.

PPARγ (Peroxisome Proliferator-Activated Receptor Gamma)

A nuclear receptor expressed primarily in adipose tissue. Master regulator of adipocyte differentiation and lipid storage. Improves insulin sensitivity by redistributing fat from ectopic sites (liver, muscle) into subcutaneous adipose.

Bile Acids

Sterol-derived molecules synthesized in the liver from cholesterol, secreted into bile, and used to emulsify dietary fats in the intestine. Reabsorbed in the ileum and recycled via the portal circulation (enterohepatic circulation). Also function as signaling molecules via FXR and TGR5 receptors.

Major Transport Pathways

Exogenous Lipid Pathway

Dietary fat packaged into chylomicrons in intestinal enterocytes → secreted into lymph → enters bloodstream → TG hydrolyzed by LPL in peripheral tissues → chylomicron remnants taken up by liver via LDL receptor and LRP1.

Endogenous Lipid Pathway

Liver secretes VLDL loaded with TG and ApoB-100 → peripheral LPL converts VLDL → IDL → LDL → LDL taken up by LDL receptor-expressing cells (liver, adrenals, gonads, etc.).

Insulin resistance increases VLDL secretion and impairs LPL activity simultaneously — a double hit that explains why metabolic syndrome so reliably produces elevated TG and the full atherogenic dyslipidemia triad.

Reverse Cholesterol Transport (RCT)

The pathway by which cholesterol is transported from peripheral tissues and macrophages back to the liver for excretion. Steps: (1) ABCA1 effluxes cholesterol to lipid-free ApoA-I → forms nascent HDL; (2) LCAT esterifies free cholesterol → matures HDL; (3) CETP may exchange CE for TG with LDL/VLDL; (4) SR-B1 on hepatocytes accepts CE from HDL; (5) Liver excretes cholesterol as bile acids or free cholesterol.

RCT efficiency — not just HDL-C — is the current mechanistic focus of HDL research. Cholesterol efflux capacity has shown associations with CV events independently of HDL-C in multiple prospective studies.

3. Atherogenesis & Vascular Biology

Plaque Biology & Imaging

Low Attenuation Plaque (LAP)

On coronary CT angiography (CCTA), plaque with CT attenuation below 30 Hounsfield units. Corresponds histologically to lipid-rich or necrotic core content. A high-risk plaque feature predictive of acute coronary syndrome risk beyond stenosis severity.

TAV / PAV (Total Atheroma Volume / Percent Atheroma Volume)

IVUS-derived measures of plaque burden. TAV is the absolute volume of plaque across a standardized coronary segment. PAV normalizes TAV to total vessel volume, enabling comparison across patients.

NCPV (Non-Calcified Plaque Volume)

The CCTA-derived volume of non-calcified (soft) plaque in the coronary arteries. Includes lipid-rich, fibrous, and mixed plaque. Regarded as more dynamic and more susceptible to rupture than calcified plaque.

Thin-Cap Fibroatheroma (TCFA)

A high-risk (vulnerable) plaque morphology: large lipid-rich necrotic core with an overlying fibrous cap measuring less than 65 µm. The thin cap is mechanically susceptible to rupture under hemodynamic stress. TCFA rupture and subsequent thrombosis is the most common mechanism of acute MI.

TCFAs can now be identified non-invasively on CCTA via high-risk plaque features: low attenuation, positive remodeling, napkin-ring sign, and spotty calcification — predicting ACS risk beyond stenosis severity alone.

Necrotic Core

The lipid-rich, acellular region inside an advanced plaque formed by accumulated apoptotic foam cells and extracellular lipid. Grows when efferocytosis (clearance of dead cells) is overwhelmed. A large necrotic core is the defining feature of a high-risk vulnerable plaque.

Necrotic core growth is driven by both lipid supply (LDL, remnants) and impaired efferocytosis. Reducing circulating lipids and reducing inflammation are both mechanistically relevant to plaque stabilization.

Fibrous Cap

The layer of smooth muscle cells, macrophages, and collagen that overlies the necrotic core. Thickness determines mechanical stability. Maintained by collagen synthesis (SMCs) and degraded by matrix metalloproteinases (MMPs) produced by activated macrophages.

Spotty Calcification

Small microcalcifications within lipid-rich plaque. On CCTA, identified as small, punctate calcium deposits within non-calcified plaque regions. Associated with local inflammation and macrophage activity rather than stable organized calcium.

Spotty calcification is considered a high-risk plaque feature that may seem counterintuitive given that high CAC scores generally indicate stable plaque. The distinction is that microcalcification indicates active, ongoing, inflammation, which isn’t inherently implied by macrocalcification.

Positive / Negative Remodeling

Positive remodeling: outward expansion of the vessel wall to accommodate growing plaque (the Glagov phenomenon). The arterial lumen is preserved despite significant plaque accumulation, causing standard angiography to miss the burden. Negative remodeling: inward vessel constriction, leading to lumen reduction out of proportion to plaque volume.

Positive remodeling is considered a high-risk CCTA feature associated with vulnerable, lipid-rich plaque. It may explain why many patients have their first acute MI without any prior warning on functional stress tests — the lumen looks fine until the plaque ruptures.

Cellular Mechanisms

Endothelial Dysfunction

Typically thought of as the earliest and most reversible stage of atherogenesis. Characterized by reduced nitric oxide (NO) production by eNOS, increased endothelial permeability, upregulation of adhesion molecules (ICAM-1, VCAM-1, E-selectin), and a pro-inflammatory, pro-thrombotic endothelial phenotype.

Endothelial dysfunction is detectable before structural plaque forms and can be measured by flow-mediated dilation (FMD) of the brachial artery.

Transcytosis

The vesicle-mediated transport of ApoB-containing lipoproteins (primarily LDL) across the endothelial cell into the subintimal space.

Monocytes → Macrophages → Foam Cells

The cellular sequence at the heart of early plaque development. (1) Circulating monocytes are recruited to the endothelium via chemokines (MCP-1/CCL2) and adhesion molecules. (2) They transmigrate into the subintimal space and differentiate into macrophages under M-CSF signaling. (3) Macrophages ingest modified LDL via unregulated scavenger receptors (SR-A, CD36, LOX-1), accumulating intracellular lipid and becoming foam cells.

Foam cell formation is not just a passive lipid-filling process — foam cells are metabolically active, secreting cytokines, proteases, and reactive oxygen species that drive plaque progression and instability.

Smooth Muscle Cells (SMC) in Atherosclerosis

Vascular smooth muscle cells (VSMCs) migrate from the media to the intima in response to PDGF and other growth signals. In the intima they proliferate and produce extracellular matrix — contributing to the fibrous cap. However, SMCs can also take up oxidized LDL via scavenger receptors and become foam cells.

SMC-derived foam cells have only recently been recognized as a significant component of advanced plaques. Single-cell RNA sequencing studies suggest SMCs contribute substantially to the foam cell pool.

T-Lymphocytes in Atherosclerosis

Activated T cells (primarily Th1 subtype) infiltrate plaques and secrete pro-inflammatory cytokines (IFN-γ, TNF-α) that amplify macrophage activation, increase MMP production (destabilizing the fibrous cap), and impair efferocytosis. Regulatory T cells (Tregs) play a counterbalancing anti-inflammatory role.

The adaptive immune system plays a meaningful role in plaque progression and vulnerability.

Oxidation, Modification & Inflammation

OxLDL (Oxidized LDL)

LDL that has undergone oxidative modification of its lipid and protein components in the subendothelial space. Recognized by scavenger receptors (LOX-1, SR-A, CD36) on macrophages, driving foam cell formation. Also activates endothelial cells and promotes inflammation.

Some lipidologists suggest that clinically meaningful oxidation of LDL occurs within the subintima, and that antioxidants in the serum would largely prevent oxidation of LDL in the bloodstream. On the other hand, serum oxidized LDL can be measured in humans, and has been induced in vivo in some animal models. In any case, antioxidant defenses (PON1, vitamin E) modulate the rate of LDL oxidation and may partly explain variation in CV risk at the same LDL-C level.

OxPL-ApoB (Oxidized Phospholipids on ApoB)

Oxidized phospholipids covalently bound to ApoB-containing lipoproteins — particularly Lp(a). Among the most potent pro-inflammatory lipid signals known. Activate endothelial cells, drive macrophage foam cell formation, stimulate osteoblast-like differentiation in valvular tissue, and promote vascular calcification.

OxPL-ApoB is strongly correlated with Lp(a) levels and evidence suggests it has independent predictive value for aortic valve stenosis and coronary events beyond standard lipid panels.

Lp-PLA2 (Lipoprotein-Associated Phospholipase A2)

An enzyme carried on LDL (and to a lesser extent HDL) that cleaves oxidized phospholipids in the subendothelial space, generating lysophosphatidylcholine and oxidized free fatty acids — both potent pro-inflammatory mediators. Used clinically as a marker of plaque inflammation.

PON1 (Paraoxonase 1)

An HDL-associated esterase/lactonase with antioxidant properties. Hydrolyzes oxidized lipids on LDL and HDL, preventing lipoprotein oxidation and reducing pro-inflammatory OxPL generation. A key component of HDL’s protective function beyond its cholesterol-carrying role.

PON1 activity is reduced in smokers, diabetics, and those with chronic inflammation — contributing to a more pro-atherogenic LDL and dysfunctional HDL phenotype that standard cholesterol measurements miss.

hsCRP (High-Sensitivity C-Reactive Protein)

An acute-phase protein produced by the liver in response to IL-6 signaling. A marker of systemic inflammation. Elevated hsCRP (>2 mg/L) is associated with higher residual CV risk independent of LDL-C. May also be elevated by recent exercise, illness, or other conditions.

IL-6 / IL-1β

Key pro-inflammatory cytokines in atherogenesis. IL-1β (produced by macrophages and the NLRP3 inflammasome) drives hepatic IL-6 production, which in turn drives CRP synthesis and acute-phase responses. Both amplify endothelial activation, promote monocyte recruitment, and destabilize plaques.

NLRP3 Inflammasome

A cytosolic multiprotein complex in macrophages activated by cholesterol crystals, oxidized lipids, and other danger signals. Upon activation, it processes pro-caspase-1 which cleaves pro-IL-1β into active IL-1β — a potent local and systemic inflammatory driver. Cholesterol crystals in plaques are a direct NLRP3 activator.

The NLRP3 inflammasome contributes to our understanding of plaques as not just lipid deposits but active inflammatory lesions.

GlycA

An NMR-derived composite marker of systemic inflammation reflecting glycan modifications across multiple acute-phase proteins simultaneously (haptoglobin, alpha-1-acid glycoprotein, transferrin, others). More stable and reproducible than hsCRP, which can spike acutely with minor illness and recent exercise.

In some individuals who adopt low carbohydrate diets, GlycA may decrease substantially if elevated— one of the metabolic improvements tracked in LMHR and ketogenic diet research that tends to be underappreciated when some discussion focuses exclusively on LDL-C changes.

4. Genetic & Causal Inference Concepts

Genetic Conditions

heFH (Heterozygous Familial Hypercholesterolemia)

The most common monogenic lipid disorder (~1 in 250 globally). Caused by LDLR mutations (~85%), ApoB-R3500Q mutations (~5–10%), or PCSK9 gain-of-function variants (~1–2%). One defective LDLR allele reduces LDL clearance by ~50%, producing LDL-C typically 190–400 mg/dL from birth. Caused by LDLR mutations (~85%), ApoB-R3500Q mutations (~5–10%), or PCSK9 gain-of-function variants (~1–2%).

hoFH (Homozygous Familial Hypercholesterolemia)

Rare (~1 in 300,000). Most commonly caused by biallelic LDLR dysfunction which produces LDL-C typically above 400–500 mg/dL, sometimes exceeding 1000 mg/dL. MI in childhood is the natural history. Planar and tendon xanthomas are hallmarks.

Familial Combined Hyperlipidemia (FCHL)

The most common familial lipid disorder (~1 in 100–200). Characterized by elevated LDL-C, elevated TG, or both — varying within and between family members. Driven by hepatic ApoB overproduction and influenced by metabolic status. Polygenic with strong environmental modifiers.

Familial Dysbetalipoproteinemia (Type III Hyperlipidemia)

Generally amplified by the ApoE2/E2 genotype (~1% of the population carries two E2 alleles, but only ~10% of them develop Type III). ApoE2 binds poorly to the LDL receptor, impairing remnant clearance. The result: dramatic accumulation of IDL and chylomicron remnants, elevated cholesterol and TG, with distinctive palmar xanthomas (xanthoma striata palmaris) and high atherosclerotic risk.

The penetrance is incomplete — most ApoE2/E2 individuals do not develop Type III without a metabolic ‘second hit’ (obesity, hypothyroidism, diabetes). Some lipidologists recommend testing ApoE genotype in anyone with mixed hyperlipidemia and both TC and TG elevated.

Familial Chylomicronemia Syndrome (FCS / LPLD)

A rare autosomal recessive disorder caused by biallelic loss-of-function mutations in LPL or its essential cofactors (ApoC-II, ApoA-V, LMF1, GPIHBP1). Results in near-complete failure of TG hydrolysis. TG routinely exceeds 880 mg/dL. Clinical hallmarks: milky plasma, eruptive xanthomas, lipemia retinalis, recurrent pancreatitis.

Tangier Disease

A rare autosomal recessive disorder caused by ABCA1 loss-of-function mutations. Without ABCA1, cells cannot efflux cholesterol to form nascent HDL. Resulting phenotype: near-absent HDL-C, orange tonsillar deposits, peripheral neuropathy, hepatosplenomegaly, and premature atherosclerosis.

Tangier disease demonstrates that ABCA1-mediated cholesterol efflux is non-redundant — other efflux pathways cannot compensate when ABCA1 is absent. It validated the mechanistic importance of ABCA1 in human lipid biology.

Sitosterolemia

Rare autosomal recessive disorder caused by ABCG5 or ABCG8 mutations. Massively elevated plasma plant sterols, tendon and tuberous xanthomas, and premature atherosclerosis — sometimes presenting in childhood. Sometimes misdiagnosed as FH because plant sterols cross-react in cholesterol assays.

Lp(a) Elevation

Unlike most lipid traits, plasma Lp(a) is considered to be ~90% genetically determined by variants at the LPA locus, although lifestyle may additionally influence levels via inflammatory signaling and some diet-related factors. For example, some evidence suggests that low carbohydrate diets may decrease Lp(a) in certain individuals, although currently the mechanism is unclear.

Small isoforms of apo(a) tend to result in a higher genetic baseline of Lp(a). Median population Lp(a) is ~20 mg/dL but the distribution is highly skewed — ~20% of people carry levels above 50 mg/dL which is associated with elevated CV risk.
Because of the mass of Lp(a) can vary by the isoform, some lipidologists suggest that Lp(a) should preferentially be measured in nmol/L which measures the number of particles.

Polygenic & Complex Genetics

Polygenic Risk Score (PRS)

An aggregate genetic score summing the weighted effects of thousands of common variants on a quantitative trait (LDL-C, TG, HDL-C, or ASCVD risk directly). Each individual variant has a small effect; in aggregate they explain a meaningful fraction of the population variance in lipid levels and cardiovascular risk.

Causal Inference

Mendelian Randomization (MR)

An epidemiological method using genetic variants as instrumental variables — proxies for a lifetime exposure — to estimate causal effects on outcomes, exploiting the random allocation of alleles at conception as a natural experiment. Key assumptions: (1) The genetic variant is robustly associated with the exposure (relevance). (2) The variant is independent of confounders (independence). (3) The variant affects the outcome only through the exposure (exclusion restriction).

Horizontal Pleiotropy

When a genetic variant affects the outcome through pathways other than the primary exposure of interest. This violates the exclusion restriction assumption and can bias MR estimates. For example, if a ‘LDL-C genetic instrument’ also independently affects blood pressure, its effect on CV events cannot be attributed entirely to LDL-C.

Methods like MR-Egger regression, weighted median, and MR-PRESSO are sensitivity analyses designed to detect and correct for horizontal pleiotropy — essentially attempts at quality checks when interpreting any MR study.

Vertical Pleiotropy

When a genetic variant affects both the exposure and the outcome, but only through the causal pathway under study. This is acceptable in MR and does not violate the exclusion restriction.

Exclusion Restriction Assumption

The third and most frequently violated core assumption of Mendelian randomization. Requires that the genetic instrumental variable affects the outcome only through the exposure of interest — not through any other biological pathway.

The exclusion restriction is untestable directly. MR analyses rely on biological plausibility, sensitivity analyses, and testing instruments from different biological mechanisms to evaluate whether results are robust to potential violations.

Instrumental Variables (IV)

Genetic variants used in Mendelian randomization as instruments for a modifiable exposure. A valid IV must be (1) strongly associated with the exposure, (2) independent of confounders, and (3) only related to the outcome through the exposure. Weak instruments (low F-statistic) introduce bias toward the observational estimate.

Colocalization

A statistical analysis that tests whether the same genetic variant is driving association signals for two traits simultaneously (e.g., LDL-C and coronary artery disease). If the same SNP drives both associations, this strengthens the causal argument; if different SNPs underlie each, it may suggest confounding or pleiotropy.

5. Biomarkers & Advanced Testing

Lipid Metrics & Ratios

ApoB

The direct plasma measure of total particle number across all ApoB-containing lipoproteins. One ApoB per particle. Captures LDL, VLDL, IDL, Lp(a), and chylomicrons in a single number. Measured by immunoassay. Endorsed as a primary or co-primary treatment target by multiple international lipid guidelines.

In mainstream lipidology, ApoB is considered the most informative single lipid test for cardiovascular risk stratification. It outperforms LDL-C in people with elevated TG, insulin resistance, or discordant lipid patterns — though as with all lipid markers, the metabolic context in which it is elevated may be relevant.

Non-HDL-C, LDL-C, HDL-C, TG

These terms are defined in detail in Section 1. In the context of the biomarker panel, the key clinical point is how they relate to each other.

The standard lipid panel (TC, LDL-C, HDL-C, TG) provides a starting point. Its limitations: LDL-C is calculated (not directly measured in most labs), fails at high TG, and captures neither particle number nor remnant burden fully.

LDL-C/ApoB Ratio

The ratio of LDL cholesterol to ApoB concentration. Reflects the average cholesterol content per LDL particle. A low ratio = many small, cholesterol-poor particles (Pattern B, high ApoB per unit LDL-C). A high ratio = fewer, cholesterol-rich, large buoyant particles (Pattern A, lower ApoB per unit LDL-C).

In LMHR individuals, the LDL-C/ApoB ratio is often high (>1.2) — consistent with large, buoyant LDL. This distinguishes the LMHR pattern from the metabolic syndrome pattern where the ratio is low (many small dense particles).

TG/HDL-C Ratio

A simple surrogate for insulin resistance and small dense LDL predominance. A higher ratio is associated with Pattern B dyslipidemia and metabolic dysfunction. Easy to calculate from a standard fasting lipid panel.

The TG/HDL-C ratio is often considered one of the most informative simple metrics derivable from a standard lipid panel because it can be used as a proxy measure for metabolic health. However, it can be helpful to take the ratio in context, as some factors increase triglycerides without decreasing HDL-C - like habitual alcohol consumption, or caffeine during the fasting period.

Remnant Cholesterol

(Defined in Section 1.) Calculated as TC − HDL-C − LDL-C from a standard fasting panel.

Lp(a)

(Defined in Section 1.) Measured in mg/dL or nmol/L — an important distinction. Nmol/L measures particle number; mg/dL measures mass (which includes Apo(a) protein, size of which varies between individuals). Many guidelines now prefer nmol/L.

Advanced Lipid Testing

NMR Lipoprofile (LDL-P, LPIR)

Nuclear Magnetic Resonance spectroscopy of a plasma sample measuring lipoprotein particle concentrations directly (LDL-P, HDL-P, VLDL-P) and particle sizes. Also generates LPIR (Lipoprotein Insulin Resistance Index) — a score derived from six particle metrics that correlates with insulin resistance and predicts type 2 diabetes.

In LMHR individuals on low-carb diets: LDL-P and ApoB rise, but many LMHR report LPIR falling substantially — a metabolic divergence that standard panels cannot capture and that may factor in to the LMHR risk question.

Ion Mobility

An alternative method to NMR for lipoprotein particle sizing and quantification. Uses differential electrical mobility of particles in a gas phase. Provides high-resolution particle size distribution data. Pioneered by Ronald Krauss.

ApoB Immunoassay

Direct immunoturbidimetric or immunonephelometric measurement of ApoB in plasma. Widely available in clinical laboratories, inexpensive, and robust. Does not require fasting. Now endorsed by the American Heart Association and European Atherosclerosis Society as a preferred risk marker.

Sterol / Cholesterol Balance Testing

Plant Sterols (Sitosterol, Campesterol)

Absorbed from dietary plant foods via NPC1L1; normally kept very low by ABCG5/G8 efflux back into the gut. Elevated plasma levels indicate high intestinal cholesterol absorption efficiency — the hyperabsorber phenotype.

Synthesis Markers (Desmosterol, Lathosterol)

Precursors in the cholesterol biosynthesis pathway. Lathosterol is a late-stage intermediate in the Kandutsch-Russell branch of cholesterol synthesis; desmosterol marks the Bloch branch. Both are elevated when endogenous cholesterol synthesis is high — identifying the hypersynthesizer phenotype

Metabolic Markers

Insulin / C-Peptide / HOMA-IR

Fasting insulin reflects insulin secretory demand and is elevated in insulin resistance. C-peptide is co-secreted with insulin and is a more stable measure of endogenous insulin production. HOMA-IR = fasting insulin (µU/mL) × fasting glucose (mmol/L) / 22.5 — a validated surrogate for insulin resistance.

C-peptide is one of the most sensitive early markers of metabolic dysfunction, often rising years before fasting glucose or HbA1c crosses clinical thresholds. Elevated c-peptide is associated with VLDL overproduction, high TG, low HDL-C, and Pattern B — the full atherogenic dyslipidemia.

HbA1c

Glycated hemoglobin — reflects average blood glucose over the preceding ~3 months. Standard diagnostic and monitoring tool for diabetes and prediabetes. A component of comprehensive cardiometabolic risk assessment alongside lipids.

HbA1c is less sensitive than fasting insulin for detecting early insulin resistance — it only rises once glycemic dysregulation is established.

Adiponectin

An adipokine (fat tissue-derived hormone) with insulin-sensitizing, anti-inflammatory, and fatty acid oxidation-promoting properties. Inversely associated with BMI, visceral adiposity, and insulin resistance. Low adiponectin is a feature of metabolic syndrome.

Adiponectin is suggested to be anti-atherogenic through multiple pathways: improved insulin sensitivity, reduced VLDL overproduction, suppressed endothelial inflammation, and increased fatty acid oxidation in muscle. Its reduction with visceral fat accumulation may be a mechanistic link between obesity and dyslipidemia.

Homocysteine

A sulfur-containing amino acid produced during methionine metabolism. Elevated plasma homocysteine (hyperhomocysteinemia) is associated with endothelial damage, increased thrombosis risk, and CV events. Raised by B12, B6, and folate deficiency, chronic kidney disease, and certain genetic variants (MTHFR).

GlycA

(Defined in Section 3 under Inflammation.) In the biomarker context: measured on the same NMR lipoprofile panel that generates LDL-P and LPIR. Provides a composite systemic inflammation score from a single blood draw alongside particle data.

The practical value of GlycA on an NMR panel: it allows simultaneous assessment of particle burden (LDL-P), metabolic health (LPIR), and systemic inflammation (GlycA) — a three-dimensional metabolic snapshot from one test.

6. Phenotypes & Models (TFP_-Relevant)

Phenotypes

LMHR (Lean Mass Hyper-Responder)

A phenotype observed predominantly in lean, metabolically healthy, and often physically active individuals who adopt very-low-carbohydrate (ketogenic or near-ketogenic) diets. Characterized by a distinct lipid triad: LDL-C exceeding 200 mg/dL, HDL-C above 80 mg/dL, and TG below 70 mg/dL — occurring simultaneously. First systematically described and named by Dave Feldman.

The central scientific question surrounding LMHR is not whether LDL-C is elevated — it clearly is — but whether this elevation in this specific metabolic context (lean, insulin-sensitive, low TG, high HDL, low inflammation) carries equivalent atherosclerotic risk to LDL-C elevation occurring in the context of insulin resistance, obesity, and dyslipidemia. Prospective CAC and CCTA data are accumulating to answer this.

Hyperabsorber

An individual with constitutively high intestinal cholesterol absorption, identifiable by elevated plasma sitosterol and campesterol on a sterol panel.

Hypersynthesizer

An individual with constitutively elevated endogenous cholesterol synthesis, identifiable by elevated lathosterol and/or desmosterol on a sterol panel. HMG-CoA reductase activity is high, driving excess hepatic cholesterol production.

Hypersynthesizers are at the other end of the synthesis/absorption spectrum from hyperabsorbers. Some clinicians use their understanding of these phenotypes to individualize lipid therapy selection to the phenotype of the patient.

Metabolic Syndrome / Insulin-Resistant Phenotype

A cluster of interrelated metabolic abnormalities driven by insulin resistance and visceral adiposity. Lipid phenotype: elevated TG, low HDL-C, small dense LDL (Pattern B), elevated remnant cholesterol, and elevated ApoB — often with normal or only modestly elevated LDL-C.

In metabolic syndrome, ApoB level may be higher than the individual’s LDL-C level may suggest — an individual may have LDL-C of 100 mg/dL but ApoB of 130 mg/dL or above, with many small dense LDL particles. This is the opposite of the LMHR pattern and illustrates why phenotypic context may play a role in interpreting lipid values.

Conceptual Models

Lipid Energy Model (LEM)

A hypothesis developed by Dave Feldman and colleagues proposing that in lean, metabolically healthy, carbohydrate-restricted individuals, the observed elevation in LDL-C and ApoB reflects upregulated lipid mobilization and systemic transport to deliver fatty acids and ketones for energy — a physiological adaptation to absence of dietary carbohydrate, not a dysregulated atherogenic state. The model generates specific, testable, directional predictions: LDL-C should attenuate when carbohydrate intake is sufficiently increased; it should track with energy demand markers.

The LEM remains a hypothesis — it has reasonable mechanistic plausibility and directional n-of-1 predictions, but long-term prospective imaging data in LMHR individuals are required (and being collected) to establish whether this LDL-C elevation translates to plaque accumulation at the rate standard risk equations would predict.

Discordance

The clinical phenomenon where two markers expected to correlate do not. In lipidology, most commonly: LDL-C vs. ApoB. High LDL-C with low ApoB = large, buoyant, cholesterol-rich particles (Pattern A — fewer trucks, more cargo each). Low LDL-C with high ApoB = small, dense, cholesterol-poor particles (Pattern B — more trucks, less cargo each). The second pattern has higher CV risk per unit of LDL-C.

The LMHR phenotype characteristically shows high LDL-C with proportionally lower ApoB — a Pattern A discordance suggesting fewer (in proportion), larger, cholesterol-rich particles. The metabolic syndrome shows the opposite. Understanding discordance direction may be helpful in guiding risk interpretation in non-standard lipid patterns.

Response-to-Retention Hypothesis

The foundational mechanistic model of atherosclerosis initiation endorsed by mainstream cardiology. Proposes that the primary initiating event is the retention of ApoB-containing lipoproteins in the subendothelial matrix — via binding of ApoB to proteoglycans — where they are then modified, oxidized, and taken up by macrophages. The endothelium is not simply a passive barrier; it is a selective filter, and ApoB retention behind it is the pathogenic trigger.

7. Clinical Outcomes & Disease States

Major Cardiovascular Events

MACE (Major Adverse Cardiovascular Events)

A composite clinical trial endpoint. 3-point MACE: cardiovascular death, non-fatal MI, non-fatal ischemic stroke. 4- or 5-point MACE adds revascularization and/or hospitalization for unstable angina. The primary endpoint in most large CV outcomes trials.

The specific MACE definition varies between studies — always verify components before comparing effect sizes. A 15% MACE reduction means something different when the composite includes softer endpoints like revascularization versus harder ones like CV death.

CHD Events / STEMI / NSTEMI

CHD events include any manifestation of coronary heart disease: stable angina, unstable angina, NSTEMI (non-ST-elevation myocardial infarction — partial coronary occlusion, troponin elevation, no ST elevation on ECG), and STEMI (ST-elevation MI — typically complete coronary occlusion, ST elevation, requiring emergent revascularization).

STEMI and NSTEMI differ in mechanism as well as ECG appearance: STEMI most commonly results from plaque rupture and complete thrombotic occlusion; NSTEMI more often reflects plaque erosion or partial occlusion. Lipid-rich plaque burden is the shared upstream risk factor for both.

Ischemic Stroke / CV Death

Ischemic stroke — cerebral infarction due to arterial occlusion — is part of the MACE composite and is causally linked to atherosclerosis (carotid and intracranial) and cardioembolic sources. CV death includes death from coronary heart disease, stroke, arrhythmia, and heart failure.

Other Lipid-Related Conditions

Acute Pancreatitis (TG-driven)

Severe pancreatic inflammation triggered by extreme hypertriglyceridemia (typically >880–1000 mg/dL). Chylomicrons obstruct pancreatic microvascular flow; locally hydrolyzed free fatty acids from TG are directly toxic to acinar cells. Recurrent pancreatitis can lead to exocrine insufficiency and chronic pain.

Pancreatitis risk requires a different TG threshold than atherosclerosis risk. Modest TG elevation (150–500 mg/dL) is associated with increased atherosclerotic risk; pancreatitis risk becomes substantial above ~500–880 mg/dL and severe above 1000 mg/dL.

Xanthomas / Xanthelasmas

Xanthomas: lipid deposits in skin and tendons. Types include tendon xanthomas (Achilles, extensor tendons — classic for FH), tuberous xanthomas (over joints — seen in hoFH and Type III), and eruptive xanthomas (small papules over buttocks/trunk — pathognomonic for severe hypertriglyceridemia). Xanthelasmas are periorbital cholesterol deposits, more common but not specific for hyperlipidemia.

Lipemia Retinalis

Creamy-white appearance of retinal blood vessels on fundoscopy caused by light scattering through TG-laden chylomicrons in retinal capillaries. Occurs at TG levels typically above 2000–3000 mg/dL. Pathognomonic for severe chylomicronemia.

NAFLD / MASLD

Non-Alcoholic Fatty Liver Disease / Metabolic Dysfunction-Associated Steatotic Liver Disease. Hepatic fat accumulation (steatosis) driven by insulin resistance, de novo lipogenesis, and excess free fatty acid flux to the liver. The spectrum includes simple steatosis, MASH (steatohepatitis with inflammation), fibrosis, and cirrhosis. MASLD is the updated nomenclature emphasizing metabolic etiology.

NAFLD/MASLD is strongly associated with elevated TG, low HDL-C, high ApoB, and elevated remnant cholesterol — the full atherogenic dyslipidemia. It is both a consequence of and contributor to insulin resistance and VLDL overproduction.

Calcific Aortic Valve Stenosis

Progressive calcification and stiffening of the aortic valve leaflets leading to obstruction of left ventricular outflow. Shares pathological features with atherosclerosis. Strongly associated with Lp(a) elevation (via OxPL-ApoB-driven osteoblastic differentiation in valve tissue) and with LDL-C.

8. Imaging & Direct Disease Measurement

Imaging Modalities

CAC Scoring (Coronary Artery Calcium Score)

Non-contrast CT quantifying calcified coronary plaque by the Agatston score (area × density weighting). Reflects cumulative lifetime atherogenic burden. Score of 0 = very low near-term event risk; above 100 or at or above the 75th percentile for age/sex/ethnicity = elevated risk.

CAC = 0 is often considered to be one of the most powerful negative risk predictors — sometimes called the “Power of Zero” due to the very low short-moderate term heart disease risk associated in this context.

CCTA (Coronary CT Angiography)

Contrast-enhanced CT of the coronary arteries. Visualizes both calcified and non-calcified plaque, stenosis severity, and high-risk plaque features (LAP, positive remodeling, napkin-ring sign, spotty calcification). The only non-invasive modality that directly characterizes plaque composition.

IVUS (Intravascular Ultrasound)

Catheter-based ultrasound deployed inside the coronary artery during invasive angiography. Provides cross-sectional images of the arterial wall, enabling volumetric plaque quantification (TAV, PAV). The gold standard for serial plaque progression/regression imaging in clinical trials.

OCT (Optical Coherence Tomography)

A high-resolution catheter-based coronary imaging technique using near-infrared light. Spatial resolution ~10x higher than IVUS. Provides detailed imaging of fibrous cap thickness, lipid pool composition, and micro-features of plaque vulnerability (including visualization of TCFA).

OCT is the best available tool for identifying and measuring thin-cap fibroatheromas in vivo. Its high resolution allows direct cap thickness measurement below the 65 µm threshold that defines TCFA — something IVUS cannot reliably do.

CIMT (Carotid Intima-Media Thickness)

Ultrasound measurement of the combined thickness of the intima and media layers of the common carotid artery. A non-invasive marker of subclinical atherosclerosis. Correlates with cardiovascular risk factors and predicts future CV events at the population level.

CIMT has fallen somewhat out of favor as a primary risk stratification tool (CAC scoring outperforms it) but it is still considered useful in younger populations (where CAC is typically 0) and as a research tool for studying early vascular changes.

Cardiac MRI

Magnetic resonance imaging of the heart. Provides excellent assessment of myocardial structure, function, tissue characterization (edema, fibrosis, scar), and great vessel anatomy without radiation. Also enables aortic and carotid plaque characterization in research settings.

Cardiac MRI is the reference standard for myocardial viability assessment after MI, guiding revascularization decisions. In lipid research, it is used for carotid and aortic plaque characterization alongside more commonly used modalities.

Key Measurements

TAV, PAV, LAP, TCFA, and Positive Remodeling are defined in their primary sections (Section 3 and Section 8 above).

9. Study Designs & Evidence Framework

Study Types

Prospective Cohort Study

Participants are enrolled and followed forward in time, with exposures (lipid levels, diet, medications) measured at baseline. Outcomes (MI, stroke, death) are ascertained prospectively. Examples: Framingham Heart Study, UK Biobank, MESA.

Prospective cohort studies establish associations and generate hypotheses. Confounding — the presence of unmeasured factors that correlate with both exposure and outcome — is their primary limitation. They cannot definitively establish causality.

Retrospective Cohort Study

Uses existing data (medical records, claims databases, registries) to reconstruct exposure histories and outcomes for a defined group. Faster and cheaper than prospective studies. Subject to information bias and limited by quality of available data.

Case-Control Study

Compares individuals with a disease (cases) to those without (controls), looking backward to assess prior exposures. Efficient for rare outcomes. Subject to recall bias and selection bias in control selection.

Case-control studies were critical early in establishing the relationship between cholesterol levels and MI risk — before long-term prospective cohort data were available at scale.

Randomized Controlled Trial (RCT)

Participants are randomly assigned to treatment or control conditions. Randomization distributes known and unknown confounders equally between groups, making the RCT the gold standard for establishing therapeutic efficacy. Double-blinding prevents outcome ascertainment bias.

Mendelian Randomization

(Defined in detail in Section 4.) In the study design context: a hybrid approach combining genetic epidemiology with causal inference methodology. Uses genetic variants as natural experiments, bridging the gap between observational associations and RCT-level causal evidence.

Meta-Analysis / Systematic Review

A systematic review comprehensively identifies and synthesizes all studies on a topic using predefined criteria. A meta-analysis statistically combines results across studies to generate pooled effect estimates with greater statistical power than any individual study.

Key Concepts

Confounding

A confounding variable is associated with both the exposure and the outcome, creating a spurious or distorted association. The classic example in lipid epidemiology: people who eat more saturated fat may also exercise less, smoke more, and have higher BMI — making it difficult to isolate the specific effect of dietary fat on CV outcomes in observational studies.

Bias (Selection, Information, Publication)

Selection bias: systematic difference between study participants and the target population. Information/recall bias: systematic errors in measuring or reporting exposures. Publication bias: tendency for positive results to be published and negative results to go unreported, inflating apparent effect sizes in meta-analyses.

Publication bias in nutrition research may be substantial — small studies showing dramatic dietary effects are more likely to be published than null results. This is one reason dietary epidemiology findings must be interpreted with considerably more caution than large RCT data.

Intention-to-Treat (ITT) vs Per-Protocol Analysis

Intention-to-treat analysis includes all randomized participants in their assigned groups regardless of whether they completed treatment. Per-protocol analysis includes only participants who adhered to the protocol. ITT preserves randomization and is the primary approach in RCTs; per-protocol can assess efficacy in ideal adherents.

ITT is conservative by design — it dilutes apparent treatment effects because non-adherent participants in the treatment arm are still counted as treated. This is still considered appropriate for real-world effectiveness questions but can underestimate biological efficacy.

Odds Ratio (OR)

A measure of association used in case-control studies and logistic regression. The odds ratio compares the odds of an outcome occurring in an exposed group to the odds of it occurring in an unexposed group. An OR of 1.0 means no association; above 1.0 suggests the exposure is associated with higher odds of the outcome; below 1.0 suggests lower odds. In rare-disease settings, the OR closely approximates the relative risk (RR). In common-outcome settings, the OR and RR diverge — and ORs may overstate the magnitude of association compared to RR.

Odds ratios are frequently reported in observational lipid studies — and may be misread as relative risks. When an outcome is common (>10% prevalence), an OR of 1.5 does not mean 50% more likely in the way a relative risk of 1.5 would.

Absolute vs. Relative Risk / NNT / NNH

Relative risk reduction (RRR): the proportional reduction in event rate between treatment and control groups. Absolute risk reduction (ARR): the difference in event rates. Number needed to treat (NNT) = 1/ARR — how many patients must be treated for one to benefit. Number needed to harm (NNH) is the equivalent for adverse effects.

RRR is consistent across risk levels; ARR and NNT vary enormously by baseline risk. A 25% relative risk reduction sounds the same in primary and secondary prevention — but the NNT may be 200 in low-risk primary prevention vs. 15 in very-high-risk secondary prevention. Both numbers matter for informed shared decision-making.

10. Landmark Cohorts & Registries

Framingham Heart Study + Offspring

The Framingham Heart Study (est. 1948) followed residents of Framingham, Massachusetts, prospectively establishing the foundational concept of cardiovascular risk factors — including serum cholesterol, blood pressure, and smoking. The Offspring Study extended surveillance to the second generation, enabling familial lipid research.

Framingham generated the first risk factor paradigm in cardiology. Its limitation: predominantly white, northeastern US population limits generalizability.

MESA (Multi-Ethnic Study of Atherosclerosis)

A diverse prospective cohort (6,814 participants, 45–84 years, six US sites) specifically designed to study subclinical cardiovascular disease across white, Black, Hispanic, and Chinese-American populations. Strong data on CAC, carotid ultrasound, and lipid markers including advanced testing.

MESA data on coronary artery calcium has been foundational for understanding CAC’s risk-reclassification value across ethnic groups and for calibrating CAC-based risk prediction algorithms used in current ACC/AHA guidelines.

UK Biobank

A large-scale biomedical database of over 500,000 UK participants with extensive genetic, imaging, biochemical, and health record data. A major resource for Mendelian randomization, genome-wide association studies, and polygenic risk score development in lipid research.

Western Denmark Heart Registry

A large population-based registry with extensive data on cardiac imaging including coronary CT and CAC scoring. Provides real-world evidence on lipid management, statin use, and imaging-based risk stratification in a large Northern European population.

Particularly valuable for CAC-based risk reclassification data in clinical practice settings — bridging the gap between academic trial populations and routine clinical populations.

ARIC (Atherosclerosis Risk in Communities)

A prospective cohort of ~15,800 adults from four US communities, followed from the mid-1980s. Strong data on LDL-C, lipoprotein subclasses, inflammation markers (hsCRP, fibrinogen), and incident MI, stroke, and heart failure outcomes.

ARIC has been a key source of data on advanced lipid biomarkers (ApoB, LDL-P, sdLDL) in relation to CV outcomes, and on racial disparities in cardiovascular risk and lipid management.

Copenhagen General Population Study (CGPS)

A large Danish prospective cohort (over 100,000 participants) that has generated some of the most cited Mendelian randomization data on remnant cholesterol, Lp(a), and HDL-C causality. The Copenhagen City Heart Study is an earlier related cohort.

11. Dietary & Fatty Acid Context

Major Fatty Acids

Saturated Fat (SFA)

Fatty acids with no double bonds in their carbon chains. Found primarily in animal products (meat, dairy, butter) and some plant oils (coconut, palm). SFA consumption raises LDL-C in most people by reducing hepatic LDL receptor expression and increasing VLDL production — though the magnitude varies substantially between individuals.

MUFA (Monounsaturated Fatty Acids)

Fatty acids with one double bond. Found in olive oil, avocado, nuts, and some animal fats. When used to replace SFA in the diet, MUFA generally produces no change or a slight decrease to the LDL-C/HDL-C ratio without raising TG. The primary fat in the Mediterranean dietary pattern. The primary fat in the Mediterranean dietary pattern.

PUFA (Polyunsaturated Fatty Acids)

Fatty acids with two or more double bonds. Include both omega-6 (linoleic acid and derivatives) and omega-3 (ALA, EPA, DHA) classes. Replacing SFA with PUFA, particularly omega-6, lowers LDL-C more than MUFA. However, the effects on HDL-C, TG, and inflammation vary by PUFA type and context.

Omega-3 Fatty Acids (EPA / DHA / ALA)

EPA (eicosapentaenoic acid) and DHA (docosahexaenoic acid) are long-chain omega-3 fatty acids found in marine sources (fish oil). ALA (alpha-linolenic acid) is the short-chain plant-derived omega-3 found in flaxseed, walnuts, and chia. Human conversion of ALA to EPA/DHA is inefficient (<5–10%).

Omega-6 Fatty Acids (Linoleic Acid)

The dominant PUFA in most Western diets, found primarily in seed oils (soybean, corn, sunflower, safflower). Linoleic acid (LA) is the essential omega-6 fatty acid. High dietary omega-6 relative to omega-3 is a feature of the modern Western dietary pattern.

The omega-6/omega-3 ratio in the typical Western diet is estimated at 15–20:1 versus an evolutionary estimate of 1:1 to 4:1.

12. Key Dietary Intervention Trials (TFP_-Relevant)

MHERO Study (VLCD vs DASH)

A prospective randomized trial comparing a very-low-carbohydrate diet (VLCD) against the DASH diet in patients with hypertension, examining effects on blood pressure, body composition, and cardiometabolic biomarkers including lipids. Ongoing — represents an important head-to-head comparison of two clinically endorsed dietary approaches.

MHERO is notable for its focus on a population (hypertension) where both dietary approaches have clinical credibility, enabling a genuinely informative comparison of their differential lipid effects in a pre-defined high-risk group.

DIETFITS (Gardner et al., JAMA 2018)

A Stanford University 12-month RCT in 609 adults comparing a healthy low-fat diet versus a healthy low-carbohydrate diet for weight loss. Both diets achieved similar mean weight loss and reduced refined starches and added sugars. The low-carbohydrate arm produced greater TG reductions and HDL-C increases; LDL-C responses were heterogeneous across both arms.

KETO-MED

A head-to-head RCT comparing a ketogenic Mediterranean diet (very-low-carbohydrate with Mediterranean fat sources) against a standard Mediterranean diet. The ketogenic arm showed superior improvements in glycemic control, TG reduction, and HDL-C increases, with broadly comparable LDL-C changes in most participants.

Virta Health Trial (Continuous Care Intervention)

A non-randomized but rigorously tracked prospective study of a ketogenic diet delivered through continuous digital care in patients with type 2 diabetes. Two-year results: sustained HbA1c reduction, substantial medication reduction (including insulin), improved TG, improved HDL-C, and weight loss — despite frequent increases in LDL-C in some participants.

The Virta trial is the most detailed longitudinal cardiometabolic dataset on ketogenic diets in a clinical population to date.

A Note on Accuracy

Every definition in this glossary aims to accurately represent both the current scientific evidence and clinical consensus as of 2026.

This is a living reference. As new data emerge, entries will be updated.

Share

Leave a comment

What is a Lean Mass Hyper-Responder?

A lean mass hyper-responder (LMHR) is a distinct blood lipid pattern (phenotype) observed in some individuals who adopt a very low-carbohydrate or ketogenic diet.

This observed pattern is defined by 3 cut points (the “lipid triad”):

• LDL cholesterol (LDL-C): > 200 mg/dl

• HDL cholesterol (HDL-C): > 80 mg/dl

• Triglycerides (TG): <70 mg/dl

While low triglycerides and high HDL-C are generally considered favorable markers of cardiovascular risk, LDL-C levels this high typically raise significant concern for long-term cardiovascular disease risk in mainstream medicine.

Each of these findings can show up on their own but seeing them together is remarkably unusual. When they do cluster like this, it suggests a coordinated metabolic response rather than a set of unrelated anomalies.

Subscribe now

Why does this happen?

A possible explanation for this pattern is the Lipid Energy Model (LEM), which focuses on lipoproteins in their role of energy delivery vehicles and how it can potentially explain this pattern.

When carbohydrate intake is very low, the body shifts toward fat as its primary fuel.

The liver responds by transporting more fat through the bloodstream in particles called very-low density lipoproteins (VLDL). This response is often more pronounced in lean individuals, who have less stored fat to draw from directly.

Under the LEM, VLDL particles carry these fats, including triglycerides (TG), to energy-hungry tissues, which take them up rapidly — leaving less TG in circulation (low TG). The VLDL remodel into cholesterol-rich LDL (high LDL-C) as their surface components are transferred to HDL (high HDL-C).

The result is the familiar lipid triad: high LDL-C (and ApoB), high HDL-C, and low triglycerides.12

Share

Why Does This Matter?

LDL-C, or better yet, ApoB is associated with cardiovascular disease and is considered a cornerstone of preventive cardiology.

ApoB (apolipoprotein B) is a protein that sits on the surface of LDL and other cholesterol-carrying particles. Because each particle has one ApoB, measuring ApoB tells us how many particles are circulating in the bloodstream that could deposit cholesterol into the artery wall.

Of note, LDL-C and ApoB typically rise together in LMHRs, meaning that if LDL-C is high, ApoB is usually elevated as well.

However, a given ApoB level does not explain why this level is elevated. In our view, that's the critical question.

Metabolic Dysfunction

In most populations, high ApoB clusters with abdominal obesity, inactivity, insulin resistance, hypertension, high triglycerides, low HDL-C, and small, dense LDL.

In this setting, ApoB may reflect metabolic dysfunction rather than independent risk.3

Genetic Abnormalities

ApoB may be elevated in certain genetic conditions (e.g. Familial Hypercholesterolemia) that alter lipoprotein handling. This may also result in a constellation of other pathological changes including broader disruptions in lipid metabolism, immune function, blood clotting, etc.45

LMHR: A Whole New Context

LMHRs may represent a completely different category: adaptive ApoB elevation, where higher particle numbers are disentangled from insulin resistance, abdominal obesity, physical inactivity, or other lipid disturbances, including genetic abnormalities.

In LMHR, ApoB may be higher due to more fat energy utilization and often tracks with favorable markers such as low TG/HDL-C ratios and lower BMI. Whether cardiovascular risk is equivalent across these different reasons remains an open and important question.

Therapeutic Nutritional Ketosis

This distinction is increasingly important because ketogenic diets are now used for therapeutic purposes beyond weight loss or type 2 diabetes.

Applications for therapeutic nutritional ketosis (not exhaustive):

• Bipolar disorder

• Depression

• Anxiety

• Schizophrenia

• Parkinson’s disease

• Migraine

• Inflammatory bowel disease

• Disordered eating behaviors

• Lipedema

• Alzheimer’s disease

• Multiple sclerosis (MS)

As these dietary patterns are adopted across a wider range of body types, LDL-C and ApoB elevations, particularly in leaner individuals, may become more common.6

Bottom Line

The emergence of LMHRs reinforces the need to understand the underlying biology to determine what this specific change in LDL-C means for these individuals.

Leave a comment

From the desk of Own Your Labs

As a subscriber to TFP_ Newsletter, you can submit questions for upcoming podcast guests, “Ask Me Anything” episodes, general inquiries about the research, and more.

Submit your question

What Is the Feldman Protocol_?

The Feldman Protocol_ (TFP_) is a podcast about metabolism, nutrition, and the philosophy of science, with deep dives into cholesterol and heart disease. It also spans topics from mental health to self-experimentation to the pros and cons of scientific debate on social media.

The Feldman Protocol podcast, hosted by Dave Feldman, is built around long-form conversations. This gives guests time to settle in and allows listeners to understand what first set them on their path, how their thinking developed, and where they stand today.

The Origins of TFP_

Dave was a software engineer with no medical background or formal training in medicine who started paying closer attention to his own metabolic health after noticing his blood sugar creeping up year after year alongside a strong family history of diabetes.

When he flagged it with his doctor, the response was: This is the second year in a row of high glucose, but we’ll keep monitoring it.

But that response didn’t quite sit right.

At some point, he came to a simple realization: As an engineer, I should be able to figure this out…and figure it out now.

He dug into the literature on his own, found online forums discussing a low-carbohydrate, high-fat (LCHF) diet, and tried it — and it worked. His blood sugar came down, his health improved. By every measure he was tracking, it appeared he had solved the problem.

Then he got the next blood test…

The 8.5×11” Sheet of Paper That Changed Everything

The lab results, on an inconspicuous sheet of 8.5×11" paper, completely threw him: while his blood sugar had improved dramatically, his LDL cholesterol (the "bad" cholesterol often linked to heart disease) had shot through the roof.

He’d thought he was on a new path to health. Now none of it made sense.

When he tried conventional fixes like cutting saturated fat and reducing calories, his LDL went up. Not down. Up.

That’s when concern turned into something closer to obsession.

“It was the 8.5×11” piece of paper that changed my life.” - Dave

What followed was years of self-experimentation: dozens and dozens of blood draws, a photo of every single bite of food he ate, spreadsheets tracking how each dietary change moved his lab numbers, and reverse-engineering his own metabolism.

He started a blog on CholesterolCode.com, where he chronicled his metabolic journey.

This work led to two things that have since made their way into the peer-reviewed literature: the term Lean Mass Hyper-Responder (LMHR) and the Lipid Energy Model (LEM), a proposed explanation for the phenomenon.

Subscribe here

Dave went from a career as a senior software engineer to becoming a citizen scientist, eventually collaborating with established academic researchers, founding Own Your Labs — a platform that provides individuals with direct, affordable access to laboratory testing, establishing a fully qualified 501(c)(3) public charity called the Citizen Scientist Foundation, crowd-funding a groundbreaking research study, and contributing to a growing body of peer-reviewed work.

And Now…The Newsletter!

TFP_ Newsletter takes the podcast topics into written form — deeper dives, more nuance. Content ranges from foundational overviews for newcomers to close readings of complex lipid physiology for the specialists.

What to Expect

New weekly posts, straight to your inbox. You can also log in anytime to explore the archive.

Along the way, we’ll cover:

• Foundational overviews that clarify essential concepts so you can stay up to speed with the articles.

• Detailed explorations of complex metabolic and physiological mechanisms, tied back to foundational concepts.

• Critical appraisals of emerging and established literature, evaluated through a distinct metabolic framework shaped by Dave’s research.

• Unique perspectives spanning mental health, biochemistry, lipidology, nutrition, and heart disease assessment through imaging.

• Thoughtful discussions that encourage curiosity and open scientific dialogue.

Subscription options_

Free ($0/mo): article previews, select unlocked articles, podcast show notes previews.

Paid ($8/mo or $80/yr): full weekly newsletter, early access to video content, full podcast show notes, AMAs, community conversations, and more.

There’s also an optional founding tier for those who want to support the work at a higher level as we build this out. Founding members receive everything in the paid tier, plus recognition as early supporters and a founding-member thank-you gift.

Subscribe now

Looking Ahead

TFP_ Newsletter builds on the original blog and Dave's evolving body of work, connecting core principles to newer — sometimes unconventional — ideas.

The goal isn't to push a single viewpoint. Ultimately, it's to help people find their own path to better health.

Glad you’re here. Let’s dig in.

From The Desk of Own Your Labs: your opportunity to be heard

As a fan of TFP_, you can submit questions for upcoming podcast guests, “Ask Me Anything” episodes, general inquiries about the research, and more.

Submit your question

Start Survey

Share

Leave a comment

Who is Lily Johnston, and the book that came from a career spent hoping it would get better [0:56]

• Dr. Lily Johnston is a vascular surgeon at Scripps in La Jolla and the founder of a cardiometabolic prevention clinic within that practice. Her book, Disconnected, is forthcoming at the time of recording, with an anticipated formal launch in October. It is, in her words, “an exploration of my frustration with medicine and the practice of medicine and the culture of medicine in the 2020s.” She is the daughter of two physicians — an OB/GYN mother and a cardiac surgeon father — which makes her someone who should, theoretically, have known exactly what she was getting into. She did not. The book traces the gap between what medicine promises and what it delivers, both to patients and to clinicians, and is written as much for patients trying to understand why their doctor is a jerk as for colleagues trying to understand why they are unhappy.

Subscribe now

Growing up in a medical family: why the plan was always medicine, and what she could not see from inside it [7:37]

• Lily does not remember a time when medicine was not the plan. Her parents adored their careers and modeled what a life in medicine could look like at its best: her mother ran a thriving private OB/GYN practice, her father was a cardiac surgeon whose fond memories of his career remain intact even in retirement. She looked at these two people and thought: who would not want this? What she could not see as a young person was how much their experience of medicine — the autonomy, the time with patients, the longitudinal relationships — was already being eroded by the structural changes overtaking the field during her childhood. She would enter a completely different game than the one she had been watching from the sideline.

• Her sister, 13 years older and a social studies teacher in Albuquerque, was the self-described “smart one” and the family rebel who always worried that Lily was not rebellious enough. In the acknowledgments of the book, Lily tells her: “You can now say you told me so.”

The gap year that became three years: how a rejection at a science fair led to the intelligence community [11:10]

• Lily built a gap year into her medical school application timeline from the start — she would apply late, giving herself one year to grow up a bit before beginning the long tunnel of training. Her first plan was to be a ski instructor in Colorado. Her premed advisor immediately and explicitly told her not to: not a surf instructor, not a ski instructor. The indoctrination, she notes, starts before medical school. You are supposed to be making yourself bigger and better for the dedicated service ahead. Frivolous and non-academic is not the path.

• Attempting instead to find science consulting work, she went to a science and tech career fair and was turned away by every booth — a psychology major whose lab work was on neuroscience but whose resume did not read as a scientist. At the very last table, a government recruiter asked her to interview. She said yes, largely as practice for consulting interviews she expected to get. She did not get the consulting interviews. The government hired her as a science and technology analyst. She deferred medical school for a year, then two, then was told by her only remaining admitting school that there would be no third deferral: come now or reapply. She chose to come, having spent three years instead of one in a role that was, in retrospect, the best preparation she could have had for what followed.

Intelligence community training: what it actually means to declare a conclusion with calibrated uncertainty [20:32]

• The four-month training program Lily underwent as a new science and technology analyst was, in her description, the best brain training she has ever had. The context: she entered shortly after the WMD intelligence failure. The entire intelligence community had undergone its own root cause analysis of that catastrophic miss, and the training she received was designed to systematize thinking in response to it.

• The shift from academic training was 90 degrees. Academia builds up from first principles: here is the existing knowledge, here are the gaps, here is what we have done, here is our conclusion, after many pages or hours of argument. Nobody in a working environment has time for this. What intelligence work demanded was the opposite: lead with your conclusion, state your level of certainty, and tell the reader what you might have missed. Figure out what you know, figure out your gaps, have a process for testing your conclusions, and then declare — with appropriate and explicit uncertainty — what people need to know to make the next decision.

• Dave notes this connects directly to something he finds genuinely maddening in research medicine: the resistance to probability thinking, to stating conclusions as points on a spectrum of confidence rather than as binary truths. Lily agrees that medicine, unlike the intelligence world she came from, does not reward acknowledgment of uncertainty. It treats nuance as wishy-washy and curiosity about whether a given finding might be wrong as a personal challenge to authority rather than as normal scientific reasoning. She walked into medicine already trained in the other approach and spent years relearning to suppress it.

Back to medicine: still not cardiac surgery, eventually vascular surgery by accident [25:52]

• Lily returned to medical school aware that her utility as a science analyst had a shelf life — without a PhD, without a continuing academic credential, her ability to access scientific conferences would expire. Medicine was still the right next move. She had expected to become a cardiac surgeon like her father, planned her general surgery residency around cardiac fellowship access, and did a cardiac surgery sub-internship during medical school. What changed it was a rotation on vascular surgery as a resident, where several surgeons pulled her aside and told her: cardiac surgery is 1980s medicine. Look at what we can do.

• Vascular surgery offered variety she had not anticipated: leg one day, arm the next, neck the day after, open surgery alternating with minimally invasive catheter work, an on-call practice with the entire hospital as your referral network. Cardiac surgeons, by contrast, are largely beholden to interventional cardiologists, who increasingly want to do the catheter-based work themselves and pass on only the patients who are too sick and old for it. She chose vascular.

The indoctrination nobody acknowledges: what medicine does to people the way the military does to recruits [36:23]

• One of the central arguments of Disconnected is that medicine intentionally disconnects trainees from who they are as whole human beings, in a process that is functionally identical to military boot camp — with one critical difference: the military is transparent about what it is doing and why, while medicine never names it. You are told the training will be hard. You are not told that the systematic dismantling of your personal identity, community, and structure outside of medicine is deliberate, designed to mold you into someone whose primary loyalty is to the institution and the hierarchy.

• The result is a generation of clinicians whose sense of identity became fused with their credentials, their hierarchy, and the correctness of everything they learned during that long tunnel. When a patient — or a researcher, or a colleague — raises a question that might suggest part of what they learned was wrong, it is not experienced as a scientific challenge. It is experienced as a threat to the entirety of their career basis. The Feynmans of the world, Lily says, are not in medicine. The field does not reward the person who can be proudly wrong and grow from it; it rewards the person who protects what has been established.

• At the white coat ceremony — the ritual beginning of every medical school year — there is a line that appears in almost every speech across the country: “Half of what you will learn in the next four years is wrong; we just don’t know which half.” Everybody laughs. And then, Lily says, the second anyone actually tries to identify which half it might be, the response is: stop. That is settled. That is not the half.

The metabolic clinic: an hour for initial visits, half an hour for follow-ups, and what that costs [41:13]

• Lily now runs a cardiometabolic prevention clinic within Scripps in La Jolla alongside her vascular surgery practice. Initial visits are one hour. Follow-ups are 30 minutes. She is explicit that this represents a financial sacrifice relative to seeing patients every 15 minutes, and that she has accepted that cost because the difference in her own satisfaction is worth whatever she is leaving on the table. Her partners down the hall see patients every 15 minutes for surgical concerns. When anyone asks about diet, movement, sleep, stress, social connection, or supplementation, they are immediately referred to her. She sees people more frequently because the relationship needs to be high-touch to work.

• She describes what she had before moving to an insurance-based model: 90-minute to two-hour initial visits in private practice, 45 to 60 minute follow-ups. Even the current hour is a compromise, but it is a compromise she will live with to avoid the model she trained in. Dave raises the structural issue with the 15-minute visit: there is almost certainly information that only surfaces in the 16th, 17th, or 18th minute — information that could change everything for a particular patient — and the system as designed does not give the physician latitude to discover it.

The pharmacology lecture and the guitar: the moment medicine stopped feeling like a meritocracy [43:48]

• Lily’s first explicit signal that medicine was not going to be what she expected came in her first year of medical school. A pharmacology professor walked on stage, announced that supplements were garbage and a waste of everyone’s time, sat down, and started playing guitar. He told the class there were a couple of things in the syllabus they would cover briefly, but mostly they were just going to listen to music.

• Lily had actually read the chapter. She had done work on human performance supplements as a government analyst. She had questions. She had been genuinely excited for this lecture. She put her hand up. She was told — politely but unmistakably — to sit down and not worry about it. The contrast with her previous job was total: in the intelligence community, a junior analyst who had done the reading and had a question was an asset. In medicine, a first-year medical student who raised her hand during a senior professor’s improvised guitar session was a problem. She had arrived at the bottom of a very long hierarchy and needed to understand that, whatever she had done before, her thoughts were not yet welcome here.

I cannot just keep being the person who takes these people’s legs off [1:06:18]

• The turning point that redirected Lily toward cardiometabolic prevention was specific and came early in her career as a vascular surgeon. Her first practice did all amputations in the hospital, including diabetic foot amputations — cases where infection has destroyed so much foot tissue that amputation is the only alternative to death from sepsis. She had done an amputation on a patient who then appeared in the system again: he had healed the left-side amputation and was now admitted for an infection in his right foot. She pulled up the chart photographs. She knew immediately he was going to lose the second leg. He was in his 40s or 50s. She remembered his spouse talking about their cats, their dogs, the life they wanted to be living.

• “It was the moment for me,” she says. “I this has to change. Like, something has to be different. I have to have a practice that in some way helps with this problem because I cannot just keep being the person who takes these people’s legs off. Like, that is going to crush my soul forever and I will never recover if that is all that I offer these people in this world.” Her first introduction to carbohydrate reduction as a therapeutic tool came from the Low Carb MD podcast — Tro Kalayjian, Brian Lenzkes, and Brian Sanders — during COVID, when she was already looking at her own health and feeling like she could not stand in front of patients and advocate for lifestyle changes she was not herself living.

Vascular disease is systemic and it does not end: the whack-a-mole reality of surgical interventions [1:23:47]

• The deeper frustration Lily developed, beyond the amputation cases, was an understanding that vascular surgery in the heart disease context does not fix the problem. It manages one expression of a systemic disease. She stents the artery in the neck and the patient has a heart attack during the general anesthesia. The blood pressure drops managing the heart attack, and a vessel in the leg clots off. The toe goes black. New plaque builds in the stent within the year. She goes back to decide whether to balloon the stent again or replace it. It never ends. Her patients are patients for life. And the reason they are patients for life is not that surgery failed; it is that surgery was never going to fix the underlying disease. “These are patients for life, and while yes, I love the ability to have a long-term relationship with my patients, I don’t want it because they’re not getting better.”

Lipids, diabetes, and the question of what sucks all the oxygen out of the room [1:27:51]

• Lily is clear on her position: she generally agrees with the preponderance of evidence that elevated lipids — especially in patients with established plaque, established disease, or prior events — are worth monitoring and potentially reducing as one of several risk-management strategies. She has sat on an American Heart Association committee writing guidelines on medications for peripheral arterial disease. She is not a lipid skeptic.

• But she raises a concern Dave shares strongly: LDL tends to suck all the oxygen out of the room. The number of physicians whose cardiovascular risk assessment begins and ends with LDL — who do not look at fasting insulin, who do not look at triglyceride/HDL ratio, who do not register the severity of insulin resistance or metabolic syndrome as the far bigger proximate driver of the patients she actually sees getting amputated and stented — is too high. She has watched colleagues prescribe a statin to bring LDL down while simultaneously prescribing medications that worsen insulin resistance, with the diabetes managed by someone else’s silo. Her practice has become, in large part, a rejection of that silo. She wants to know what the whole metabolic picture looks like, not just what the lipid panel says.

• Dave’s framing: wherever you stand on the lipid hypothesis, diabetes is not even a close call. The debate about LDL has enormous range for reasonable disagreement. Whether severe, untreated insulin resistance and hyperinsulinemia are dangerous independent of any lipid value is not a debate at all. A fasting insulin of 40 scares him regardless of what the LDL is. This point gets lost because the lipid conversation is louder.

Carotid intima-media thickness: what it measures, why it is operator-dependent, and what cardioRisk does differently [1:37:51]

• Lily performs carotid IMT (intima-media thickness) testing in her practice and carries an ultrasound probe to conferences to do bedside screening. She explains the underlying anatomy: arteries are layered like an onion, with three layers. The inner two — the intima and media — begin to thicken when there is inflammation or subendothelial lipoprotein deposition. IMT measures this thickness, typically in the common carotid artery, as an early signal of arterial disease before frank plaque is visible.

• The test has a mixed reputation in cardiology, and she agrees with the criticism: ultrasound is operator-dependent, results vary enormously based on who is performing the scan and how, and without a standardized protocol, IMT measurements cannot reliably track change over time. She uses a lab called CardioRisk, run by Dr. Todd Eldridge, an engineer PhD who applied his manufacturing process background to the problem of minimizing variability in IMT testing. To scan for CardioRisk, she had to pass a blinded certification: she scanned five people twice, her images were anonymized and compared against their 20-year sonographer’s results, and a statistician confirmed her reproducibility. People who do not pass must practice more and retest. This level of standardization is what allows CardioRisk to report measurements precise to hundredths of a millimeter and to detect year-over-year change.

• She distinguishes two components of the full exam: the IMT measurement itself, done in the common carotid, and the detection of actual plaque, done in the carotid bulb where the artery bifurcates — the site of most turbulent flow and most early plaque formation. IMT and plaque are different findings and are tracked separately. She also images the femoral arteries, following Dr. Valentin Fuster’s work showing that femoral and carotid plaque detection together achieve approximately 90% concordance with coronary imaging for detecting systemic atherosclerosis.

Dave’s CIMT engineering problem and the neck mold that did not work [1:45:48]

• Dave discloses something he has not shared publicly before: he went through a period of trying to solve the operator-dependency problem of CIMT himself. He bought a Clarius ultrasound probe, attempted to make a fixed mold of his own neck and head to hold the probe in an exactly reproducible position for repeat scans, and ran into multiple problems including probe pressure, gel type, and gel quantity as additional uncontrolled variables. The mold did not work. He has since concluded that the variability problem he was trying to solve from first principles was real and that the CardioRisk protocol represents a better solution, though one that is not yet accessible in most locations in the way that a CAC scan is.

• His current recommendation for someone who is concerned about cardiovascular risk and wants the most information at lowest invasiveness: CAC and a high-quality CIMT performed under a standardized protocol if available. CT angiogram is the gold standard for soft plaque detection but requires contrast dye (with a small but real kidney risk), radiation, and access to a facility that can read it appropriately. His own keto CTA study has taught him something important about minimal detectable change: in a population where plaque volumes are low, the noise floor of the imaging tool itself becomes a limiting factor, and he now has a paper in preparation specifically on this. He flags that CT angiogram’s ability to track change over time may be further limited by algorithmic drift in the AI overlay tools, an issue raised by Dr. Nadir Ali that has not yet been addressed with published validation data.

The keto CTA study, minimal detectable change, and what Dave is and is not comfortable claiming [1:57:23]

• Dave and Lily discuss the LMHR CTA study in the context of measurement precision. His candid assessment: the most important finding — that people with sky-high LDL and an average of 4.7 years on a ketogenic diet did not have the expected elevated baseline plaque levels — is a major result regardless of the measurement precision questions. The expected finding per the prevailing lipid hypothesis was substantially more plaque at baseline than was found. That signal is large enough to be clinically meaningful independent of the noise floor question.

• Where the measurement precision issue bites is in tracking change over time. In a population where plaque volumes are already very low, distinguishing real change from measurement noise becomes much harder. The minimal detectable change threshold — the amount of difference between baseline and follow-up scans at which you can be 95% confident a real change has occurred — may be above the actual change levels present in many of their participants. This is not a catastrophic limitation; it is a known challenge of imaging low-disease populations longitudinally. It is also why he is pursuing five-year follow-up scans: a longer interval means a larger potential signal that is more likely to cross the noise floor.

• Lily finds this limitation clinically useful rather than discouraging: it has changed how she approaches CT angiography in her own patient practice, making her more thoughtful about what conclusions she can and cannot draw from sequential imaging and more focused on using CIMT for longitudinal tracking in patients with lower initial disease burden.

Guidelines, liability, standard of care, and the legal distinction most physicians do not know [2:06:11]

• Lily does medical malpractice work on the side and makes a legal distinction most clinicians are not aware of: guidelines are not the same as standard of care. Standard of care is what a similar physician with similar training would do under similar circumstances — a community standard, not a document standard. Guidelines can be used both for and against plaintiffs or defendants in a malpractice context. A brand-new guideline (she uses the recent recommendation that all patients should be tested for Lp(a) as an example) cannot be applied retrospectively to cases that predate its publication. Practicing outside guidelines is not automatically malpractice; it is, however, a risk that requires a clear documented rationale.

• The mechanism she recommends for physicians who want to practice outside guidelines without undue liability exposure: document a shared decision-making conversation with the patient. The patient has been told what the guidelines say, understands the risks and benefits, and has chosen a different course. If that is genuinely in the chart — in a way that can be verified — the physician has substantial protection. She acknowledges this note often reads as punitive to patients who can now access their own charts, but it is protective in the event of a legal challenge. And she acknowledges the uncomfortable truth that the low-carb and cardiometabolic prevention community has seen physicians have their licenses challenged and their reputations publicly attacked for practicing outside standard of care in good faith and with patient benefit — and that guidelines, whether or not they are the legal standard, function as institutional cover that a doctor practicing outside them simply does not have.

Nick Norwitz’s case report, the medication critics, and what it costs people to share personal health decisions publicly [2:19:29]

• At the time of recording, Dave and the TFP_ team have just released a case report documenting Nick Norwitz’s total cholesterol of approximately 700 mg/dL sustained over seven years on a ketogenic diet, alongside recently disclosed information that Nick has chosen to start bempedoic acid and ezetimibe based on his own research into potential brain health benefits given his ApoE4/4 genotype. Critics in the space immediately used the medication disclosure to suggest the case report was undermined or that Nick had implicitly conceded something about high LDL.

• Dave describes being more bothered by this than usual, and Lily articulates why: Nick did not have to share this. He is a researcher and a public figure making a personal health decision based on his specific genetic context, and he chose to be transparent about it. The use of that transparency as rhetorical ammunition against the research — rather than as exactly what it is, which is an individual making an informed choice about his own body — is precisely the pattern that chills physicians and researchers from sharing anything that could be used against the position they are publicly associated with. It teaches people that honesty about the complexity and individuality of health decisions is a liability. That is the opposite of the culture either of them wants to build in this space.

AI in medicine: what it will do well, what it cannot do, and whether it might eventually be dangerous not to use it [2:29:28]

• Lily is optimistic about AI in medicine, primarily because AI does the boldface-type medicine — the algorithmic matching of symptoms to standard-of-care diagnoses and prescriptions — better than humans do. She uses Open Evidence, a medical AI platform trained on the peer-reviewed literature and run in partnership with NEJM and JAMA, as a tool she trusts more than general chatbot interfaces for clinical questions because its training set is bounded and auditable. Her vision for AI’s highest-value contribution: offload the cognitive work of differential diagnosis, documentation, and guideline checking onto a system that does not get tired, does not have a diagnosis it likes better for personal reasons, and updates in near-real time when new evidence drops. Free the human clinician to do the thing AI cannot do: be actually present with another human being who is frightened and suffering.

• Dave raises the framing he finds most clarifying: we already accept that full self-driving cars are measurably safer than human drivers in controlled conditions, and that some jurisdictions might eventually make human-only driving illegal in certain contexts because the AI is simply safer. The medical analog may not be as far off as it seems. The scenario where it becomes genuinely dangerous for a human physician to be the sole diagnostician — because an AI with access to a patient’s full wearable data, labs, imaging, and real-time physiological monitoring is so much more accurate — is not science fiction. The catch, as Lily notes, is the training set: who decides what the AI is trained on, and whether the top-down directives embedded in that training will allow it to diverge from existing guidelines in cases where a metabolically healthy lean mass hyperresponder with elevated LDL and no plaque might not need pharmacological management.

Exercise-induced laryngomalacia: a decade of being told she was out of shape, a PubMed search, and what AI could have found in 10 minutes [2:39:56]

• Lily shares a personal story that illustrates the AI diagnostic point concretely. From childhood through college and into her medical training, she experienced severe exercise-induced breathing difficulty — not asthma, not vocal cord dysfunction (which she was eventually told), but something that reliably appeared whenever her heart rate exceeded 160 bpm and caused breathing so stridorous during a martial arts tournament that her opponent’s coach stopped the match out of fear for her safety. She underwent stress echos, allergy testing, asthma testing, and multiple physician visits across more than a decade. The consistent answer was: you are out of shape, work harder.

• She found the correct diagnosis herself, scrolling PubMed in a laboratory one day: exercise-induced laryngomalacia, described in only a handful of published case reports at the time. She emailed the lead authors, connected with an ENT surgeon in New York, and got the answer more than ten years after her symptoms began. Her conclusion: had she been able to type her symptoms into any current large language model, she would almost certainly have had the correct diagnosis within ten minutes. This is not a small deal. It is a decade of being told by multiple physicians with access to the same information that she was simply unfit. AI does not get tired of asking questions and does not have a prior diagnosis it prefers.

Errors of commission versus errors of omission: why doing nothing feels different from doing something, even when nothing is correct [3:05:58]

• Lily raises a surgical psychology reality that applies broadly to how medicine handles uncertainty: errors of commission — doing something that causes harm — feel catastrophically worse than errors of omission — not doing something that might have helped. As a surgeon, her hands are literally in the patient. When she does a procedure and it goes poorly, the weight of responsibility is immediate and personal. When she recommends against a procedure and the patient deteriorates from the natural progression of disease, she can tell herself it was not her fault; that was just the disease. Both outcomes can be equally bad for the patient. The psychological experience of them is not the same for the doctor.

• This asymmetry drives over-intervention. Physicians are uncomfortable with observation and watchful waiting even when observation is the most evidence-supported strategy, because the failure mode of doing nothing feels morally heavier than the failure mode of doing something that does not work. This matters directly for LMHR patients: an asymptomatic patient with elevated LDL, no plaque, and excellent metabolic markers is presenting a scenario where watchful waiting with repeat imaging may well be the most appropriate path — but it is a path that requires a physician willing to live in that uncertainty without the psychological cover of a guideline-compliant intervention.

Mortality, health span, and the best patients she has ever had [3:11:24]

• The episode closes with a thread Lily weaves throughout her book and throughout this conversation: none of us are getting out of this alive, and the best death she has witnessed clinically is not the one where everything possible was done. It is the one where the patient says, “It’s been a great run. Thank you so much for everything. I’d like to go home now.” She has watched colleagues do amputations and procedures on patients approaching 90 who then never return to anything resembling the life they had before. She has watched families push for intervention because they cannot accept that their mother, who was mobile and playing with grandchildren last week, might not go back to that life on the other side of the surgery.

• Her framing is not anti-intervention; it is pro-honesty. She talks about mortality with her patients. She thinks medicine does not talk about it enough, and that clinicians who have spent decades surrounded by death and suffering and processed it largely in isolation have lost access to the human conversation about it that patients actually need. Part of what Disconnected is trying to do is return that conversation to medicine — for clinicians, for patients, and for anyone navigating the healthcare system on behalf of someone they love. She lost a cousin to metastatic breast cancer in her late 20s while Lily was in surgical training. She thinks about what her cousin would want for her nearly daily. It is part of why she changed her career, and part of why she is trying to live a life she would be at peace with if it ended tomorrow.

• The book is Disconnected. She can be found at Lily Johnston MD on YouTube and as EL Johnston MD on LinkedIn.

Share

Leave a comment

Low-density lipoprotein cholesterol (LDL-C) has long been used as a central marker in cardiovascular risk assessment. But one of the persistent questions about LDL-C is how well it separates who will actually develop heart disease from who will not.

Many people with elevated LDL-C may never experience a cardiovascular event, while others with “normal” levels still do.12

This, of course, is not an issue unique to LDL-C. Many blood-based biomarkers struggle to reliably identify who truly has disease developing in their arteries.

This raises an important question:

What if, instead of estimating risk indirectly from a blood test, we looked directly for evidence of disease itself? Wouldn’t that be a better way to identify someone at risk?

That is where imaging enters the picture.

Moving beyond the blood…enter imaging

A coronary artery calcium (CAC) scan is a specialized CT scan that detects calcified plaque within the coronary arteries. Rather than measuring a risk factor, CAC scoring measures the actual presence of heart disease.

So which is more useful for predicting future cardiovascular events? For example, if someone’s LDL-C is high with no disease based on imaging, how does their risk compare to low LDL-C but evidence of disease on imaging?

An article3 from the Multi-Ethnic Study of Atherosclerosis (MESA) provided some evidence to help answer this question…

Researchers broke up people with different levels of LDL-C into four groups called quartiles, going from lowest (Q1) to highest (Q4):

• Q1: less than 70 mg/dl

• Q2: 70-100 mg/dl

• Q3: 100-130 mg/dl

• Q4: greater than 130 mg/dl

They also broke up CAC scoring into three groups:

• CAC score of 0 (no calcified plaque)

• CAC score of 1-99 (lower-moderate calcified plaque)

• CAC score of greater than or equal to 100 (higher calcified plaque)

At first glance, you’d expect those with higher and higher LDL-C to possibly have more heart attacks and strokes, irrespective of their CAC score, right?

Share

The MESA study tells us a different story…

When CAC scoring enters the picture, LDL-C levels became much less helpful in predicting who is at risk for heart disease vs who is not.

Take a moment to look at the following figure:

The numbers on the bars are essentially the percentage of people having a heart disease-related event over a 10-year period. So, in other words, 4.7% or ~5 people out of 100 with LDL-C <70 mg/dl with a CAC score of 0 followed for ten years had a heart attack or stroke or some other event related to heart disease (black bar on the far left).

As you see in the figure, as LDL-C increases in the context of a CAC score of 0 (black bars from left to right), the rate of heart events goes from 4.7% for LDL-C <70 mg/dl to 3% for LDL-C > 130mg/dl. This is generally considered to be a very low rate of heart disease.

The Power of Zero

This is commonly known as, “The Power of Zero.” In other words, those with a calcium score of zero usually have a very low incidence of having a heart attack or stroke, at least over a 5 to 10-year period, irrespective of many established risk factors such as LDL-C.

As the MESA study authors noted, “when CAC=0, absolute event rates remain relatively low across varying levels of dyslipidemia.”

But what happened when the calcium score starts to creep up even with different levels of LDL-C?

CAC tends to take over predicting risk.

For example, notice those with low LDL-C <70 mg/dl and a CAC ≥100 had an incidence rate of ~26%. That’s nearly nine times the incidence of those with LDL-C ≥ 130 and CAC=0 (see figure below).

And this finding held up across multiple measures of cholesterol levels, ratios, etc.

The calcium score was a better predictor of whether an individual experienced a heart disease-related event.

So what’s the takeaway?

LDL-C shows a very modest association at lower CAC score, but a more notable association at a higher CAC score.

But as far as predicting the incidence of having a future heart attack or stroke…CAC appears to be the clear winner in this study.

But what about other cholesterol/lipid disorders like high triglycerides (TG), low high-density lipoprotein cholesterol (HDL-C) with high LDL-C (similar to atherogenic dyslipidemia)? What about apolipoprotein B levels? How do these compare to CAC scoring? Do they do any better?

That's where the paid subscription picks up — if you want to follow the thread all the way through, consider subscribing!

Subscribe now

From the desk of Own Your Labs: your opportunity to be heard

As a subscriber to TFP_ Newsletter, you can submit questions for upcoming podcast guests, “Ask Me Anything” episodes, general inquiries about the research, and more.

Submit your question

Read more

Who is Austin Dudzinski, and what a clinical pharmacist actually does [1:08]

• Austin Dudzinski is a board-certified clinical pharmacist working in primary care alongside four physicians (two internal medicine, two rheumatology). His role centers on deprescribing — working to de-escalate medication burden — with nutrition therapy as a primary tool, the ketogenic diet being one of the most powerful interventions in his toolkit.

• His actual function is closer to an advanced prescribing practitioner under a collaborative practice agreement: he has autonomy to start, change, or de-escalate certain therapies and order labs. He functions as a liaison between physician visits — a patient with an A1C of 12% is not waiting three months for follow-up; they typically reach him within 24 hours for a full conversation about pathophysiology and lifestyle mitigation before any medication decision is made.

• He distinguishes his role from retail/community pharmacy (drug interaction review, dosing checks, point-of-sale education) and other subfields — compounding, specialty, nuclear, hospital pharmacy — noting his niche specifically involves polypharmacy management: understanding how a large medication list interacts and finding ways to consolidate and reduce it.

• His night job: part owner of Digital Conversation, a company building digital healthcare texting infrastructure to streamline communication between outpatient facilities and patients between visits — explicitly framed as solving the same problem Tro Kalayjian’s app addresses, via texting rather than an app, with the goal of breaking down the wall created by the standard 15-minute office visit.

Subscribe now

The insulin problem that started Austin’s nutrition journey [8:22]

• Austin’s entry point into nutrition research was professional frustration, specifically around type 2 diabetics being escalated to exogenous insulin (long-acting basal insulins like Lantus, eventually mealtime insulin) while their own pancreas is still secreting insulin. Adding exogenous insulin to an already hyperinsulinemic state, he explains, is borne out in observational data as likely harmful rather than helpful.

• He distinguishes the two major categories of diabetic damage: microvascular (eye, nerve, kidney damage), which correlates more tightly with blood glucose values, versus macrovascular (heart attacks, strokes), which is what actually kills most people with type 2 diabetes. The argument for exogenous insulin rests on lowering glycemia to prevent microvascular damage — but when you look at the data on macrovascular outcomes, insulin therapy in this context does not help and often makes patients sicker: weight gain, and a predisposition toward dangerous hypoglycemia that can cause seizure, coma, or death.

• His professional conclusion, stated plainly: at the point where insulin is the tool being reached for in this specific context, you are causing harm. That conviction is what sent him searching for an alternative — reading broadly in nutrition science to find a way to prevent patients from reaching that point at all.

The plant-predominant detour: Willett, the Sonnenburgs, and Longo before low-carb [12:20]

• Before any low-carb exposure, Austin found Stefani Reinagel’s Red Pen Reviews — a site applying rigorous fact-checking criteria to popular nutrition books — and used it to select his starting material: Walter Willett’s Eat, Drink, and Be Healthy; The Good Gut by the Sonnenburgs; and Walter Longo’s The Longevity Diet. The consistent takeaway across all three: eat more plants, less saturated fat, fewer animal products, more fiber, with protein treated as a secondary consideration.

• He brought this framework directly to patients and found their eyes glazing over — they had already heard “more whole grains, less steak, fewer eggs, less bacon” repeatedly, found it largely unactionable, and even among those who did adhere, he did not see the outcome improvements he expected. This is the tension he flags as worth returning to throughout the conversation: the balance between a diet’s theoretical healthfulness and its real-world adherence.

Becoming his own patient: 154 triglycerides, My Fitness Pal, and finding Peter Attia’s early episodes [14:47]

• Around the same period, with his wife pregnant with their first child, Austin took stock of his own labs: triglycerides at 154, low HDL-C, abdominal obesity, mild hypertension. He tried the conventional calorie-counting playbook — My Fitness Pal, a 500-calorie deficit, joining a gym five days a week — and lost 35 pounds while being, in his words, miserable: pervasive hunger and the social awkwardness of meticulously tracking food around his newborn and family.

• At the gym, he discovered Peter Attia’s podcast The Drive in its early run — “old Peter Attia,” the keto-era Attia who told stories about mainlining saturated fat off cuts of meat at Brazilian steakhouses, featuring guests like Robert Lustig, Gerald Shulman, and Richard Johnson on low-carb, fructose, and insulin resistance. He tried the approach himself, lost another 10 pounds with ease, and was never hungry. Around the same time, several physicians in his own practice were independently exploring keto with patients, creating the opening to start deploying it clinically — placing this timeline around 2018, just as “keto” was becoming a mainstream buzzword.

• The clinical results, in his account, were repeatedly described as miraculous for patients who could adhere: taking people off 200 units of exogenous insulin in under two weeks, a result he calls unheard of in conventional medicine. He flags the under-discussed cost dimension of this directly — exogenous insulin is expensive, requires daily (sometimes multiple-daily) injection, and is one of the two medication classes (alongside blood thinners) most associated with hospitalization.

The salt journey: fructose malabsorption, the soda-drinking trio, and eight tablets a day [27:21]

• Austin walks through what he calls the hardest single thing he has had to figure out in his entire low-carb journey: sodium. The difficulty, he argues, is precisely because it is so easy to rule out — “I’m already salting my food, that can’t possibly be it” — and because symptoms (foot cramps, lightheadedness, fatigue) are more intuitively attributed to potassium or magnesium.

• His path to the answer ran through George Henderson, who connected Austin’s pre-existing, independently diagnosed fructose malabsorption to impaired sodium absorption — the same intestinal transport mechanism implicated in fructose malabsorption may also reduce sodium absorption. Austin’s account of discovering the fructose malabsorption itself is a story in three parts: his friend Leroy visiting Las Vegas and immediately identifying Austin’s symptoms as matching his own high-fructose-corn-syrup intolerance; Austin’s own two-years-earlier failed self-experiment of quitting soda but replacing it with apple juice (symptoms persisted); and a controlled apple-eating test with his friend Andrew that produced a severe symptomatic flare that evening — confirming the diagnosis. All three members of the same childhood soda-drinking, video-game-binging friend group developed fructose malabsorption and IBS independently, which Austin treats as more than coincidence.

• The diagnostic trap fructose malabsorption sets: symptoms typically do not appear immediately. The delay is inconsistent but commonly runs 12–36 hours after the triggering food, meaning sufferers reflexively (and incorrectly) blame their most recent meal rather than something eaten the day before.

• Once Austin committed to a genuine two-week salt-with-abandon experiment — deliberately consuming 10–12 grams of salt daily, including drinking salt water, with a blood pressure cuff on hand to monitor for salt sensitivity — his symptoms disappeared completely. He has since settled into a routine of roughly two to four salt tablets per meal (eating twice daily, so up to eight tablets total), scaling back to three once stabilized, on top of normal food salting. He flags that adequate sodium is also, in his clinical experience, by far the most effective intervention he has found for the severe constipation commonly seen in patients starting GLP-1 agonists — more effective than fiber, Miralax, or Senna, with what he estimates as a 95% hit rate using roughly two cups of broth daily.

The Drive episode, Peter Attia’s “pre-buttal,” and what changed and didn’t [52:01]

• Austin describes his own journey discovering Dave through Attia’s podcast — a roughly three-and-a-half-hour episode he has, by his own account, relistened to more times than almost anyone. His honest first impression: confusion about whether Dave’s evident kindness on the podcast was genuine or some kind of performance, rooted in what he now calls an embarrassing but honest old mindset that “the doctor is the smartest person in the room” — making an engineer who seemed to understand lipids better than Attia himself genuinely disorienting to process.

• What changed his read on Dave specifically was a separate appearance on Ivor Cummins’s podcast, recorded shortly after the Attia episode, where Attia had delivered what neither Dave nor Austin had ever seen a podcast host do before or since: a “pre-buttal” summarizing in advance the three core objections he held to Dave’s position (mass balance and genetic-alteration arguments) before the conversation had even concluded — effectively editorializing the episode’s framing for listeners before they had heard Dave’s case made in full. Dave is candid that this did create real friction within the low-carb community and likely discouraged some listeners from engaging with the argument on its own terms — while still expressing genuine gratitude for the platform and his belief that Attia’s concern, at bottom, comes from a sincere worry about potential harm rather than malice.

• Austin’s assessment, having relistened repeatedly: Dave’s substantive positions have remained remarkably consistent from that early episode to today, with the most meaningful evolution being a stronger, better-developed answer to the mass balance critique — specifically Dave’s now-clearer articulation of his structural/endocytosis hypothesis (discussed in depth below), which did not yet exist in its current form at the time of the original Drive recording.

The structural demand hypothesis: why lean mass hyperresponders might transcytose more [1:01:07]

• Dave lays out, with the explicit framing that this is speculative, his structural demand hypothesis for why lean individuals in particular see the largest LDL rises on a ketogenic diet: leaner people have smaller adipocytes, and the cyclical expansion and contraction of those smaller fat cells — even in small absolute increments — may require more net endocytosis of ApoB-containing lipoproteins, since their cholesterol and phospholipid cargo is structural material the membrane needs to accommodate that expansion.

• He flags the testable prediction this generates: if true, biopsy or mRNA-expression data should show measurably elevated transcytosis-related receptor activity (e.g. SR-B1) in lean mass hyper responders specifically, reflecting genuine cellular demand for the cargo rather than passive accumulation. He notes this directly contradicts the textbook teaching that “every cell can synthesize all the cholesterol it needs” — acknowledging the position is, in his own words, “blasphemy” relative to current orthodoxy, while pointing to a response-to-retention paper by one of that hypothesis’s own originators conceding cells likely do uptake LDL components for various structural purposes under certain circumstances (his “Home Depot” analogy: delivery trucks circulating a neighborhood get flagged down for small material needs constantly, even outside formal deliveries).

• The mechanistic stakes Dave names explicitly: if structural-demand-driven transcytosis is real and operating at a higher net rate in LMHRs, that would put his hypothesis in direct opposition to the concentration-gradient version of the present-day lipid hypothesis — which treats higher transcytosis rates as proportionally increasing the odds of an ApoB particle becoming irreversibly “stuck” and triggering the macrophage/foam-cell cascade (what Dave calls, semi-jokingly, “the lottery you don’t want to win”). If LMHRs are genuinely transcytosing more and yet show no elevated plaque development, that would be a striking piece of evidence against the simple stochastic version of that model.

Bo Fan’s “flypaper, not flies” thread and why diffusion-mediated entry doesn’t hold up physically [1:06:42]

• Both credit “Lipid Pan” (Bo Fan), an engineer by training, as an underappreciated independent analyst in this space — Austin notes the irony of another engineer “knowing more about lipids than 99.9% of lipidologists.” Dave highlights a specific thread of his: a recent publication suggesting the response-to-retention hypothesis needs updating, because the rate of transcytosis itself does not appear to predict how much ApoB-containing material actually gets retained in the intima — the proteoglycan binding sites themselves (”the flypaper,” not “the flies”) appear to be the differentiating factor, for reasons not yet mechanistically understood.

• Austin independently raises the pure physics objection to a passively diffusion-mediated entry model: an LDL particle would need to navigate the glycocalyx, locate an endothelial cell, and squeeze through tight junction gaps measured at 3–9 nanometers — while the particle itself is three to four times that diameter. Simple steric hindrance, he argues, makes that mechanically implausible as a passive process; it has to be active, which both take as further evidence against any model where raw ApoB concentration alone is treated as a direct, linear driver of plaque formation.

Dave’s explicit defence of statins, and the conversation he refuses to have on someone’s behalf [1:09:24]

• Both push back directly on a recurring social-media strawman: that Dave’s advocacy is responsible for patients refusing statins against their doctor’s advice. Austin offers a direct rebuttal from his own clinical experience: the two medication classes he fields the most fear-driven phone calls about are bisphosphonates and statins, and patients refusing a clinically appropriate statin because “Dave said so” are, in his clinical assessment, going to refuse it regardless of the specific source — and if a clinician cannot walk a patient through the actual evidence well enough to overcome that fear, “you’ve failed as a clinician,” in Austin’s words, not Dave’s.

• Dave’s explicit position, stated for the record: he holds a default assumption that every medical intervention carries trade-offs, with no exceptions, and he is deliberately uninterested in being the deciding voice on any individual’s treatment decision — that conversation belongs between a patient and their own doctor, full stop. He is careful to distinguish this from claiming metabolic health is more important than lipid management for everyone; rather, his consistent clinical priority ordering is metabolic health first, with lipid intervention remaining on the table as a secondary consideration for those who, after addressing metabolic health, still want to “hedge their bets.”

The documentary’s hardest case: when lowering LDL brought the mental illness back [1:22:31]

• Austin, who has seen an early cut of the Cholesterol Code documentary, raises (with care to avoid third-act spoilers) one of the film’s most difficult stories: a participant with debilitating bipolar disorder who, after seeing LDL rise on keto, worked with their doctor to reintroduce carbohydrate specifically to bring LDL down — and the mental illness returned alongside the falling LDL. Austin describes being moved to tears watching it, framing the story’s real subject as not just the patient but her family’s relationship with her, and the visible contrast between her affect during remission and during relapse.

• Austin’s clinical hypothetical for a patient like this: ezetimibe, or a PCSK9 inhibitor if affordable, purely as a hedge alongside continued imaging — while citing Nick Norwitz’s own n=1 high-intensity rosuvastatin (20mg) experiment as a relevant data point: Norwitz saw a comparatively weak LDL response and experienced side effects, illustrating that the mechanism by which statins lower LDL (inhibiting hepatic HMG-CoA reductase, an enzyme plausibly important for individuals trafficking large amounts of fat) may make them a poor tool specifically for the highest-trafficking LMHR phenotype — quite apart from whether the resulting LDL reduction would even be clinically meaningful over a lifetime for someone in that specific case.

“Cholesterol focus is a disastrous distraction”: the LPIR comparison [1:27:50]

• Dave states his clearest formulation of his core critique: cholesterol focus, even granting every assumption of the lipid hypothesis, functions as a disastrous distraction from addressing insulin resistance syndrome — a condition associated with catastrophically larger effect sizes that gets comparatively little attention relative to the “spotlight” placed on lipid management.

• Austin supplies the specific numbers from the Women’s Health Initiative analysis using Bill Cromwell’s LPIR (Lipoprotein Insulin Resistance) score: the adjusted hazard ratio for a woman under 55 is 6.51 per standard deviation of LPIR, versus 1.81 per standard deviation of ApoB — a roughly three-and-a-half-fold difference in magnitude. He flags this alongside the broader multicollinearity problem with treating ApoB as analogous to smoking or hypertension as an independent risk factor: nearly every other metabolic syndrome marker worsens in lockstep with rising ApoB on a population basis — except in lean mass hyper responders, where the more they exercise and the leaner they get, the higher their ApoB climbs while triglycerides fall and HDL rises, the exact opposite clustering pattern. This divergence is, in Austin’s assessment, exactly why he considers the LMHR phenotype far more clinically generalizable than critics tend to credit it for.

Remnant cholesterol: why a tiny number change produces an outsized risk signal [1:45:06]

• Austin introduces remnant cholesterol — the cholesterol carried on triglyceride-rich, non-LDL, non-HDL ApoB particles, typically VLDL-derived precursors to LDL — as outperforming ApoB and LDL-C directly when pitted against plaque progression metrics in available data. He cites Boørge Nordestgaard’s “excess ApoB” research: even a very small absolute increase in the remnant fraction is associated with a disproportionately large jump in cardiovascular risk.

• Dave’s explanation for why this small number carries such outsized signal, without conceding the remnants are intrinsically more atherogenic per particle: a higher remnant count is itself a marker of the same underlying “traffic jam” — elevated remnants cluster with elevated triglycerides and a higher proportion of small dense LDL through CETP-mediated remodeling, meaning the remnant elevation and the cardiovascular risk it predicts likely share a common upstream cause (impaired lipid trafficking) rather than the remnant particle itself being the causal agent. His critique of ApoB as a metric follows directly from this: because ApoB sums LDL particles and remnant particles together, any condition that elevates remnants (hypertriglyceridemia, metabolic dysfunction) will marginally inflate ApoB and get statistically “blamed” on the LDL fraction it is bundled with — which is precisely why isolating a population with extremely high LDL/ApoB and simultaneously very low remnants (only achievable, as far as either is aware, in lean mass hyper responders) is so valuable: for the first time, ApoB and LDL-C can be examined in genuine isolation from the remnant confound.

• Austin adds the corroborating mediation-analysis literature: one study found roughly 75% of cardiovascular death attributable to HOMA-IR (a glucose/insulin-based insulin resistance proxy) once both HOMA-IR and remnant cholesterol were included in the same model — suggesting the remnant elevation is substantially a downstream marker of insulin resistance rather than an independent driver. A second study found that adjusting ApoB’s association with cardiovascular risk for VLDL particle count rendered the ApoB association statistically non-significant — again pointing toward the “milieu” (the broader inflammatory, insulin-resistant metabolic state) as the operative variable rather than the minority remnant particle subtype itself.

• The Nordestgaard “excess ApoB” finding Austin flags as most frequently misunderstood on social media: excess cholesterol carried per ApoB particle (i.e. cholesterol-rich, “fluffier” particles relative to particle count) was associated with reduced myocardial infarction and reduced all-cause mortality in women — the opposite direction many assume. Framed differently: when non-HDL cholesterol divided by ApoB falls below a 1.4 ratio threshold, risk rises sharply; above 1.4, risk falls. Dave’s shipping-manifest analogy for why this makes physiological sense: an LDL particle’s lipid composition at any given moment is a “manifest” of its prior metabolic journey — a particle that has accumulated triglyceride cargo rather than cholesterol ester cargo is telling you something about upstream CETP exchange activity and lipid trafficking efficiency, which is precisely the lipid energy model’s core claim.

Mendelian randomization’s hidden assumption: environmental equivalency [1:55:29]

• Austin raises Mendelian randomization’s core methodological vulnerability directly: the technique assumes a genetically-driven elevation in a biomarker is causally equivalent to the same biomarker level reached through any other pathway over a lifetime — formally known as horizontal pleiotropy or the exclusion restriction assumption, what Adrian Soto-Mota’s COI talk termed “environmental equivalency.” Once a metric is in the bloodstream, the technique treats its etiology as irrelevant to its downstream risk — which is precisely the assumption that produces the clean, “pretty” log-linear plots seen in papers like the 2017 EAS consensus statement.

• Dave’s direct empirical challenge to that assumption: working with Siobhan, he identified roughly 26 SNPs that change LDL and/or total cholesterol specifically without any associated change in triglycerides or HDL, and which are not otherwise implicated in lipid metabolism pathways. Using the MR-Base tool to run a forest plot against these isolated SNPs, the result showed no association with worse all-cause mortality — if anything, a slightly favorable direction. Austin’s reaction: this would be a direct empirical violation of one of Mendelian randomization’s foundational assumptions, and he explicitly encourages Dave to formally publish the analysis.

• A second concrete violation Austin raises independently: genetic variants affecting ApoB through different mechanistic pathways (lipoprotein lipase variants vs. LDL receptor variants, as in classic familial hypercholesterolemia) carry meaningfully different cardiovascular risk per identical ApoB particle count — which should not be possible if the technique’s core assumption (that the route to an elevated biomarker is irrelevant to its downstream risk) actually holds.

The three-tier FH chart: same LDL, wildly different risk depending on genetic etiology [2:00:13]

• Dave describes the slide he uses regularly to make this point visually concrete: three groups, all sharing the same elevated LDL level, but with dramatically different cardiovascular risk depending on genetic etiology. Monogenic FH (the classic Brown-and-Goldstein-style single-gene defect, e.g. LDL receptor or ApoB mutations) carries the highest risk by a wide margin; polygenic FH (elevated LDL driven by an accumulation of many smaller-effect genetic variants, captured by polygenic risk scores) carries meaningfully lower risk at the identical LDL level; and elevated LDL with no identified genetic component at all carries the lowest risk of the three — a clean, stepwise gradient that strongly implies genetic etiology itself, not just the resulting blood level, is doing independent causal work.

• Dave’s pointed logical challenge built on this: a prominent lipidologist (left unnamed) holds both that “it matters how you lower ApoB” (citing LDL receptor upregulation specifically as the validated, efficacious mechanism) and that ApoB level alone is “the whole game” — positions Dave argues cannot be coherently held simultaneously. If mechanism of arrival genuinely matters, that is a meaningfully different (and more interesting, in Dave’s view) hypothesis than the simple concentration-snapshot version of the lipid hypothesis, and one that would logically also need to account for how a lean mass hyperresponder’s elevated LDL arrived — not just its absolute level.

Inside the LMHR paper: what Austin saw first, and the “plaque begets plaque” finding [2:06:51]

• Austin, given early access to the paper ahead of its public release, opens with his reaction to its title — “Plaque Begets Plaque, But ApoB Does Not” — and confirms the result matched his own prior expectation closely. The single strongest predictor of future plaque presence or progression in the dataset was baseline plaque itself (an R² of roughly 0.33 for baseline calcified plaque predicting future presentation) — not ApoB, not LDL-C, not any lipid metric tested.

• ApoB and all tested LDL metrics (LDL-C, LDL-P) showed no association with non-calcified plaque volume, percent atheroma volume, or total plaque progression across the cohort.

• Dave’s reflection on the achievement itself: zero participant dropout across both scan visits in a 100-person cohort — a retention rate the Lundquist Institute statisticians told him they had never previously observed in a study of this size, particularly notable given participants were uncompensated beyond travel and flown cross-country twice. He credits Dr. Matthew Budoff, Nick Norwitz, Adrian Soto-Mota, Tommy Wood, and the broader low-carb donor community by name.

• A key contextual finding Dave flags as novel in its own right, independent of the lipid question entirely: this is, to his knowledge, the first longitudinal dataset in a healthy (non-diseased) population confirming that baseline plaque predicts future plaque progression — that relationship had previously only been established in already-diseased populations.

The fifth-year extension, and what would and wouldn’t change Dave’s mind [2:09:50]

• Dave confirms fundraising is nearly complete for a third round of scans, targeted for October of the following year — exactly five years from the study’s original October 2021 baseline. He flags openly that he does not expect full 100-participant retention for this round given the longer interval, and is candid that this round could, in principle, surface an LDL/ApoB-plaque association that the current one-year data does not show — either across the whole cohort or specifically within the fast-progressor subgroup.

• His standing position on plaque trajectory generally, stated as a hypothesis rather than a finding: he does not believe plaque progression is necessarily linear, and considers it plausible that some degree of non-calcified plaque increase represents an ongoing reparative process (analogous to a healing wound) rather than a purely one-directional decline — a hypothesis the planned third scan round is specifically positioned to help test, given enough additional follow-up time to observe whether early non-calcified plaque is being progressively converted to stable calcified plaque or resolving outright.

The third Miami Heart re-analysis: stripping out the statin-treated third [2:13:30]

• Austin raises the most substantive critique the match analysis received publicly (from Bill Cromwell and Simon Hill specifically): roughly one-third of the original Miami Heart comparison cohort was on lipid-lowering therapy, which Dave acknowledges is a legitimate methodological concern — his team simply could not source a sufficiently large statin-free comparison population through Miami Heart alone.

• In response, Dave requested (and Lundquist complied with) a follow-up re-analysis: re-matching the LMHR cohort against only the statin-free two-thirds of the original Miami Heart pool. He presented the result at Low Carb San Diego — still no statistically significant difference in plaque burden between groups. He walks through why the steelman version of the original objection does not hold up arithmetically: for the excluded statin-treated third to be meaningfully skewing the original comparison toward a falsely favorable result, that subgroup’s pre-treatment LDL would need to have averaged roughly 190 mg/dL — a figure Dave considers implausible for what should be a roughly representative general population sample.

• Austin notes he has not yet seen a public response from Simon Hill specifically acknowledging the follow-up analysis, despite having been the most vocal in raising the original concern.

PCSK9 loss-of-function: an 88% relative risk reduction that should give everyone pause [3:41:09]

• Austin raises a finding from the landmark PCSK9 sequence-variant paper (New England Journal of Medicine) studying African-American populations with naturally occurring PCSK9 loss-of-function variants: the LDL difference between the variant carriers and controls was a comparatively modest 30–40 mg/dL, yet the relative risk reduction in major adverse cardiac events was 88% — wildly disproportionate to what standard LDL-lowering dose-response models (which predict roughly 25% relative risk reduction per 38.7 mg/dL, or 1 mmol/L, of LDL lowering) would predict for that magnitude of change.

• His explanation for the discrepancy leans on two factors: first, methodological — the study’s composite endpoint included softer outcomes like silent MI and EKG changes, which can meaningfully inflate an apparent benefit; second, mechanistic — PCSK9 itself, independent of its effect on LDL receptor recycling, appears to carry direct pleiotropic effects on inflammation, angiogenesis, and smooth muscle cell recruitment (PCSK9 blood levels predict cardiovascular events even after statistically adjusting for LDL-C), meaning at least part of the apparent benefit of PCSK9 loss-of-function may not run through LDL lowering at all.

• Both flag this as a direct violation of the “consistency” criterion within Bradford Hill’s framework as commonly invoked in EAS-style consensus papers: if multiple independent lines of evidence (statins, ezetimibe, PCSK9 inhibitors, genetic variants) are claimed to converge cleanly on the same dose-response relationship between LDL lowering and event reduction, findings like this — wildly disproportionate benefit relative to the LDL change observed — represent a genuine inconsistency the framework should have to reckon with, not explain away after the fact.

2004: the regulatory hinge point, and CARDS vs. ASPEN as a natural experiment [3:46:01]

• Austin and Dave walk through the regulatory shift both consider an underappreciated turning point in statin trial history, occurring around 2004: stricter trial registration requirements (reducing the ability for unfavorable trials to simply go unpublished and unknown), paired with a shift in clinical-equipoise standards that effectively eliminated genuine placebo-controlled statin trials going forward — from that point on, even control arms in new cholesterol-lowering drug trials were generally required to receive some baseline statin therapy, since withholding it was no longer considered ethically defensible given existing evidence.

• Austin presents what he considers a striking natural experiment bridging that exact line: CARDS and ASPEN, two trials with nearly identical design and intervention (10mg atorvastatin vs. placebo in diabetic patients), differing mainly in patient population (CARDS was purely primary prevention; ASPEN mixed primary and secondary prevention, a somewhat sicker population in which drug benefits typically appear larger, not smaller) and in which side of the 2004 dividing line each trial fell on. CARDS — the earlier, pre-2004 trial — is the study most frequently cited as definitive proof of statin benefit in diabetics (40–44% relative risk reduction in major adverse cardiac events). ASPEN, the later, ostensibly higher-risk-population trial, found a completely non-significant result for the same endpoint.

• Austin extends the analysis to a third trial, 4D (diabetic patients on hemodialysis), where statins are independently known to show no event-reduction benefit at all. Pooling all three trials specifically designed around a diabetic inclusion criterion in a meta-analysis shows no significant benefit — and even excluding 4D entirely and pooling only CARDS and ASPEN together, the combined result still fails to reach significance. Austin’s conclusion, offered with the explicit caveat that this should not be read as blanket advice to discontinue statin therapy in diabetics (there remains a body of subgroup and post-hoc analyses supporting use): with respect to the highest quality data available, there does not appear to be a significant benefit of statins in individuals with diabetes as a primary-evidence matter — a position he anticipates will generate pushback, and notes he received no substantive rebuttal when raising the same point directly with a prominent and famously combative cardiology Twitter account.

PESA dissected: N of 2, N of 9, and the bar chart that visually implies something the data does not support [2:44:09]

• Austin walks through his specific methodological objections to PESA (Progression of Early Subclinical Atherosclerosis), the study most commonly invoked as a comparison cohort against the LMHR research given its similarly low-baseline-risk population. First: no pre-registered analysis plan is identifiable in the published methodology, and the study used stepwise regression — a variable-selection technique Austin flags (citing epidemiologist Miguel Hernán’s work on causal inference methodology) as particularly prone to producing spurious correlations depending on the order and manner variables are entered and removed from the model.

• The statistical power problem in PESA’s most widely circulated graph (LDL-C bucketed in 10-point increments against subclinical atherosclerosis burden): the lowest LDL bucket (50–60 mg/dL) contains only 2 participants; the next bucket (60–70 mg/dL) contains 9. Both are visually displayed as full-height bars identical in format to buckets containing 233, 261, and 275 participants further up the distribution — a visualization choice Austin considers actively misleading regardless of intent, since it implies even statistical weighting across genuinely incomparable sample sizes.

• Austin cites a separate, larger analysis from Miami Heart (Cromwell) directly relevant to PESA’s implied conclusion: 1 in 12 individuals with LDL-C below 70 mg/dL — a statin-free cohort, eliminating confounding from treatment — had a calcium score above 100, a considerable atherosclerotic burden given presumably lifelong low LDL exposure.

• Drilling into PESA’s “optimal cardiovascular risk factor” subgroup specifically (the closest available proxy to a genuinely healthy population, with blood pressure, A1C, and other markers in clearly normal range): the stepwise regression model still returned roughly a 16% increased risk of subclinical atherosclerosis per 10 mg/dL increase in LDL-C even within this subgroup. Austin’s caveat, which he is explicit is not an attempt to smuggle triglycerides or HDL back in as “the real cause”: the same analysis showed parallel changes in triglycerides and HDL tracking with the higher-atherosclerosis-burden participants within that same “optimal” subgroup — raising the live possibility that LDL-C is functioning as a shadow marker for a marginally less metabolically optimal profile within a group that, by the criteria used, still does not fully exclude milder insulin resistance.

• The clinically relevant punchline, contextualizing PESA against meaningful endpoints rather than subclinical imaging alone: 94% of PESA’s “optimal cardiovascular risk factor” subgroup had a calcium score of zero, and the companion MESA cohort (16.6 years of follow-up) found no significant LDL-C association with actual cardiovascular events — only with subclinical imaging findings — reinforcing the distinction both Austin and Dave return to repeatedly throughout the conversation between subclinical atherosclerosis as an imaging finding and atherosclerotic cardiovascular disease as a clinical, event-producing condition.

The number needed to treat that almost nobody discusses: 3,571 [3:01:09]

• Austin supplies the number-needed-to-treat figure for statin therapy in a calcium-score-zero population: 3,571 patients would need to be treated over ten years to prevent a single major adverse cardiac event. For the calcium-score 1–100 group (still a low but non-zero burden), the NNT improves substantially but remains modest — roughly 100 patients treated over ten years to prevent one event, meaning 95 of every 100 people in that group see no benefit at all over the studied period.

• Both flag this NNT context as essential background for evaluating “primordial prevention” — the more recently popularized framing advocating LDL-lowering even in metabolically healthy populations with no detectable disease, on the theory that orphaning atherosclerosis entirely is achievable purely through sufficiently aggressive ApoB reduction regardless of metabolic context.

Closing: the origin story, the Nobel Prize joke, and what comes next [3:09:43]

• Austin recounts how he and Dave actually met in person: a Twitter exchange after Austin sent Dave a paper on an SGLT2-inhibitor knockout mouse model relevant to the lipid energy model, followed by phone calls, and finally an impromptu in-person meeting in Omaha during Jamie Seeman’s Hard to Kill Summit — Dave DMing to ask if Austin could meet immediately, Austin agreeing despite not yet having showered that day, and the two talking lipids and transcytosis for roughly two hours in Austin’s home before his son Paxton woke up, at which point Austin introduced his infant son to “a future Nobel laureate.”

• Austin’s closing reflection on what he considers Dave’s most underrated trait: not raw intelligence or persistence alone, but a consistent, deliberately cultivated commitment to cordial, good-faith engagement even under direct attack — citing Dave’s Ivor Cummins interview, where Dave brought his own laptop specifically to quote Attia’s “pre-buttal” comments verbatim from the transcript rather than paraphrasing, as a small but telling example of intellectual honesty under provocation.

• Dave’s own closing reflection, offered with visible emotion: the experience of capturing his real-time, upstaged reaction to seeing the very first LMHR baseline dataset on camera — mid-presentation from Dr. Budoff, distracted enough by what he was absorbing that he failed to properly respond to the documentary crew’s direction to reposition for a different camera angle — stands out as one of the most meaningful moments of the entire eight-year project, not for any scientific reason but simply as an honest record of what discovery actually felt like in the moment it happened.

Share

Leave a comment

Who is Dr. Eric Westman, and why his YouTube channel works [1:04]

• Dr. Eric Westman is an associate professor of medicine at Duke University, where he runs the Keto Medicine Clinic. He describes his current practice plainly: he is practicing the best internal medicine of his career, and the entire mechanism of that improvement is changing the food.

• His YouTube channel grew out of a practical need: patients were arriving citing things they had seen online, so his company built a channel to get more information to people who could not make it to the physical clinic. The channel now produces reaction videos with deliberately funny thumbnails his same-age patients roll their eyes at — and pulls roughly a million views a month, a volume he notes is not typical for a practicing physician.

• His read on why it works where so many other doctor-hosted channels do not: most physician YouTubers fall into two camps — doctors still defending the old pharmaceutical paradigm, or physicians/PhDs who are credentialed but not actively practicing. Grassroots clinical practice, in his view, teaches things that go well beyond book learning, and he considers his ongoing, decades-long direct patient contact the differentiator. Patients regularly tell him “I saw you on TV,” not realizing they mean YouTube — evidence, he says, of how completely cable and network television have been displaced for his patient demographic.

Subscribe now

1998: the first two Atkins patients, and the cholesterol result nobody expected [7:01]

• Westman’s entry point into low-carbohydrate medicine was not theoretical. In 1998, working as an ambulatory care physician and smoking-cessation clinical trialist at the Durham VA, two patients — mostly Vietnam veterans, with a few WWII veterans in his panel — independently lost over 50 pounds on their own. One patient, somewhat sheepishly, told him: “All I did is eat steak and eggs.” Westman had never heard of the Atkins diet at that point.

• He read the books, talked to colleagues, and went in expecting the worst: he assumed cholesterol would rise given the volume of saturated fat being consumed by very overweight patients. He offered to measure it. In both patients, LDL went down and HDL went up — the opposite of what 1998-era medical training predicted. His own statistical instinct kicked in: lightning could strike once, but two patients in a row defying the expected direction was a real signal worth investigating rather than dismissing.

• He called his own mentor in nutrition and obesity, Dr. Jim Anderson at the University of Kentucky (an oat-bran cholesterol researcher), to ask whether he should study the Atkins diet. Anderson’s answer, delivered with visible hesitation: “It’s a balanced diet... it’s not my favorite, though.” A hedge, not an endorsement — but enough of a green light, combined with the in-clinic results, for Westman to proceed. One of the hospital dietitians complained to the hospital director attempting to shut the study down; the research committee did not close it, but required monthly reports demonstrating Westman was not harming anyone.

Visiting Dr. Atkins, the 2004 randomized trial, and the freak accident that killed him [12:17]

• Westman and Jeff Volek (then at Ohio State) approached Dr. Atkins directly for research funding around the same period. Westman travelled to Atkins’s four-story Midtown Manhattan office at the peak of his career — thirty years into the practice — and observed patient visits with nurse Jackie Eberstein before being walked down the hall to see Atkins himself, treated with the deference of a senior teaching physician.

• A BBC documentary covered the early Atkins research, including Westman running laps at Duke to visually establish himself as “fit” on camera, and researcher Gary Foster (later moving to Weight Watchers). The documentary closed on Dr. Atkins’s death in 2003 — he slipped on ice during a freak snowstorm in New York City during Holy Week, sustaining a brain bleed from the fall. Westman personally knows Dr. Keith Berkowitz, the physician who found Atkins on the sidewalk and later inherited much of his practice. Atkins never lived to see Westman’s second study — the randomized controlled trial published in the Annals of Internal Medicine in 2004 — though he knew before he died that the results were going to be positive.

• The “diet doctor dies obese” media event followed almost immediately: a vegan-advocate physician obtained Atkins’s death certificate (improperly, by Westman’s account — they are not supposed to be released this way), reporting his weight after a week or two of ICU fluid resuscitation as though it reflected his actual living condition. The story circulated globally.

• Mrs. Atkins, who had married Dr. Atkins relatively late in life, endowed seven professorships at US universities following his death and personally requested, in front of the Duke dean, that one go to Westman. It did not happen, and she never returned to Duke. Westman stayed anyway, in part because of Duke’s own unusual historical relationship to dietary medicine.

Duke’s buried history: the rice diet and how Westman found his way to obesity medicine [15:10]

• Westman recounts the origin of the Rice Diet, founded at Duke in the 1930s: a European-accented physician treating a hypertensive patient (no blood pressure drugs existed at the time) was startled when she returned with dramatically improved blood pressure. Misunderstanding her own thick-accented instructions, the patient had simply eaten rice — and the radical sodium restriction reversed her hypertension.

• Duke later used the rice diet framework for weight loss, including residential programs Westman worked within during the early 2000s at the Duke Diet and Fitness Center. Watching diabetes, hypertension, and obesity reverse using simple low-calorie, exercise-based protocols — not keto, not even low-carb — is what initially pulled Westman from smoking-cessation trial work toward obesity medicine, roughly eight years before his research shifted his own clinical practice fully to a keto model.

• He recalls being on an American Heart Association press panel in Chicago for the publication of his first 50-patient study, where a reporter pointed out that millions of people had already done the Atkins diet, implying the study was redundant. Westman’s reply — “millions of books have been sold; that doesn’t mean everyone’s doing it” — became something of a recurring posture: be honest that book sales are not adherence data, and do not pretend to be a shill for a phenomenon you are trying to study honestly.

The decade-long taboo on studying high-fat diets [18:49]

• Westman describes attending Obesity Society meetings for ten consecutive years asking colleagues directly: “Tell me I shouldn’t study this. Tell me something that’s bad about it.” Nobody could give him a reason not to — but almost nobody was willing to actually study it either. He became, in his own words, an irritant for repeatedly asking why a diet nobody could disprove also was not being investigated.

• His diagnosis of the resistance: a deep institutional inability to conceptually accept “giving people so much fat in the food,” reinforced by decades of saturated-fat-as-pathogenic literature that, in his framing, being widely published does not make true. Dave draws the parallel to his own early skepticism a decade ago — reading about Atkins and thinking it sounded “myopic,” worrying specifically about dietary cholesterol and red meat — until his own climbing A1C and the diabetic-forum testimonials he kept encountering (running directly counter to major institutional guidance recommending up to 60g of carbs per meal) pushed him to actually test it.

• Westman’s own due-diligence requirement before proceeding, framed in “first do no harm” terms: he needed institutional cover proving carbohydrates were not essential before he could ethically proceed. He found it in an underappreciated Institute of Medicine section explicitly stating carbohydrates are not essential nutrients — a citation he says almost nobody references, even though it came from one of the few genuinely unbiased sources available at the time, predating the popular paleo/primal/hunter-gatherer framing entirely.

The LDL response is not what either side expects: the BMI-stratified meta-analysis [22:53]

• Westman’s randomized trials, later replicated worldwide, consistently showed average LDL did not change on a low-carb diet — though it moved in both directions for individuals, cancelling out in the aggregate. He notes most physicians are not well-equipped to parse this nuance and simply round it to “LDL goes up.”

• Dave brings in the meta-analysis he and Adrian Soto-Mota published, pooling RCTs with a keto-naive or low-carb-naive arm and stratifying results against BMI under the lipid energy model framework: below a BMI of roughly 25, carbohydrate restriction is associated with rising LDL — the area of concern. Between BMI 25–35, LDL typically does not move much. Between BMI 35–45, LDL typically goes down. This BMI-stratified pattern explains why the dramatic LDL spike phenomenon was barely visible in the original Atkins-era research: in that era, low-carb was almost always a late-resort intervention chosen by people who were already overweight and metabolically unwell after exhausting other diets. Only in roughly the last decade have meaningfully lean, metabolically healthy people begun choosing to restrict carbohydrate voluntarily — creating, for the first time, the population in which LDL “goes through the roof.”

Total carbs vs. net carbs: the difference between prescription strength and over-the-counter [27:05]

• Dave credits Westman with a distinction he only recently fully adopted: total carbs versus net carbs. Net carbs — total carbohydrate minus fiber and sugar alcohols, on the theory that these do not get absorbed or affect metabolism — became common with the New Atkins book in the early 2000s, a book Westman himself worked on. But it was not how the original Atkins clinic operated.

• When Westman visited Atkins’s actual practice in 1998, the patients were not on net carbs. They followed a one-page handout limiting total carbs to 20 grams per day — maintained at that induction level, not for two weeks as the book suggested, but until the patient reached their goal, often years. Had Westman not personally visited the clinic, his research would have replicated the book’s protocol rather than the clinic’s actual practice — a meaningful divergence, since other investigators who studied “the Atkins diet” by reading the book rather than visiting the clinic (including Gary Foster’s early study, which added carbs back after two weeks) effectively studied a diluted, lower-dose version and got correspondingly weaker results.

• Westman’s framing, which he is actively trying to popularize: net carbs is “over-the-counter” low-carb or keto — not prescription strength. Total carbs, kept around 20–30g, is prescription strength. He cites Virta Health’s 30g total-carb standard as an example of an organization using the stronger formulation. His own clinical observation: getting someone off ten medications, including insulin, generally is not going to happen on the over-the-counter version.

• The other major divergence from the original Atkins protocol that Westman discovered only by visiting in person: the original sheet of paper handed to patients explicitly limited cheese, cream, oils, and mayonnaise — a detail entirely absent from popular modern “macro-counting,” “bulletproof coffee,” and fat-bomb-heavy approaches to keto that emerged well after Atkins’s death. Westman is explicit that he is not a fan of bulletproof coffee or oil-stacking generally, considering these later additions a kind of unnecessary decoration on what should be a simple core principle: keep total carbs low.

Banting before Atkins: low-carb medicine predates the epilepsy ketogenic diet by a century [38:14]

• Westman makes a point he considers frequently muddled by people teaching low-carb and keto online: the modern ketogenic diet for epilepsy is not the historical root of therapeutic carbohydrate restriction. The lineage actually runs through the Banting diet in England (a low-carb protocol for obesity), then through Oertel and Allen and Joslin at what is now the Joslin Clinic in Boston — used for diabetes roughly a century ago, predating the epilepsy ketogenic diet’s development. He treats this as a useful tell: if someone presents the keto diet’s origin story as beginning with epilepsy treatment, that is a signal they may not have dug into the deeper history.

• The practical implication of the century-old diabetes lineage: those early diabetic diets were not necessarily strict macro-counted ketogenic protocols — simply low-carbohydrate was sufficient to be clinically effective, which matters for how strict modern practitioners need to make the intervention to see benefit.

• Dave connects this to his own research into the early American Diabetes Association founding guidance, prompted by his frustration with a recent ADA dietary plate release: the original ADA dietary instructions, from its founding era, were substantially closer to a modern low-carb approach than anything the organization currently recommends. The shift away from that original position happened gradually over the 20th century, not from the organization’s actual founding.

• Both credit Gary Taubes’s work — Good Calories, Bad Calories for documenting the weakness of the low-fat evidentiary record, and his more recent book on the history of diabetes treatment — for excavating how, once insulin became measurable in blood, overweight patients with high blood sugar were already found to have high insulin, meaning insulin therapy was being added on top of an existing hyperinsulinemic state rather than correcting a deficiency, a sequencing error baked into clinical practice that both consider still under-examined today.

Why doctors default to prescriptions, and the “pitch it like an FDA drug” reframe [44:24]

• Westman’s direct challenge to physicians reluctant to discuss diet: this approach has been studied to a standard comparable to FDA drug approval, so pitch it like a drug. His test case: would you withhold an effective medication from a patient because it required taking a pill four times a day? Then why withhold a dietary intervention with comparable evidence because it requires effort?

• He flags a specific clinical gap: most doctors have simply never asked a patient to write down what they actually eat for breakfast, lunch, and dinner. A one-page intake form accomplishes more diagnostically than most physicians realize, and the absence of this basic question is, in his view, a major reason food-based intervention gets skipped in favor of medication by default.

• His key message to patients and prospective patients: “This will work if you do it.” Confidence and simplicity, not apps, ketone meters, or macro tracking, are what make the approach sustainable in his clinical experience. He references having reversed 18 years of diabetes in two weeks in a recent case — routine enough in his practice that his own nursing staff jokingly tell him he is “getting out of control” when he announces it, even though it happens regularly.

The “challenge cases,” the bagel confession, and why even obesity-medicine doctors are choosing the shot [47:03]

• Westman, a past president of the Obesity Medicine Association, observes that even doctors within his own specialty society are increasingly prescribing GLP-1 medications by default — partly because it is a higher-volume practice model, allowing more patients to be treated per unit time. He recounts a colleague, visibly improved, admitting under questioning: “I want my bagel in the morning,” having chosen the injection specifically to avoid changing the underlying diet.

• Dave’s framing of this pattern within his own extended family: there are “challenge cases” who explicitly prefer something they can take indefinitely rather than do the ongoing behavioral work — and while it is not the choice he would make for them, he is careful to note that, at minimum, it is an informed choice rather than ignorance, which is a meaningfully different situation from patients who were simply never offered the diet option at all.

• Westman is explicit that the FDA-approval process gives both doctors and patients a false sense that all possible long-term interaction effects have already been vetted — something he considers naive given how new these drug classes are and how little is known about decades-long use.

Tricks of the trade: chaffles, cauliflower rice, and why substitution knowledge is its own form of expertise [48:56]

• Westman describes the practical, granular knowledge that distinguishes an experienced low-carb clinician from a generic physician simply telling a patient “eat less”: chaffles (a cheese-and-egg waffle substitute), cauliflower rice, and sugar-free gelatin as a fruit-flavor substitute are all concrete substitutions he actively teaches and points patients toward, including pointing them to YouTubers doing detailed chaffle taste-test comparisons.

• His bicycle-learning analogy for patients who do not succeed immediately: nobody learns to ride a bike on the first attempt without falling. The same applies to dietary adherence — carb slip-ups are part of the learning process, not evidence the method has failed, and an effective clinician treats early stumbles as data to troubleshoot specific substitution gaps rather than a verdict on the patient’s capability.

• He distinguishes his own strict, in-network clinical practice from a colleague’s cash-pay, medication-first practice in the same area — patients cycle between the two depending on what they can afford and how much structure they currently want, which Westman treats as a healthy reflection of patient choice rather than a competitive problem.

Sugar addiction, AA meetings on the road, and the January relapse rush [51:54]

• Westman treats sugar addiction as functionally comparable to alcohol addiction in a meaningful subset of patients, citing colleagues who attended AA meetings even while travelling out of town for medical conferences — illustrating how critical daily contact and community support can be for maintaining abstinence from any addictive substance.

• His clinical framing of holiday relapse: it is not a moral failure, it is a relapse, treated with the same compassionate, matter-of-fact “get back on” approach used in addiction medicine. January is consistently his busiest clinical month as patients return after holiday slips. His practical advice: remove trigger foods from the house entirely rather than trying to moderate around them — “like taking a band-aid off,” in his words, less painful done all at once.

• Dave extends this with his own concrete tactic: bringing prepared low-carb food (cheese sticks, homemade low-sugar chocolate chip cookies) to family holiday gatherings specifically to avoid the all-or-nothing trap, while also naming the psychological mechanism he finds most dangerous — the moment a single planned “cheat” reopens an entire category of food as permissible going forward, undermining resolve well beyond the original event.

Carb creep, social pressure, and the “list of foods in English” teaching method [58:04]

• Westman names “carb creep” as an expected, near-universal three-to-six-month phenomenon — a whack-a-mole pattern where region-specific trigger foods (sweet tea in North Carolina, grits) quietly re-enter a patient’s diet, often without the patient consciously registering the lapse. His teaching method deliberately avoids technical nutrition vocabulary: a plain-English list of allowed foods, translated into Spanish and other languages by collaborators, removes the need for patients to understand amino acids or macronutrient percentages at all.

• A revealing anecdote from his own clinic: a patient who had reintroduced sweet tea defended herself by saying “you never said I couldn’t” — prompting Westman to reflect that either the specific instruction was missed in a particular class session, or the patient (possibly sleep-deprived from undiagnosed sleep apnea) was not retaining information in the room. Either way, it reinforced his view that teaching methodology and follow-up structure matter as much as the diet’s content.

• His broader point on fit-fluencer culture as a clinical hazard: patients who are doing better than they have ever done on their own personal trajectory can still feel like failures by comparing themselves to viral 90-pounds-in-45-days transformation stories, leading to discouragement and premature abandonment of an intervention that was, by any honest internal comparison, working.

Westman’s own A1C, the Randle cycle, and why “abnormal” blood sugar on keto might be adaptive [1:06:02]

• Westman describes an increasingly common pattern in patients who have been low-carb or keto for five to ten years: their fasting glucose and A1C never quite normalize to what would be considered “normal” by carbohydrate-eater reference ranges, despite excellent metabolic markers otherwise. Dave shares his own numbers as a parallel case: fasting glucose in the 90s when militantly keto, A1C bouncing between 5.5 and 5.9, fasting insulin between 2–5 µIU/mL, and a flat CGM line throughout the day.

• Westman’s mechanistic explanation, using the older HOMA-IR framing: risk is a function of the multiplicative product of glucose and insulin, not glucose alone. If insulin has come down dramatically even while glucose has not fully normalized to a carbohydrate-eater’s range, the calculated risk is much lower than the raw glucose number alone would suggest.

• Dave layers in his own preferred mechanism, the Randle cycle operating body-wide: a fat-adapted person has more fatty acids circulating in cellular cytosol, which down-regulates GLUT4 expression and glucose uptake in peripheral tissue, sparing circulating glucose for obligate glucose-using cells like red blood cells. He draws a direct parallel to acute infection, where blood sugar rises specifically to support immune response — a contextual, adaptive elevation rather than a pathological one. Both agree that diabetes is fundamentally a disease of dysregulation (large swings, hypoglycemic responses, climbing area-under-the-curve insulin) rather than simply elevated fasting glucose in isolation, and both want to see this distinction studied directly with insulin-area-under-curve data rather than glucose snapshots alone.

Own Your Labs as the open dataset low-carb research has never had [1:10:01]

• Dave describes Own Your Labs’s broader ambition beyond simple lab-test reselling: incentivizing users to submit anonymized blood work alongside demographic and dietary context, with the explicit goal of building an open dataset that does not currently exist anywhere for the low-carb population — specifically to test hypotheses like whether consistently low fasting insulin alongside higher fasting glucose/A1C in lean, low-carb individuals tracks differently than the same glucose numbers in a hyperinsulinemic, carbohydrate-eating population.

• Westman explicitly endorses the idea and flags the regulatory tightrope: HIPAA compliance and careful anonymization are non-negotiable, but the absence of any open-source low-carb dataset is, in his words, “a problem” that a transparent registry could meaningfully begin to solve. Both frame this as a genuine gap rather than a hypothetical one — there is simply no large, open, low-carb-specific dataset currently available to outside researchers.

Animal models, rabbit cholesterol, and the 40,000-mouse search for the right phenotype [1:14:10]

• Westman recommends the book Rigor Mortis (a journalistic investigation, in the Gary Taubes tradition, of weak animal-model science) and describes its core argument: researchers often work backward from a desired human conclusion, engineering an animal model specifically capable of producing the expected result, then treating that constructed model as independent confirmation of the original hypothesis — a circularity he considers endemic to cardiovascular research specifically.

• He relays a conversation with Gary Taubes about an investigator who reportedly screened through roughly 40,000 mice to find a strain capable of producing the desired atherosclerosis phenotype — illustrating how much selective engineering underlies “the mouse model” as commonly invoked in cardiovascular literature.

• Dave extends the critique with specifics from his own research into animal-model lipid biology: most rodent atherosclerosis models rely on specific obesity-prone lineages fed an atherogenic chow (often containing carbohydrate alongside the cholesterol load, not isolated dietary cholesterol as commonly assumed), and these species naturally run lower ApoB and make heavier relative use of HDL than humans do — compounded further by the fact that rodents lack cholesteryl ester transfer protein (CETP) entirely, a key player in the lipoprotein exchange processes central to human lipid metabolism. Many models also use transgenic mice engineered for elevated ApoB synthesis, moving the model even further from the species’ own native physiology before any dietary intervention is applied.

• Westman shares a directly relevant anecdote: visiting a primate-colony researcher (a closer model to humans) who confirmed that drugs successfully stopped atherogenesis in the colony — but when Westman asked whether the team had tried varying the diet itself (an Atkins-style or low-carb diet) rather than only testing pharmaceutical intervention against a fixed “typical American diet,” the researcher responded that diet variation was simply not the point of the study; the goal was creating atherosclerosis reliably enough to then test a drug against it, not testing whether diet itself could prevent or reverse the process in the first place.

• The original cholesterol hypothesis origin story, walked through together: Anitschkow’s early rabbit-feeding experiments used a herbivore — rabbits are not natural cholesterol consumers at all — fed purified cholesterol mixed into chow that, on closer inspection of the original methodology, also contained carbohydrate. When researchers later attempted to replicate the atherosclerosis result in dogs and rodents (omnivores, closer to human dietary physiology), it did not replicate, a finding both consider a serious and still under-acknowledged problem with the hypothesis’s foundational evidence.

Metabolic status as the confounder that touches everything: the “Dave Island” thought experiment [1:23:50]

• Dave constructs a thought experiment to isolate the core methodological problem running through the entire conversation: if you were taken to an island and forced to eat sweets all day, several lipid changes would be predictable regardless of any specific causal LDL mechanism — HDL would fall, triglycerides would rise, LDL would likely rise modestly (more so at higher BMI), and small dense LDL particle proportion would increase, the latter because triglyceride-rich lipoproteins arriving at maximally full fat cells have nowhere to deposit their cargo and remain in circulation longer, undergoing more CETP-mediated remodeling in the process.

• His point: any intervention that subsequently lowers LDL in a population already carrying this metabolic dysfunction cluster cannot be cleanly interpreted as “LDL-lowering caused the cardiovascular benefit” without first asking whether the intervention also improved appetite, induced weight loss, or shifted metabolic status more broadly — because all of those co-occurring changes would independently move HDL, triglycerides, and small dense LDL particle proportion in the same favorable direction, creating an unavoidable confounding structure between the lipid profile itself and whatever underlying metabolic state produced it.

• This is, in Dave’s explicit framing, the foundational reason lean mass hyper responder research matters: only a population where the LDL elevation is decoupled from metabolic dysfunction — elevated for the opposite reason, via fat-adaptation in an already metabolically healthy person — allows the field to isolate how much of the LDL-atherosclerosis association in conventional populations is being driven by LDL itself versus the metabolic dysfunction that typically travels alongside it.

Inside the new LMHR paper: heterogeneity, the second scan, and why no clean topline number exists yet [1:30:46]

• Westman, having read the newly published paper closely (he mentions handing it out a dozen times that week), asks pointed questions about the underlying plaque scoring methodology. Dave explains the semi-quantitative scoring system: cardiology readers assess 15 predetermined coronary segments on a 0–3 scale each (no plaque, mild, moderate, heavy), summed into a total plaque score ranging 0–45, alongside separate quantitative measures (non-calcified plaque volume, calcified plaque volume, percent atheroma volume) read via the CLARiFY methodology.

• The headline finding from the first scan: roughly two-thirds of the cohort scored zero on total plaque score despite self-reported high LDL sustained for an average of 4.5+ years. But the critical complication emerging from the second scan, taken approximately a year later: pronounced heterogeneity. The cohort does not move together as a single block — there is a clear, statistically distinguishable separation between slow-to-no progressors (the majority) and a smaller subset of fast progressors, a pattern Dave notes mirrors what has been previously observed in monogenic familial hypercholesterolemia populations, lending weight to the idea that imaging-based risk stratification may be more clinically meaningful than LDL level alone, regardless of how that LDL became elevated.

• Dave is explicit and repeated about why he cannot yet give Westman (or anyone) a single clean topline “the average change was X” number: doing so would misrepresent a population that statisticians on his team insist cannot be treated as one homogeneous group. He uses his wife’s side of the family (notably shorter than himself) as an illustrative analogy: the mean height of a group can be 5’4”–5’5”, yet literally nobody in the group is actually that height — a vivid illustration of why reporting only a population mean can actively obscure rather than reveal the real underlying structure of the data, and why a second, follow-on paper specifically built around identifying and clinically characterizing these subgroups was already in submission by the time of this recording, built directly on the heels of the first.

• Why the paper title and framing emphasize ApoB rather than LDL specifically: not because LDL was excluded (both are reported, alongside non-calcified plaque volume and percent atheroma volume, the latter two now favored over the originally-planned primary endpoints as the team’s statistical sophistication around multiple-comparisons handling matured over the course of the study), but because ApoB has risen sharply in clinical and research prominence specifically within the last two to three years, well after the study’s original design phase began.

What would actually move the needle: Brown and Goldstein revisited, and the call for a fifth-year scan [1:42:16]

• Westman raises the Brown and Goldstein homozygous familial hypercholesterolemia case directly: a three-year-old child with xanthomas, stable angina at three, an LDL around 700 mg/dL, and a first MI at six, with no confounding risk factors — not a smoker, not diabetic, not Type A by personality. This is, in Westman’s words, “the one case where we see that LDL is causal,” meeting Bradford Hill’s criteria for both strength and biological isolation at the time. Dave’s position is not to dispute the case but to interrogate how far it generalizes: he notes parallel reasoning failures in other domains of medicine (a 250 systolic blood pressure clearly causing strokes does not by itself establish the correct treatment threshold or magnitude at far lower pressures), and flags that David Diamond and others have separately raised the hypercoagulable state present in many HoFH patients as an under-examined potential confounder alongside LDL itself — plus the practical impossibility of retroactively measuring inflammatory markers in patients studied decades ago.

• The numeric comparison Dave draws out explicitly: several LMHR participants in the current cohort have individual LDL levels in a similar range to the historical HoFH cases (one participant reportedly around 690 mg/dL, several others above 400), and by the time a planned fifth-year scan is completed, the cohort’s cumulative LDL-years exposure will exceed what Brown and Goldstein’s child patients had already accumulated by the time symptoms first appeared — a genuinely novel natural experiment, in Dave’s framing, precisely because it isolates extreme LDL exposure in a population without the accompanying five-and-a-half decades of conventional metabolic dysfunction that defines the broader population from which the original cholesterol hypothesis was built.

• Both are careful to draw the line precisely: the current paper’s finding of no LDL/ApoB association with plaque presentation or progression at the population level does not, on its own, constitute evidence that intervening to lower LDL in a fast-progressor subgroup would be harmful or pointless — only that it complicates a “simple,” scale-independent version of the LDL hypothesis. Dave is explicit he is finalizing a study extension, hoping to bring as many original participants back as possible for five-year scans, specifically to test whether the heterogeneity resolves into a detectable dose-response signal with more follow-up time, or whether the slow-to-no-progression pattern holds even longer term.

Heart failure, SGLT2 inhibitors, and the cognitive dissonance of accidental ketosis [2:10:34]

• Westman recounts a patient calling from the hospital reporting ketoacidosis — not from a ketogenic diet, but from an SGLT2 inhibitor (a glucose-leaking diabetes drug, e.g. Jardiance, Invokana) that he had never previously seen trigger ketoacidosis in his keto-experienced clinical population. These drugs are independently being studied and found effective at reducing heart failure recurrence, and the mechanism researchers have converged on is the resulting state of mild, drug-induced ketosis itself.

• The dissonance Westman finds professionally striking: trials are now underway infusing exogenous ketones directly into patients with heart failure and observing short-term functional improvement — yet when he points out that a ketogenic diet achieves the same endogenous ketone elevation without a drug or infusion, the response from colleagues is reliably “no, that’s not good for you.” The same metabolic state is being pursued pharmacologically while being actively discouraged nutritionally.

• Dave extends the mechanism into cardiac fuel preference specifically: roughly 70% of cardiomyocyte fuel on a mixed diet comes from fatty acids, much of it delivered as triglyceride cargo directly off chylomicrons — which, critically, route through the thoracic duct and reach the heart before undergoing first-pass hepatic metabolism via the portal vein, a fact some drug formulators already exploit by attaching compounds to fat for absorption-routing purposes. He cites an animal model in which cardiomyocyte-specific lipoprotein lipase was selectively deleted, eliminating the heart’s ability to extract triglyceride cargo from circulating lipoproteins — producing severe heart failure — and notes corroborating human case literature on lipoprotein lipase deficiency specifically affecting cardiac tissue, with documented structural and functional cardiac abnormalities.

• Westman shares a clinical case he presented as an abstract: a 45-year-old female patient with heart failure, ejection fraction starting at 20%, lost 140 pounds on a ketogenic diet and saw her ejection fraction rise to nearly 50% (near-normal). He is careful to flag the obvious confounding question — was the improvement driven by the weight loss itself, the ketones themselves, or both inseparably — a familiar limitation Westman has encountered before in low-carb research generally, where investigators are sometimes told the intervention “does too much” to cleanly isolate a single mechanism.

• He references an active Department of Defense-funded research effort (with Dom D’Agostino reportedly involved) studying both endogenous and exogenous ketones specifically in heart failure patients, and notes that many of his own heart-transplant-track referral patients — too heavy at baseline to qualify for transplant — are sent to him specifically because the transplant team appreciates that he can produce significant weight loss through diet alone, without the drug interactions a transplant candidate’s already-complex medication regimen would otherwise have to absorb.

Documentary films as patient education: from Cereal Killers to Cholesterol Code [2:23:36]

• Westman walks through the documentary lineage that shaped both his own thinking and his patient communication over the years: Cereal Killers and Cereal Killers 2 (the latter following Steve Phinney coaching Sami Inkinen and his wife to break the world record for rowing San Francisco to Hawaii after full keto-adaptation); The Magic Pill; Fat Fiction; and First Do No Pharm, on physician Georgia Ede. He highlights one specific lesson he absorbed from Cereal Killers 2 and conversations with Phinney directly: full keto-adaptation for athletic performance can take up to six months, meaning athletes should not begin the transition immediately before competitive season but during the off-season instead — a practical training-periodization detail he has since incorporated into his own patient guidance around exercise.

• Vinnie Tortorich’s documentary work is referenced twice: Fat (an apocalyptic-toned entry in the genre) and his more recent Dirty Keto, specifically calling out “keto” junk foods marketed around the net-carb loophole — reinforcing Westman’s “prescription strength” framing, since dirty-keto products built around net-carb math generally do not qualify under his total-carb standard.

• The Cholesterol Code documentary gets extended discussion. Dave explains its origin as a 2016 blog — essentially a public lab notebook of his own self-experimentation — that evolved alongside the Citizen Science Foundation’s formation and the LMHR study itself. Filmmaker Jen Eisenheart (whose team also made Fat Fiction) and her crew arranged to be present, cameras rolling, for Dave’s real-time reaction to the very first complete dataset from Dr. Budoff weeks before private review, capturing his genuine unscripted response on film, and continued following the project for the subsequent two years through every major data checkpoint and emotional swing along the way.

• Westman highlights what he considers the film’s most important editorial choice: deliberately centering patients who adopted ketogenic diets for efficacy reasons unrelated to weight loss — bipolar disorder, anorexia, eating disorders, type 1 diabetes — who became incidental lean mass hyperresponders as a side effect of treating their primary condition, putting a human face on exactly the clinical dilemma the underlying LMHR research exists to address. He confirms attending the first private screening at the Citizen Science Foundation fundraiser weekend and calls it, simply, “a good flick.” Target release window discussed: this fall, pending finalization of a streaming distribution partner.

Westman’s bookshelf: the required reading list [2:32:01]

• Walking through his physical bookshelf on camera, Westman highlights: Ketogenic (textbook, lead editor Tim Noakes, to which Westman contributed chapters on basic low-carb nutrition and diabetes/obesity); The Art and Science of Low Carbohydrate Living and Performance (Volek and Phinney, which he assigns to curious students and residents); Food Junkies and Jen Unwin’s Fork in the Road on food and sugar addiction frameworks that simply were not part of the clinical conversation twenty years ago; the original Atkins books; Dr. Bernstein’s Diabetes Solution; South Beach Diet Phase One; Protein Power (the Eadeses); Jason Fung’s fasting-focused books; and the near-complete Gary Taubes catalogue, including Good Calories, Bad Calories, The Case Against Sugar, The Case for Keto, and his diabetes history work.

• Ben Bikman’s Why We Get Sick gets a specific compliment: Westman attributes Bikman’s communicative effectiveness partly to having an active university lab of students running new experiments he can fold into talks in near-real-time — including a finding Westman cites that beta cells begin increasing insulin secretion within a day or two, which he connects to why someone keto-adapted for a while might initially “fail” a standard glucose or insulin tolerance test.

• Thomas Kuhn’s The Structure of Scientific Revolutions occupies a special place on the shelf: Westman recounts showing it to Dr. Atkins personally at a dinner, making the case that a genuine paradigm shift requires a framework open-ended enough for other people to solve their own problems within it — a structural feature Westman believes Atkins himself never quite built, since Atkins positioned himself as already having the answers rather than building infrastructure for others to extend the work, in contrast to what Westman sees happening now as other physicians independently adapt and extend the approach within their own practices.

Filming in Shenzhen: a hospital floor with a complete low-carb library and 30,000 people watching live [2:37:34]

• During a 2020 round-the-world documentary filming trip (36 interviews across 17 countries and 28 cities, undertaken before the formal Cholesterol Code production began), Dave’s first stop was Beijing in early January — arriving just as early reports of a novel illness in Wuhan were beginning to circulate, a detail that in retrospect bookends the entire trip with unintentional historical timing.

• In Shenzhen, Dave visited a hospital where a physician was leading a low-carb/keto-interested clinical team, discovering an entire hospital floor with a dedicated library containing what Dave describes as essentially every low-carb book he had ever encountered. He was asked to give an impromptu presentation to roughly 30 people physically present in the room — with a translator simultaneously relaying the talk to a connected Facebook group of 30,000 people watching live, illustrating the scale at which Chinese audiences were already engaging with this material. Westman and Jackie Eberstein had separately visited the same clinic system previously and recall being shown the same small library room, underscoring how closely the international low-carb clinical community tracks itself even across language and institutional barriers.

• Both credit Jackie Eberstein — Dr. Atkins’s original clinical nurse, now retired to Richmond, Virginia, and still an active mentor figure for Westman over the following two decades — as one of the most knowledgeable people either of them has encountered in the space, frequently sounding more clinically fluent in lectures than many credentialed physicians they have separately observed. Her own account of being hired by Atkins despite open skepticism about his reputation (telling him directly she did not want to work for him “because I found out who you were,” to which Atkins reportedly replied simply, “Okay, you’re hired”) is offered as a small but illustrative window into Atkins’s personal character beyond his public caricature.

Public health schools, Aramark, and why hospitals still serve carbohydrate-heavy “heart healthy” meals [2:44:08]

• Westman names a structural barrier rarely discussed publicly: schools of public health, even at institutions geographically adjacent to his own low-carb-friendly clinical practice (he cites UNC, ten miles from Duke), remain staffed predominantly by faculty trained within and committed to the low-fat dietary model. He recounts presenting low-carb outcomes data to a UNC public health group, only to be told afterward that the same researchers, when they raised the topic with their own cardiologist colleagues, were directly told “you can’t do that.”

• Most hospital food service nationally is contracted out to large food-service corporations (Aramark is named specifically), which Westman says generally defer to existing official dietary guidelines rather than independently evaluating diet-specific clinical evidence — meaning a hospital can serve a high-carbohydrate, low-fat-labelled “heart healthy” meal to a newly diagnosed diabetic patient without any single party in the chain (treating physician, hospital dietitian, contracted food service) taking ownership of whether that specific meal pattern is appropriate for that specific patient’s condition.

• Westman’s summary of the systemic failure, stated plainly: outpatient doctors have largely given up discussing diet; medical schools have largely given up teaching it; and hospitals serve what he considers the wrong diet by default — not from any single actor’s malice, but from an accumulated absence of training, incentive, and institutional ownership at every link in the chain. Dave shares a parallel personal experience visiting a hospitalized family member, where “heart healthy” framing was used to describe a carbohydrate-heavy diet plan with no apparent consideration of the patient’s diabetes status, reinforcing Westman’s point that the labelling itself often substitutes for actual clinical reasoning.

• A counterpoint example offered as evidence the system can be changed locally: a nursing-home/assisted-living operator who implemented a low-carb dietary approach for residents and has seen some patients improve enough to graduate out of needing assisted living entirely — an outcome Westman considers a powerful, underreported existence-proof against the assumption that elderly populations cannot meaningfully benefit from dietary intervention.

Polypharmacy, deprescribing, and the average of seven medications by age 65 [2:59:36]

• Westman cites a striking statistic from his own clinical population: the average number of medications a 65-year-old patient is taking, across both his indigent-care and high-end clinic populations, is seven — commonly two or three for blood pressure, one or two for diabetes, plus arthritis and heartburn medications layered on top, with drug-drug interaction data for many of these specific combinations simply not formally studied.

• His clinical practice of “deprescribing” — systematically reducing medication burden as patients improve metabolically — is, in his words, one of the more enjoyable parts of his job, though it takes real time and is not always fully achievable; sometimes the diet helps a patient get off one or two medications as part of a combination approach rather than eliminating pharmacotherapy entirely. He is careful never to promise full medication elimination upfront, instead framing it honestly to patients as “let’s find out” rather than a guarantee.

• Both flag the structural incentive problem underlying drug-trial design specifically: run-in periods (where the full study population takes the active drug for several weeks before formal randomisation, allowing those who experience side effects to self-select out before the trial officially begins) mean published trial results may understate real-world side-effect rates compared to what clinicians like Westman observe once a drug reaches general practice. Dave raises this as one of his standing methodological objections across drug literature generally, not specific to any one drug class.

What Westman wants to see next: head-to-head trials and a placebo-controlled LMHR drug study [3:19:34]

• Westman articulates the trial he most wants to see funded: a genuine head-to-head randomization of type 2 diabetics with elevated cardiovascular risk into a ketogenic-diet arm, a GLP-1 medication arm, and a usual-care (cholesterol-medication-only, no structured weight-loss support) arm — a study design he considers entirely feasible with current infrastructure, simply not yet funded or prioritized.

• Dave names the specific study he would most want to run within the LMHR research program: a placebo-controlled randomized trial of ezetimibe (rather than a statin) specifically in lean mass hyper responders, reasoning that ezetimibe’s comparatively mild side-effect profile would make a credible placebo-blinded design more achievable than with statins, where symptom awareness might more easily break the blind — paired with longitudinal CTA scanning to directly test whether pharmacological LDL-lowering changes plaque trajectory in this specific population, independent of the metabolic-status confounding discussed earlier in the conversation.

• Both acknowledge the practical funding obstacle this specific trial faces: a genuinely low-baseline-risk population (sky-high LDL, but otherwise textbook-healthy by every other cardiovascular metric) requires a very large sample size to detect a measurable event-rate difference, which makes it a poor commercial fit for pharmaceutical sponsorship despite being, in Dave’s words, exactly the kind of study a thoughtful drug manufacturer with a public-health mandate should want answered.

• Westman closes the substantive research discussion by directly questioning Dave on the underlying imaging methodology’s reliability: is CTA, used this way in a relatively young, low-risk population, a reproducible and validated proxy for actual cardiovascular events the way it is in older, higher-risk populations where it is already well-established? Dave’s answer leans on the statistical logic of event-rate power calculations directly: populations with a higher expected event rate require smaller sample sizes to reach significance, which is precisely why prospectively imaging an extreme, well-characterized exposure population like LMHR — rather than waiting for a comparably large higher-risk cohort to accumulate hard outcomes — is a methodologically efficient way to generate genuinely novel data, even from “only” 100 participants, given how extreme and well-documented their underlying LDL exposure is.

Closing: gratitudenomics, the Ketofest origin of the Lipid Energy Model, and what comes after the first study [3:29:01]

• Dave traces the project’s actual origin point back to Ketofest 2017 in New London, Connecticut (organized by Carl Franklin and Richard Morris of the 2 Keto Dudes podcast), where he ran his first experiment involving people beyond himself and his sister: roughly 22 volunteers, blood-tested simultaneously via a home cholesterol device (PTS Diagnostics) plus LabCorp and Quest draws for cross-validation, following a defined protocol — either fasting or eating hypocalorically for three days beforehand, then eating as much fat as desired (while remaining ketogenic) for three full days following, with blood drawn again on the Monday after. Nineteen of 22 participants showed a pronounced LDL drop; three showed an increase — the pattern that became the founding empirical observation later formalized as the lipid energy model, published roughly three years after this Ketofest weekend.

• Dave flags an unintended consequence of that early protocol becoming public: people began using it specifically to manipulate life-insurance LDL screening results by temporarily overfeeding before a blood draw, prompting his ongoing concern about people sharing the protocol informally without appropriate medical context, particularly for anyone already on cholesterol medication who would need professional guidance to safely adjust dosing around any such experiment.

• On life and reinsurance industry incentives specifically: Dave recounts learning, somewhat to his surprise, that primary life insurers are themselves insured by reinsurance companies (Swiss Re is named specifically, alongside a related conference series organized by ProfetOaks in Cape Town), and that reinsurers — who directly bear the financial consequence of policyholders living shorter lives than actuarially projected — are a more genuinely aligned audience for longevity-focused metabolic health data than primary insurers, drug manufacturers, or general healthcare payers, none of whom Dave believes are structurally incentivized by current business models to priorities long-term wellness over acute-disease treatment revenue.

• Westman closes with a direct invitation and a reflection on how far the collaboration has come: from Dave showing him fluctuating same-day cholesterol readings years ago — a finding entirely outside Westman’s own training, where cholesterol had been taught as a slow-moving, essentially static marker — to a peer-reviewed paper Westman now hands out to colleagues and patients on a near-daily basis. Both explicitly credit the open, collaborative, internet-native, citizen-science model — small recurring donations from thousands of supporters, Patreon seed funding, conference-circuit relationship-building, and consistent public gratitude toward collaborators — as the only realistic way a result this institutionally unwelcome could have been funded and completed at all.

Share

Leave a comment

Who is Alex Leaf, and how wrestling led to nutrition [1:07]

• Alex Leaf describes himself first as a scientific communicator. For the past decade he has worked as a freelance researcher and research writer: people bring him a health question or a supplement formulation problem, and his job is to find, analyse, and synthesise the research and communicate it to audiences ranging from medical professionals to interested lay readers. He started at Examine.com in 2014, came on full-time in 2016, and left in 2019 to pursue other projects, crediting Examine with establishing him in the field.

• His path into nutrition started in competitive wrestling from elementary school through high school, where body image and performance pressure pulled him into fitness and nutrition. He developed an undiagnosed eating disorder — bulimia — as a direct result of wrestling’s weight-cutting culture. When he stopped wrestling in college, he tried to use nutrition to recover, and one of his first resources was Mark’s Daily Apple, which started him on a paleo framework. From there his interest deepened into asking why the recommendations existed at all — a line of inquiry that eventually moved him away from rigid paleo positions (no dairy, no vinegar) that did not hold up to scrutiny.

• Dave probes the wrestling-bulimia connection directly: the sport structurally requires cutting to a lower weight class, with real social motivation attached. Eating disorders are commonly assumed to be a female-coded condition tied to body image obsession, but Alex confirms this is not at all uncommon in wrestling, and that these conditions are broadly underdiagnosed in men because men underreport and underseek treatment. He distinguishes eating disorders from body dysmorphia — the latter being especially common among athletic men and bodybuilders, who can have abs and significant muscle mass and still feel fat or small, driving continued unhealthy behaviour past any reasonable goal.

• Alex’s personal account: in his final wrestling year he weighed about 156 and cycled between three weight classes (164, 152, 145), each with both a floor and ceiling. To make weight for tournaments he sometimes went without food for two days while running three-hour practices plus 10-mile jogs in sweats. He is unambiguous: this is brutal and not healthy by any standard, and because adolescents are still growing and developing, this kind of practice actively damages developmental processes during a critical window.

Subscribe now

Protein, leucine, and why mTOR longevity fears never made sense to Dave [8:20]

• Dave raises a trend he watched develop and fade over roughly three or four years: the idea that mTOR activation from protein intake should be minimised for longevity reasons. He never found this compelling — he is far more concerned about under-consuming the full complement of amino acids than over-consuming them.

• Alex agrees completely and supplies the numbers. The common retort is that people in developed nations already eat “enough” protein — true in the narrow sense that average US intake (roughly 1.1–1.2 g/kg body weight) is sufficient to prevent muscle wasting. But research on what is optimal for building muscle mass, losing fat, and optimising body composition points to roughly 1.6 g/kg as a starting point, with strength athletes and bodybuilders needing a minimum of 2.2 g/kg on rest days to support ongoing protein turnover. These figures come from indicator amino acid oxidation studies, which feed participants amino acid mixtures replicating egg protein and track oxidation until a break point is reached, signalling sufficient intake.

• Dave zeroes in on amino acid specificity rather than the protein category broadly: leucine is the primary amino acid activator of mTOR and is in disproportionately high demand for muscle protein synthesis relative to other amino acids — they are not interchangeable. His 3D printer analogy: cells are protein factories, ribosomes are the original biological 3D printers, and amino acids are the colour cartridges. Just as a red-heavy print job burns through the red cartridge fastest, animal tissue burns through leucine fastest because animal amino acid composition is closer to human composition than plant amino acid composition is. This gives animal protein a numerical compositional completeness advantage — not an absolute superiority, but a “gets you there sooner” advantage.

• Alex’s reframe of the mTOR longevity fear: mTOR is meant to cycle on and off. Chronic overfeeding with insufficient fasting time to let the countersuit (AMPK-driven) process activate is detrimental — but so is excessive fasting without adequate refeeding. Both degrade the body, just through different mechanisms. The point of eating, in his framing, is to maximise the muscle protein synthesis response every time you eat, regardless of meal frequency, and the fasting period is when quality control and cellular maintenance happen.

• Dave’s term for this maintenance window: “the cell’s closing hours.” He is not specifically pro-intermittent-fasting as a protocol; he is pro giving the immune system enough uninterrupted time to do its housekeeping — managing apoptosis throughout the body. His one piece of gut-instinct health advice: not eating within three to four hours of bedtime, which he believes has an ancestral basis and would likely generalise across most people.

What hunter-gatherers actually tell us — and the ancestral fallacy [15:48]

• Alex shares research he wrote on indigenous hunter-gatherer sleep patterns: the most common reason these groups stay awake past 11pm, sometimes until 1 or 2am, is that they did not get enough food during the day, so they forgo sleep to forage or hunt small prey at night instead. Dave connects this to the survival show Alone, where contestants in extreme food scarcity scenarios end up hunting nocturnal prey (slugs near a shoreline, in one case) purely out of necessity — illustrating that ancestral behaviour patterns observed today often reflect acute survival pressure, not a deliberately chosen “healthy” eating window.

• Both agree on a key distinction: there is a difference between what humans did to survive throughout evolutionary history and what optimises health or thriving in the modern environment. Humans cycled through radically different dietary patterns depending on what was available — heavily fat-and-meat-based during the mammoth-hunting era, shifting toward more foraging and carbohydrate as megafauna died out. Nearly every equatorial and tropical hunter-gatherer tribe alive today consumes honey. There is no single ancestral human diet, especially once geography is considered — someone living far north was not eating the same ancestral diet as someone in equatorial Africa.

• Dave’s steelman of the meat-centric ancestral narrative: cave art is the earliest evidence of cultural celebration, and what gets depicted is overwhelmingly the hunt. There was clearly a premium placed on meat as a desirable, celebrated food — whether from clever nutrient-density awareness or simple caloric density and palatability, the desirability was real. But foraging filled the gaps when hunting was insufficient, and the adaptability of the human body to do both is itself remarkable.

• Alex’s sharpest formulation of the ancestral fallacy: the fact that ancestors did or did not do something carries no inherent health verdict on its own. Ancestors also engaged in violence that we do not consider it healthy or correct to replicate today. The evolutionary fact establishes only that something was not encountered, which can motivate a hypothesis worth investigating with modern science — but it cannot substitute for that investigation. We do not apply ancestral-absence reasoning to cell phones or air travel, and applying it selectively only to diet is inconsistent. Modern science’s actual finding that ultra-processed diets are linked to poor health outcomes is what gives the “we are not adapted to this” explanation its teeth — the evolutionary mismatch framing is a leading explanatory mechanism, but it earned that status through direct modern investigation, not through ancestral absence alone.

• Dave’s parallel point on consistency: by the same logic, modern feedlot beef is also “so far removed from a wild animal” that strict ancestral consistency would require rejecting it too — yet almost no one applies the argument that way. Agriculture gets the same even-handed treatment: agriculture allowed population explosion and permanent settlement (hunter-gatherers could not sustain that), which Alex calls a genuine double-edged sword — it also produced new deficiency diseases from increasingly monotonous diets, with pellagra in the post-colonial American South (from a corn-heavy diet low in niacin) as a concrete historical example that simply would not occur in hunter-gatherer populations eating a small but adequate amount of meat.

Amino acid completeness, the carnivore conference talk, and why supplementation does not invalidate a diet [22:35]

• Dave describes presenting at the first carnivore-focused conference (hosted by Amber O’Hearn), where — somewhat ironically given the audience — his topic was the difficulty of running innovative diet experiments without a protein shake to hit the full amino acid complement, and his anticipation of criticism for using a “processed food” workaround.

• Alex’s position on the supplementation-as-weakness argument: it is not a strong critique. People supplement constantly to achieve health goals across every dietary pattern, and there is no principled reason animal-based diets needing a protein shake is different from plant-based diets needing one. He is explicit that animal proteins are, on the whole, superior to plant-based proteins — particularly whole-food plant proteins — due to digestibility, amino acid availability, and compositional completeness. But isolated plant protein powders close most of that gap because processing removes the anti-nutrients that interfere with digestion and absorption in whole plant foods, concentrating amino acids enough to overcome the limiting-amino-acid problem.

• The dose-dependent crossover point he cites: at 20–30g of protein, whey outperforms soy, pea, or rice protein for stimulating muscle protein synthesis. At 40g and above, they equalise — the process saturates, and lower amino acid quality can be compensated for by consuming more total amino acids (at the cost of more total calories and food volume, which does not work for everyone but is achievable).

• Dave’s own honest motivation for not running a pure whole-food plant-based version of his experiment: less about an ancestral-purity argument and more that he simply does not enjoy eating that way as much, and did not want critics able to say the experiment used a “processed” approximation rather than a true whole-food version. He is explicit that he has metabolically healthy plant-based friends whose blood work looks great, alongside friends on both plant-based and low-carb diets whose blood work suggests the diet label itself is not sufficient — they could be doing better regardless of which side of the diet-tribe line they sit on. That nuance, he says, gets lost constantly in nutrition discourse.

The personal fat threshold and the twin cycle hypothesis [44:04]

• Alex lays out what he considers the most evidence-based explanation currently available for the development of type 2 diabetes, built around Roy Taylor’s personal fat threshold concept.

• The model: subcutaneous fat tissue evolved to expand and contract with feast and famine cycles, storing triglycerides during abundance and releasing free fatty acids during scarcity. This capacity has a limit, set by a combination of environmental/lifestyle factors (physical activity raises the ceiling) and genetic factors (some people have a greater capacity for adipocyte hyperplasia — cells splitting to create new storage capacity rather than just expanding existing cells).

• When that capacity is exceeded, fat cells become lipid-overloaded. Cellular studies show that cramming enough lipid into a fat cell causes it to shut off its own insulin receptors — even without inflammation or hypoxia, simple lipid overload triggers insulin resistance at the cellular level, refusing further energy uptake. In a living human, additional complications compound this: hypoxia in overloaded fat tissue, cell death, an immune response, and inflammation. The resulting insulin-resistant fat cells not only stop storing energy — they begin leaking fatty acids into the bloodstream when they should not be. The Randle cycle (fatty acid and glucose competing for cellular uptake, each suppressing the other’s use) then comes into play.

• The body responds to circulating fatty acid overflow with emergency ectopic storage — fat deposited in tissue not designed to store it: liver, pancreas, even skeletal muscle. This is the basis of Roy Taylor’s twin cycle hypothesis: fat in the liver causes hepatic insulin resistance specifically around glucose regulation, so gluconeogenesis is no longer properly shut off by insulin, producing continuous glucose output regardless of need. This drives the pancreas to secrete progressively more insulin to compensate — the period where fasting insulin rises while fasting glucose has not yet risen. Eventually beta cell function in the pancreas burns out (with some debate over the precise mechanism), insulin secretion declines, and fasting glucose finally rises — the late-stage marker of a process that has been building for years.

• The reversibility evidence: Taylor’s counterpoint, counterbalance, and direct trials, plus a more recent trial in normal-weight individuals with type 2 diabetes, all showed that losing body fat reversed diabetes — and the reversal held even after reintroducing carbohydrates, provided body fat was not regained. The only people who did not benefit were those with diabetes of such long duration that they had sustained genuine pancreatic damage, producing something closer to pseudo-type-1 diabetes where insulin secretion itself is now insufficient regardless of body fat status.

• Alex’s headline conclusion: losing body fat is the single most powerful lever for improving metabolic health because it directly addresses the fundamental drivers of insulin resistance, and it does not matter which diet gets you there — keto, high-carbohydrate, or anything else — because the mechanism runs through excess body fat itself, not through any particular macronutrient ratio.

Can a high-insulin diet still produce fat loss? Dave and Alex work through it live [52:36]

• Dave poses a direct question: does Alex believe a diet can be high-insulin and still succeed at lowering body fat? Alex says yes, with a qualifier — insulin from eating is not elevated at all times, so over a full day it is not necessarily “high” in aggregate.

• Dave narrows it: a high fasting insulin diet, specifically. Alex points to Roy Taylor’s diabetic study participants, who have high fasting insulin and still lose body fat successfully through dieting, with fasting insulin in some cases above 50 µIU/mL in the first counterpoint study (very low-calorie shake-based intervention).

• Dave clarifies he is borrowing a heuristic from Ben Bikman, not making a pro-carbohydrate case specifically: any diet that succeeds for weight loss is ultimately a low-insulin diet over a meaningful window, because insulin is the body’s central anabolic hormone, and a genuine 12-hour fasted period should produce low insulin if fat loss is occurring.

• Alex’s working-through: if insulin is chronically high due to insulin resistance, the signal insulin is able to transmit is blunted regardless of the absolute level, so the high number may not functionally matter the way raw insulin elevation in an insulin-sensitive person would. He raises the open question of whether normal insulin sensitivity with chronically high insulin levels exists as a real-world scenario at all (excluding cases like type 1 diabetics overdosing on exogenous insulin, which he does not consider applicable to normal physiology).

• Dave’s mechanistic case for why insulin can halt fat loss independent of energy balance: essential fatty acid deficiency researchers historically had to feed high-carbohydrate shakes every two hours specifically to keep insulin elevated and trap existing fat stores in adipocytes, because otherwise stored essential fatty acids would slowly leak out and prevent a deficiency from manifesting for a long time. This demonstrates insulin’s power to lock fat in place independent of caloric intake.

• Alex’s pushback on the practical relevance of fasting insulin as an actionable variable: if fasting insulin necessarily drops as a downstream consequence of achieving an energy deficit, then it is not something a person has direct control over — what they control is food choice, and any diet that creates an energy deficit will lower fasting insulin as a byproduct. Dave’s rejoinder: insulin area under the curve is something you do have meaningful control over through diet composition, and you can end up with materially lower net insulin exposure on a low-carb diet even without targeting it directly — while acknowledging overconsumption is still possible on low-carb.

Connecting the dots: VLDL turnover, HDL as a marker, and why LMHR free fatty acids run high even at low triglycerides [59:06]

• Dave walks Alex through the mechanistic chain he considers central to the lipid energy model. The vast majority of triglycerides found in fat cells originated from an apolipoprotein particle (acknowledging de novo lipogenesis exists in fat cells as a minor contributor). When fat cells are full and not accepting more triglyceride cargo, the lipid-rich VLDL particles delivering that cargo “can’t find parking” and circulate longer, raising blood triglycerides and eventually producing ectopic fat as the body searches for somewhere — anywhere — to offload the backlog.

• His explanation for the characteristically very high HDL cholesterol seen in lean mass hyper responders: HDL is a marker of triglyceride turnover, not just a static deposit. If turnover of triglycerides off ApoB-containing particles is extremely rapid — cargo is unloading quickly and successfully into metabolically demanding tissue — that bumps up HDL because the HDL particle population itself rises as a byproduct of that efficient cycling.

• The free fatty acid observation Dave offers from his own blood work: lean mass hyper responders are both metabolically healthy and run at relatively lower fasting insulin, yet have a high quantity of circulating free fatty acids (NEFAs) at any given point in time — including elevated free fatty acids during fasting, which is not the pattern typically described in the literature. His explanation via the Randle cycle: a cell with abundant free fatty acids available is not “dumb” — it down-regulates glucose uptake and spares the glucose for obligate glucose-using cells (red blood cells, certain neurons), because the fat-adapted system has plenty of fuel from the other source.

• Alex’s agreement with the broader implication: high triglycerides with low HDL is a legitimate early warning sign that likely precedes full-blown type 2 diabetes and deserves attention — which is precisely why the inverted LMHR lipid pattern (high HDL, low triglycerides) is the interesting research question. Dave frames the actual research question precisely: does a healthy lipid profile that reflects successful adipocyte processing of triglycerides (rather than adipocyte failure) carry the same association with atherosclerosis as the conventional dyslipidemic pattern that the lipid hypothesis was originally built around?

Statistical adjustments, the triad challenge, and why Alex would update his confidence — but not all the way [1:05:05]

• Alex’s honest answer on whether statistical adjustment for confounders like insulin resistance would satisfy his uncertainty about LDL’s relationship to atherosclerosis in metabolically optimal populations: it would move his confidence from roughly 65% to 80–85%, not to certainty. He immediately anticipates Dave’s objection — adjustments can only get you so far, and pooling heterogeneous people into one group risks averages being driven by the worse-off subset.

• Dave sets the table with the origin story of his “triad challenge,” launched in 2018: a standing, eventually monetary ($300, later $1,000) bounty for anyone who could find a study categorically showing a population with HDL ≥ 50, triglycerides ≤ 100, and high LDL associated with above-average cardiovascular disease risk. Several responses came back citing studies with matching group averages rather than individually-screened participants meeting all three cut points simultaneously — precisely the averaging problem Alex flagged. This is also why every LMHR study participant individually meets the eligibility criteria rather than the cohort simply averaging into range — it eliminates the possibility that a subset of outliers is driving the group result.

• Dave’s broader critique of sensitivity analyses and adjustment culture in nutrition epidemiology: adjustments are, definitionally, educated guesses about what the data should look like rather than what was actually collected. He is careful to frame this as a confidence-calibration problem, not a binary dismissal — he explicitly rejects the “epidemiology is all trash” position some people try to attach to his views, and affirms observational data’s real value for hypothesis generation. His one clean exception: observational data is genuinely powerful at knocking down strong causal claims, even if it cannot establish new ones — his thought experiment of an entire country of three-pack-a-day smokers who are disproportionately centenarians would not nullify a causal hypothesis about smoking, but would necessitate updating it.

Red meat and colon cancer as a calibration case for confidence in small effect sizes [1:16:34]

• Dave uses red meat and colon cancer as a worked example of his confidence framework. Alex states the causal claim is, in his view, likely true — but immediately distinguishes causality from inevitability, the same way smokers exist who never develop lung cancer. He is candid that he has not researched the red meat-colon cancer mechanism in depth, but understands it runs through colonic putrefaction of excess amino acids by gut bacteria producing toxic metabolites that damage intestinal cells.

• Dave’s test case: does Alex expect lean mass hyper responders — who eat copious amounts of red meat — to show elevated colon cancer rates? Alex says no, and offers a mechanistic reason this could be blunted: someone on a ketogenic diet eating a lot of meat may have an altered, dysbiotic-by-conventional-standards gut microbiome, but elevated blood ketones could feed colonic epithelial cells with beta-hydroxybutyrate in a way that offsets reduced butyrate production from fibre fermentation — a counteracting pathway that could blunt the causal mechanism even if the underlying causal link is real.

• The effect size acknowledgment that anchors Dave’s broader confidence calibration point: the actual reported effect for red meat and colon cancer is extremely small (roughly 0.3% to 0.4% of cancers, by Alex’s recollection, derived from observational data). Dave’s point in raising this: investing high confidence in causal directionality from an effect size this small, derived through statistical adjustment of observational data, requires trusting that the adjustment math was done correctly — a very different epistemic situation from a case with overwhelming effect size.

• The contrast case Dave draws to explain why the lipid hypothesis is so much more entrenched: homozygous familial hypercholesterolemia (HoFH). A child with HoFH developing xanthomas at age three, angina at age three, a first MI a few years later, with LDL in the 700s, and — critically — none of the conventional confounders (no type 2 diabetes, not a “Type A” personality, not hypertensive, not a smoker). Dr. Goldstein’s observation about this population is, in Dave’s view, completely understandable as the foundation for assuming an independent causal relationship. The open question Dave keeps returning to is not whether ApoB particles are part of the causal pathway — he agrees they are — but how strongly concentration alone drives the process versus acting as an accelerant on top of an existing injury or inflammatory state.

The stimulation hypothesis vs. the multiplier hypothesis, and the clone thought experiment [1:28:23]

• Dave names two positions explicitly for the conversation: the stimulation hypothesis (ApoB concentration itself is an injurious stimulus that drives atherosclerosis) versus what he calls — his own informal term — the multiplier hypothesis (ApoB particles function more like fuel/wood for a fire that some other process ignites: a site of inflammation that is then accelerated by the wide availability of circulating ApoB particles as raw material). He finds the multiplier framing more plausible but does not consider the question settled.

• The clone thought experiment Dave poses to test Alex’s actual position: if you and a clone of yourself diverge, with the clone going on lipid-lowering therapy, would Dave expect the non-treated version to develop earlier plaque? Alex agrees this would be fair — but immediately complicates it with a practically important point: what matters is the starting LDL and the magnitude of change, and real-world interventions never change just one variable. Diet, lifestyle, and drug changes shift a whole cluster of risk factors simultaneously, so the net change in overall cardiovascular risk from “lowering LDL” in practice reflects the sum of everything that changed, not LDL in isolation.

• Alex’s synthesis position, stated plainly: in the general population, he believes LDL exists on a fairly linear risk gradient because the general population is, on the whole, unhealthy — but he expects genuine exceptions to exist, and lean mass hyperresponders may well be one. He frames this as something the field needs to investigate at the level of mechanism rather than assume away in either direction. His call to action: the current paradigm is built substantially on observations made in unhealthy populations, with mechanisms then investigated in animal models that are rarely designed to replicate the experimental diets that produce these elevated lipid phenotypes in the first place.

Animal model limitations: rabbits, species-appropriate chow, and the documentary’s efficacy framing [1:33:34]

• Alex raises a structural problem with translating animal atherosclerosis models to humans that goes beyond diet composition: even when researchers attempt to replicate a macronutrient shift, the animal’s physiology was never adapted to that shift in the way humans are broadly adaptable across dietary patterns. Dave volunteers a concession he believes the low-carb community should make without defensiveness: the classic rabbit cholesterol-feeding studies that originally established the diet-cholesterol hypothesis used dietary cholesterol loads inappropriate to rabbit physiology, and the failure to replicate comparable plaque development in dogs and rodents shortly afterward was a legitimate and important piece of evidence the field should not minimise.

• His broader concern about animal model confidence: most rodent atherosclerosis models rely on specific obesity-prone lineages and highly engineered chow, which compounds the species-appropriateness problem on top of the diet-composition problem. His consistent calibration point throughout the conversation: these models are useful for hypothesis generation, but the confidence problem arises specifically when causal claims derived from them are used to foreclose treatment options for people experiencing real clinical benefit.

• Dave connects this directly to the Cholesterol Code documentary’s emphasis on patients who adopted a ketogenic diet for efficacy reasons — bipolar disorder, type 1 diabetes, and other conditions — who then became lean mass hyperresponders as a side effect. His point: if a medical team responds to a successful efficacy-driven intervention with “your LDL is now too high, so this might give you a heart attack in some years,” that response needs to weigh the documented benefit being walked away from, not just the isolated lipid number. Alex extends this directly to carnivore-diet remission cases in severe autoimmune disease, offering his own working hypothesis that plant antigens may be the actual mechanism driving symptom resolution in some of these cases rather than meat itself — meaning a full carnivore elimination may be doing the job of a more surgical removal of specific antigens, just via a blunter and easier-to-execute instrument.

Why diet evangelism becomes ideological, and the steak test for overconsumption [1:36:53]

• Both push back on a pattern they have each noticed: when someone leaves Diet X for Diet Y and finds success, defenders of Diet X frequently respond with “you just did not do it correctly” rather than accepting the person found something that worked for them. Dave’s explicit position: it should not matter what diet someone leaves or arrives at — their success in the new pattern is the relevant data point, full stop, even if they continue adjusting it over time. Paul Saladino’s public evolution — adding fruit and honey back into a strict carnivore approach after finding it lowered his testosterone and left him chronically fatigued — is cited as a healthy example of exactly this kind of non-ideological adjustment.

• Alex’s extension: genetic and physiological variation (such as polymorphisms affecting beta-carotene-to-vitamin-A conversion efficiency, which would materially disadvantage someone attempting a strict vegan diet relative to someone with efficient conversion) means there is no principled reason to expect uniform human response to any single dietary pattern — yet dietary discourse rarely accommodates this and instead treats deviation as a discipline or execution failure.

• Dave adds a second, under-discussed dimension beyond physiology: environment, family life, and culture shape what is realistic to adopt independent of a diet’s nutritional composition. His own concrete example: certain technically-keto “problem foods” (specific fat bombs) cannot be kept in his house without overconsumption, while keto chow shakes mixed with heavy whipping cream — about as processed as it gets — do not trigger the same overconsumption pattern for him. The lesson: getting the macros “right” on paper is necessary but not sufficient; food-specific behavioral responses matter independently.

• The steak test: Dave has yet to find anyone who can reliably overconsume calories on steak and eggs alone (provided the eggs are not turned into an easily overeaten preparation). His working explanation leaves room for either appetite modulation or simple monotony fatigue — he is careful to flag the second as a fair alternative explanation a good scientist should not dismiss. Alex independently corroborates from the literature: protein has a documented, under investigated special status in overfeeding research. Jose Antonio’s studies added 800 extra calories per day purely from whey protein shakes to resistance-trained subjects’ normal diets for eight weeks with no DEXA-confirmed body fat gain, and additional research shows higher protein intake actively reduces hepatic lipogenesis — protein overfeeding does not engage the same fat-synthesis machinery that carbohydrate or fat overfeeding does.

Is obesity a disease? The addiction parallel and the Vietnam veteran heroin data [1:48:46]

• Alex challenges the now-common medical classification of obesity as a disease using an extended analogy to addiction. Addiction was long treated as a fixed brain disease, but Lee Robins’s research on Vietnam veterans found that of those who became addicted to heroin in Vietnam, roughly 70% simply stopped upon returning home — despite continued access — and a further 20–27% went on to use moderately without meeting addiction criteria again. The most common pathway out of addiction generally is spontaneous remission, not lifelong management. Modern neuroimaging supports a plasticity-based account: chronic heavy drinking visibly shrinks and damages the brain, and that damage substantially reverses after roughly 12 months of abstinence — evidence of adaptive, reversible change rather than fixed disease pathology.

• Dave’s steelman challenge to the analogy: the Vietnam homecoming includes a confounding “fresh start” effect and a cultural shaming dynamic (heroin not being socially acceptable at home) that could independently explain rapid cessation — paralleling smoking cessation patterns once smoking lost social acceptability. Alex’s response sharpens rather than concedes the point: if obesity (or addiction) were a true brain disease, environmental and social factors should not be sufficient to produce remission, the way no amount of social pressure cures cancer or multiple sclerosis. The fact that behavioral and environmental change reliably produces remission in both conditions is itself evidence against the strict disease framing and in favor of a behavioral-habit account with adaptive (not pathological) neuroplastic underpinnings.

• Dave’s synthesis, drawing on real cases in his own extended family: he resists the binary between “it is 100% choice, white-knuckle through it” and “it is entirely outside personal control” framings he sees as dominant in the fitness and medical spaces respectively. Some family members find a given diet (low-carb, in several of his examples) genuinely resolves their hunger regulation with comparatively little ongoing struggle; others are what he calls “challenge cases” who remain hyperinsulinemia and constantly hungry despite trying the same interventions, requiring careful behavioral forensics rather than a single dietary fix.

GLP-1 agonists as a behaviour-change bridge, not a permanent offramp [2:01:10]

• Both agree GLP-1 agonists are, by a wide margin, the most successful weight-loss medication class encountered to date — and both are uneasy about the emerging cultural narrative that they should be framed as a lifelong medication comparable to a hypertension drug, taken indefinitely without an accompanying behavior-change goal.

• Alex’s reframe: these drugs act directly in the brain to induce satiety and reduce food noise, which makes them a tool for making behavior change easier — not a substitute for it. He draws a direct parallel to emerging GLP-1 trials in alcohol use disorder, where the drugs reduce drinking by blunting the pleasure response to alcohol; the open ethical question in both domains is whether you tell someone “now drink/eat as much as you can tolerate since the pleasure signal is blunted,” or “use this reduced food noise window to build new sustainable habits before potentially coming off the drug.”

• Dave’s pattern-recognition caution, drawing on the history of “miracle drug” cycles he has lived through (metformin’s early-2010s framing as a candidate longevity drug being the most recent prior example): as of this recording, no chemical compound has been discovered that functions as a clean, net longevity drug for the general population, and most interventions that do show population benefit turn out to be correcting an existing deficiency in that specific population rather than conferring a novel benefit. His framing of GLP-1 agonists: a rational, useful response to an obesogenic environment humans were not designed for — valuable, especially for genuine “challenge cases” in his own family who are unwilling to attempt anything else — but not an end-all solution that obviates behavioral work.

Biomarkers vs. performance: the medical system’s disease-only blind spot [2:08:46]

• Dave shares a generational story from his grandfather, who contrasted his own physician (a same-age classmate who made house calls, was in excellent health himself, and constantly lectured him about going outside, getting sunlight, and walking more) with a newer doctor focused on managing biomarkers and prescriptions — captured in his grandfather’s line that the new doctor “just seems to want to fill my pillbox.” Dave is careful to credit pharmaceutical innovation broadly and acknowledge its genuine life-saving record, while still asking the practical question almost never raised in a prescribing conversation: has this specific combination of medications, layered on top of each other over years, actually been studied together, or is the patient simply trusting an absence of known interaction?

• Alex’s diagnosis of the structural problem: modern medicine is overwhelmingly oriented around disease treatment rather than health building, with “all biomarkers in normal range” functioning as the de facto definition of healthy — even though biomarkers say nothing about fall risk in older age (a leading cause of premature mortality and disability in the elderly), daily functional capacity, energy levels, hunger regulation, or cognitive clarity. He distinguishes the “disease state” axis (Alzheimer’s and dementia as diagnosable conditions) from an entirely separate “positive health” axis (a brain that is merely disease-free is not necessarily performing anywhere near its potential), and argues the second axis is almost entirely unaddressed by conventional and even much of functional medicine, which remains biomarker-and-disease-centric.

• This is the explicit foundation of the human optimization program Alex is developing with Ari Whitten of The Energy Blueprint: building health across body systems directly — strength, cardiorespiratory fitness, detoxification capacity, heat and cold tolerance — as interventions valuable in their own right, independent of whether a specific biomarker is currently flagged as abnormal.

Why people die without a specific disease, and what hunter-gatherer longevity actually protects [2:19:34]

• Alex cites research estimating gains in average life expectancy if specific diseases were eliminated entirely: curing heart disease (the leading US killer) outright would add roughly three and a half years to average lifespan; curing cancer would bring the total to about five years; curing the full top-10 list of killers would add roughly ten years in total. His point in raising this: ten years is meaningful but far short of what biohacking culture often implies, because many elderly people do not die of any single named disease — their body simply breaks down at a systems level.

• He illustrates with his own grandfather, who lived to 101, survived the Great Depression and World War II, maintained a large garden until 97, was still driving a truck at 98, stayed continuously physically and socially active and purposeful — and then declined rapidly within roughly a week after a sudden intestinal issue required minor surgery, with his heart simply stopping shortly after. He never did dedicated cardiovascular training. Alex’s point: there is no biomarker panel that would have flagged or prevented this kind of systemic structural breakdown — what builds resilience against it is the kind of broad physical and cognitive challenge his grandfather maintained simply by living an active, purposeful, socially engaged life, which functions as a buffer against age-related degeneration that biomarker-centric medicine does not currently measure or target.

• He extends the same logic to cognitive health: elderly indigenous hunter-gatherers show notably low rates of dementia and cognitive decline, which he attributes to remaining continuously embedded in problem-solving, social interaction, and knowledge transmission to younger generations throughout old age — ongoing cognitive and social challenge functioning as ongoing stimulus for neuroplastic maintenance, in the same way resistance training provides a physical stimulus for muscle maintenance.

• Dave’s synthesis connecting this back to LMHR: he agrees lean mass hyper responders likely do carry genuinely lower cardiovascular risk than their LDL number alone would suggest — but flags an important irony in why. The lipid triad itself (high LDL, high HDL, low triglycerides) functions almost like an automatic screen for an athletic, active population: it is observationally, not definitionally, associated with leanness. The LMHR Facebook group banner is literally a collage of member photos that “looks like a gym.” Since at least 2017, Dave could often correctly guess from lipid numbers alone whether someone was lean and whether they did substantial cardio, purely from the pattern — which raises the legitimate possibility (a collider problem in statistical terms) that the lipid profile is, in some populations, more a marker of an active, metabolically resilient lifestyle than an independent causal variable in its own right.

Dayspring’s own admission, and the acute phase reactant problem [2:27:30]

• Alex raises what he considers the strongest evidence for LDL causality in the general population: the convergence of randomized controlled trials, epidemiology, mechanistic research, and Mendelian randomization — acknowledging individual RCTs show meaningfully different effect sizes, which he attributes to drugs affecting more than lipids alone (inflammation reduction alongside LDL lowering, for instance) given heart disease’s multifactorial nature.

• Dave presents what he considers a direct logical tension in that position: he has heard Thomas Dayspring state on multiple occasions, across podcasts and Twitter, that it does not matter how ApoB and LDL cholesterol are lowered — the benefit accrues regardless of mechanism. Dave’s argument: if the method of lowering genuinely does not matter, that supports independent causality. But if it does matter — if a statin’s simultaneous anti-inflammatory effect is doing meaningful independent work — then the hypothesis needs updating to acknowledge LDL is not cleanly separable from the other things changing alongside it. Holding both “it doesn’t matter how you lower it” and “the co-occurring inflammation reduction is doing real independent work” simultaneously is, in Dave’s framing, not coherent.

• Dave’s mechanistic capstone: ApoB-containing lipoproteins are acute phase reactants — proteins that rise during inflammatory states, not universally but commonly. This makes disentangling LDL’s independent effect from background inflammation genuinely difficult using the existing literature, because the very biomarker in question moves with the confounder you are trying to separate it from. Alex partially counters with a study (recalled, not cited in detail) showing similar statin-driven cardiovascular risk reduction per unit of LDL lowering in both high- and low-baseline-inflammation subgroups — to which Dave raises a chicken-and-egg objection: arterial wall damage itself can be the trigger drawing out a localized inflammatory and inflammasome response, with LDL particles aggregating preferentially at sites of inflammation by design as part of the immune playbook, which would not necessarily show up cleanly as a baseline systemic CRP-style inflammation split.

Transcytosis, endocytosis in children, and why the math on HoFH exposure gets interesting [2:33:55]

• Dave brings in the 2020 EAS mechanisms paper (sequel to the widely circulated 2017 consensus statement) for its explicit acknowledgment of transcytosis — the active, vesicle-mediated transport of lipoproteins across an intact endothelial cell, distinct from passive movement through junction gaps — and its finding that transcytosis rates increase measurably with inflammation and CRP exposure. Dave’s point: this is an important clue, because it means the rate of subendothelial LDL delivery is itself inflammation-modulated, which is precisely why ruling out systemic endothelial inflammation matters for interpreting a high-LDL, low-inflammation population.

• Dave introduces endocytosis (cellular uptake, distinct from transcytosis’s pass-through transport) as relevant to a respectful disagreement he has with conventional readings of childhood FH data: he hypothesizes endocytosis happens at a much higher rate in children specifically because of active growth, meaning children’s naturally low baseline LDL particle count partly reflects more aggressive uptake for tissue construction and repair — not necessarily a “safe” baseline level that adults should be held to. If chronically elevated childhood LDL creates more opportunities for this uptake process to go wrong, Dave argues the resulting exposure-to-plaque relationship in children should show an especially strong, clean log-linear association — essentially a more controlled natural experiment than adult observational data typically offers, given children are closer to a single dominant uptake pathway with less confounding lifestyle variation.

• Dave shares specific numbers from his own research into six published homozygous FH CTA cases: the two patients with no detectable plaque had a calculated total lifetime LDL cholesterol exposure of roughly 1,400 mg/dL-years or less; the lowest total exposure among the remaining four patients who did show some plaque was approximately 1,800 mg/dL-years. By contrast, the LMHR cohort’s total exposure in the Miami Heart match analysis was in the range of 6,500–7,000 mg/dL-years — vastly higher in absolute terms — yet the total exposure delta between the LMHR cohort and the matched Miami Heart control group was only around 700 mg/dL-years, and showed no detectable internal association with plaque presentation within either group. Dave is explicit this remains “horseshoes and hand grenades” math rather than a finished analysis, but considers the absence of any internal dose-response signal at these exposure levels, within a cohort followed for nearly five years, a genuinely strong piece of hypothesis-generating evidence regardless of which side of the debate one starts from.

Closing: the LDL-causality tweet dissected, why discussion beats debate, and the shifting Overton window [3:05:35]

• Dave walks through the specific tweet from Alex that prompted this invitation — a widely shared post citing the 2017 EAS consensus paper’s three-lines-of-evidence graph (RCTs, epidemiology, Mendelian randomization) as addressing anyone who claims “LDL doesn’t cause heart disease.” Dave’s line-by-line dissection: most viewers, without checking, assume the graph’s axes are simply “LDL” against “heart disease,” when the actual labels are “magnitude of exposure to LDL cholesterol lowering” against “proportional reduction in risk of CHD” — a specific, narrower claim dressed in the visual shorthand of the single most persuasive shape in data visualization: a clean bottom-left-to-top-right regression line. His broader point about causal claims generally: stating “A causes B” is actually several simultaneous claims — that B does not cause A (or that the reverse pathway has been quantified and accounted for), and that every other variable (C, D, E, F) that could independently affect both A and B has also been identified and accounted for. Each of those embedded claims is itself a causal claim requiring justification, not a free assumption.

• Alex’s direct response: he concedes the framing point as fair, noting individual RCTs in that very graph show wide dispersion in benefit despite similar LDL reductions — consistent with interventions doing more than just lowering LDL. Dave extends the critique with a time-scale argument: the average RCT runs three to five years; the LMHR match-analysis cohort had already accumulated 4.7 years of a far larger LDL change in the opposite direction (upward, rather than the downward changes RCTs measure) by the time of their first scan, with no internal plaque association detected — if a smaller downward LDL change over a comparable or shorter window is sufficient to register a detectable mortality benefit in RCTs, Dave argues symmetry would predict the larger upward LMHR change should register a detectable downside, and it has not, so far, within the data collected to date.

• Both explicitly credit the format itself — a long, in-person, unhurried conversation rather than a public social media exchange — for surfacing how much common ground exists that a tribal Twitter exchange would never reveal. Alex prefers the word “discussion” to “debate,” reserving “debate” for an exchange organized around winning rather than mutual understanding; both agree the format that matters most is one where saying “I don’t know” is safe and ideas can be challenged without it being read as a tribal attack.

• Alex’s closing epistemic stance, offered without resolving the disagreement: the absence of evidence on a specific subpopulation is not evidence of absence, and inferences drawn from population averages cannot be validly extended to claim a specific individual or phenotype carries a specific risk level until that phenotype has actually been studied — which is precisely the gap the LMHR research program exists to close. He flags one explicit area of continued caution regardless of metabolic health: unknown variables (an undetected genetic polymorphism, a chronic subclinical viral infection causing endothelial damage) could exist in any population and would not show up until specifically tested for, which is part of his case for building out a broader registry rather than generalizing from any single cohort.

• Dave’s closing reflection on the larger pattern across this stretch of episodes: ten years ago, announcing a cohort with average LDL of 272 mg/dL would have been read by nearly everyone as self-evidently describing a population in poor metabolic shape. The fact that even some of his sharper detractors have shifted toward “they’re probably not at dramatically elevated risk, but they would likely be better optimized at a lower LDL” represents a real, measurable shift in the Overton window around this question — a more defensible position than the one commonly held five years prior, even though it remains short of where Dave’s own research points.

Share

Leave a comment

Who is Nick Verhoeven, and how did Physionic happen? [1:03]

• Nick Verhoeven holds a PhD in molecular medicine, completed about six or seven months before this recording. He runs Physionic, a science communication channel he describes as his passion project and career: reading research, analysing it, translating it for the public.

• The Physionic origin story is one of the more striking in the science communication space. For eight years he released a piece of content every week without missing a single week. At the end of those eight years he had 20,000 subscribers and was making no money. Then the algorithm picked things up. Over the following year and a half the channel blew up, AdSense revenue got to the point where he could barely subsist on it, and he started figuring out how to build an ethical business that did not make him beholden to any company or sponsor.

• He has since rejected thousands of sponsorship offers. His reasoning: it is not just about whether having a sponsor actually biases you. It is about eliminating any possibility that a viewer could believe you are biased. He also made a deliberate decision about branding: calling the channel Physionic rather than anything diet-specific, because the moment you brand yourself as keto-this or vegan-that, you have bracketed what you can discuss and sent a strong signal about where your motivated reasoning probably runs.

• Dave shares his own version of this: his original Twitter handle was Dave Keto, an indexing decision a computer programmer would make. When his research grew beyond that label, he changed it to Real Dave Feldman, losing followers he did not care about. He makes the same point Nick does: there is no diet that is the standalone best diet for everyone, and he does not want to be read as an advocate for any single one.

Subscribe now

Seed oils: how Nick approaches a contested topic, and why short-term studies do not impress him [10:21]

• Dave raises seed oils as a hot topic. Nick’s answer models his general epistemological method: you cannot ask “are seed oils good or bad?” You have to ask outcome-specific questions. Are seed oils inflammatory? That is one question. Is canola oil associated with cardiovascular disease risk? That is a different question. He works through outcomes one at a time and acknowledges he is nowhere near done.

• Dave’s first objection to the standard framing: he does not like the claim that a particular food “causes inflammation.” Inflammation is the body’s natural response — something you would die without. What people usually mean by “inflammatory food” is chronic inflammation. And very few things are independently pro-inflammatory in the way the claim implies. The more precise question is whether a food causes some injurious stimulus that then triggers an inflammatory response as an intermediary. That distinction matters because it explains why short-term studies showing no rise in CRP are not informative: if the food is causing slow ongoing injury, you would not detect the inflammatory response in a two-week trial. The injury needs time to accumulate.

• Nick’s methodological framework for adjudicating any contested food: start with clinical outcome data. Do we see cardiovascular events, death, or other hard endpoints tracking with this food? Then backtrack through epidemiology, RCTs, and mechanism research and try to piece it together. What he sees too often is the reverse: someone finds one negative mechanism, declares the food inflammatory or dangerous, and tries to build a narrative around that single mechanism while ignoring outcome data that does not support the conclusion.

• The lipoprotein lipase example Dave uses to illustrate the mechanism problem: LPL breaks down lipoproteins and delivers their lipid cargo to cells. Cells make LPL unless something inhibits them from making it. Why would you design the system to have the inhibitor rather than just making LPL on demand? Because inhibitors to positive-feedback processes are all over biology. A lot of disease comes from missing the inhibitor to something that would otherwise run unchecked. The key example in lipid metabolism: the ANGPTL proteins, particularly ANGPTL3 and related complexes, are LPL inhibitors that enable cross-tissue communication. The liver can signal to muscle tissue to upregulate LPL activity during states where fat oxidation is appropriate. Discovering LPL without discovering the ANGPTLs would give you a wildly incomplete picture. This is Dave’s cautionary tale against extrapolating from a single newly discovered mechanism to a grand theory, especially when the mechanism comes with a product pitch.

The LMHR study design explained, and Nick V’s qualms [24:50]

• Dave walks through the Lundquist study in detail for Nick’s benefit and for listeners who may be encountering it for the first time.

• Eligibility criteria: LDL of 190 mg/dL or higher, HDL of 80 mg/dL or higher, triglycerides of 80 mg/dL or lower. Participants had to prove the hyperresponse — providing blood work showing their LDL was under 160 before starting a ketogenic diet and had risen by at least 50% or more. They had to have been on a ketogenic diet for approximately two years. Any prior diagnosis of cardiovascular disease, type 2 diabetes, or hypertension was an exclusion criterion.

• 100 participants were recruited and flown to the Lundquist Institute, where they got a CCTA (coronary CT angiography) — a high-resolution heart scan that Dave is, as of this recording, more bullish on than he has ever been in his life. One year later they return for a second scan. That longitudinal comparison — first scan to second scan — is the primary endpoint, currently under submission.

• The Miami Heart match analysis, which was announced by Dr. Matthew Budoff around a year and three months before this recording: Miami Heart is a separate study of approximately 2,400–2,500 participants, cross-sectional (one scan at baseline). Dr. Budoff had access to both datasets. Of the LMHR cohort’s 100 participants, 80 fell within Miami Heart’s age range. A statistician at Lundquist constructed a matched control group from Miami Heart’s pool — matched for age, ethnicity, gender, CRP, blood pressure, and A1C. The resulting groups were extremely closely matched.

• The result: the LMHR cohort had an average LDL of 272 mg/dL. The matched Miami Heart controls had an average LDL of 123 mg/dL — more than double. The LMHR group had been on a ketogenic diet for an average of 4.7 years. And at baseline scan, there was statistically no difference in plaque between the two groups, with the LMHR cohort actually trending slightly better. Dave is always careful to note: trending toward better is not the same as doing better — the difference is not statistically significant. The reason he mentions the trend is that if it were trending toward worse, he suspects critics would mention it even without statistical significance.

• Nick V’s main qualm: he accepts the match analysis as a good move that adds useful information. His central question is what happens over time. Does atherosclerosis progress in this cohort? Does it progress at a slower rate than expected? Does it not progress at all? One year may not be sufficient to detect a difference given the timescales involved, even with LDL levels more than double the control group. He is genuinely open to the possibility of slower progression or even no progression in metabolically healthy lean individuals with high LDL — which would challenge the lipid hypothesis substantially — but says the data to confirm or deny that do not yet exist.

Dave’s prediction for the longitudinal data, and LDL’s internal correlation to plaque [34:20]

• Dave states plainly: he expects a population-average increase in plaque with the LMHR cohort. Not because he expects a bad result, but because the average age of the cohort is 55. Find any data on 55-year-olds at a population level and there will be a population-average increase in plaque. Even people with PCSK9 loss-of-function mutations — who theoretically should have minimal cardiovascular disease — still develop plaque, just less of it. Age is a factor independent of LDL.

• What Dave considers the more interesting internal validity question: within the cohort, does LDL level correlate to plaque score? At baseline, across both the LMHR cohort and the matched controls, LDL did not correlate with total plaque score at all. In a cohort where LDLs are in the top 1% of the population, that absence of correlation is not a trivial finding.

• Nick V’s revised position relative to five years ago: he would have put a cohort with average LDL of 272 sustained for 4.7 years in roughly the same cardiovascular risk quintile as people with insulin resistance and high blood pressure — because lipidology treats ApoB/LDL as the primary or only variable worth considering. His current position: he puts LMHR in the second-lowest quintile, just above the lowest-risk group. That is, he says, already a major revision. And he acknowledges that five years ago he would not have said that.

The statin data Nick V has been reading, and why Dave finds it difficult territory [57:34]

• Nick V raises something he has been working through recently: he has been reading statin trials more carefully, and he found independently-funded studies that did address populations closer to the LMHR profile — lean individuals with normal BMI (around 23–24), no hypertension, and no type 2 diabetes. In those populations, statin therapy still showed significant effects in preventing cardiovascular mortality, total mortality, and cardiovascular events.

• His concern: people who hear the LMHR data and conclude that because they are lean and their blood pressure is fine, high LDL is simply not a risk factor for them may be making an unjustified leap. The LMHR phenotype is specific — high LDL, high HDL, low triglycerides, confirmed hyperresponse to keto. Someone who is just lean and normotensive but does not have the full LMHR profile is not necessarily in the same situation.

• Dave’s difficulty with this territory: he treats the lipid hypothesis itself as the horse and what you do about your LDL as the cart. He is more careful talking about treatment than about mechanism or risk, partly because he does not want his work to be read as telling people what drugs to take or avoid. But he raises a legitimate mechanistic complication with the statin literature: drugs do multiple things. There is no statin that acts as a magic wand that only lowers LDL. He is curious about what changes in metabolic health accompany statin use. There is modest evidence of a GLP-1 connection with statins — probably tiny, but his immediate thought is: if statins cause even a tiny bit of weight loss or nausea-driven caloric reduction, that is a metabolic confounder in studies trying to isolate the LDL effect.

• The 4S reanalysis he finds most relevant, which David Diamond cites often: the only stratification of statin data by metabolic health profile he is aware of in a major trial. People with atherogenic dyslipidemia (low HDL, high triglycerides) vs. people with HDL above 40 and triglycerides below 150 — still nowhere near the LMHR profile, but more metabolically healthy relatively speaking. The metabolically healthier group had very little benefit from the statin intervention in that reanalysis. Dave raised this with Ethan Weiss five years earlier and asked: why has no one done more of these stratification analyses in the decades since? Weiss’s answer, in a rare candid moment: they’re not really in the business of shrinking their demographic.

• Dave’s broader point on the statin data ecosystem: he is an agnostic, not a denier. The more opaque the data, the less he trusts it. The CTT (Cholesterol Treatment Trialists) collaboration holds some of the most relevant stratified data and it is essentially a walled garden — independent researchers cannot access it for reanalysis. He genuinely believes most doctors want to help their patients. He does not attribute malice. But the absence of incentive to find the subgroups who are not benefiting, or who might be harmed, is a structural problem with the system. It is not a conspiracy; it is just that there is no good business reason to find those people.

The concentration gradient problem: why Dave does not find the passive diffusion model convincing [1:11:29]

• Dave raises again the mechanistic question that has troubled him for years: the concentration gradient version of the lipid hypothesis implies that more LDL particles in circulation means more pressure to push them through the endothelial wall and into the subintimal space where plaque forms. He has never found this convincing, and the emergence of transcytosis as the dominant mechanism makes it harder to defend.

• First, the physical scale argument. Total LDL particle volume in five litres of human blood is roughly one-thousandth of the total volume — less than 0.1%. A single LDL particle is 22.5 nanometers. Compared to the surface area of endothelial cells, LDL particles are tiny. Given laminar flow in arteries, their movement is largely out of their own control. The concentration gradient is, physically speaking, extremely small.

• Second, the transcytosis problem. Until five or six years ago, mainstream lipidology assumed LDL entered the subintimal space predominantly through passive diffusion via junction gaps. That view has shifted: even people like Thomas Dayspring now think transcytosis — active, ATP-requiring, chaperoned vesicle transport — is the predominant mechanism. If that is true, there has to be a stimulatory signal. Something has to tell the endothelial cell to upregulate the transcytosis machinery. Dave has not found evidence that LDL concentration itself constitutes that signal. He notes that mRNA for the relevant receptors and chaperone proteins should be detectable and measurable — and that this is a testable prediction: biopsy lean mass hyperresponders, look at their endothelial cells, see if there is more mRNA for transcytosis-related receptors. If the concentration gradient drives transcytosis, you should see it.

• Nick V’s contribution: he steelmans the other side on junction gaps. He agrees that turbulent flow at bifurcation points — where arteries branch — increases endothelial cell turnover, creates wider gap junctions transiently, and allows LDL particles to accumulate in the area through circular eddy currents rather than being swept straight through. The European Atherosclerosis Society mechanisms paper (2020) describes how LDL can linger near bifurcation points and increase potential for subendothelial entry even without active transcytosis. This gives a plausible physical account of why plaques form preferentially at bifurcations rather than in straight arterial segments.

• Nick V also brings in the smooth muscle cell migration and calcification sequence: smooth muscle cells from the medial layer migrate toward the plaque, differentiate — essentially forgetting they are smooth muscle cells — into immune-type cells and fibroblasts, begin laying down collagen, and eventually deposit calcium. The fibrous cap that results stabilises the vulnerable lesion. This is partly an immune process, and some of the smooth muscle cells that differentiate begin to express immune markers.

• The 2020 EAS mechanisms paper, which Dave references: inflammation induces greater transcytosis. CRP exposure increases transcytosis rates. Other inflammasome signals similarly upregulate the active transport of LDL particles into the subintimal space. Dave’s read: this is consistent with his hypothesis that apolipoprotein-containing particles function as part of the innate immune response. Cells under inflammatory stress signal “come find me,” attracting immune cells and upregulating transcytosis as part of the repair playbook. The transcytosis of LDL is, in this view, by design — part of how the body marshals resources to a site of injury.

Dave’s immune hypothesis for apolipoprotein function [1:28:11]

• Dave lays out his broader mechanistic hypothesis, with the explicit caveat that it is a hypothesis:

• Apolipoprotein-containing particles may function as non-nucleated immune cells. They bind to pathogens. They carry alpha-tocopherol and other antioxidants that protrude from the monolayer, positioning them to bind reactive oxygen species at sites of oxidative stress. Macrophages, which are professional scavengers and repair cells, may use them because they are easier to collect than the smaller pathogens to which the LDL particles have already bound.

• A second function: structural demand. Monocytes differentiating into macrophages need to expand their membrane bilayer. They may soak up LDL particles for this reason — using them as raw material for structural expansion into the lysosomal compartment and cellular membrane. Dave has previously discussed this in the context of myocytes (muscle cells) and adipocytes (fat cells), which are highly elastic and need to grow to accommodate their lipid cargo when someone is fat-adapted and running a continuous flux of triglycerides through them.

• The control component of his hypothesis: atherosclerosis may represent a stage of hemostasis — not homeostasis, but blood stasis, the process of controlling a vascular breach. Primary hemostasis is the platelet plug, simple and transient. Secondary hemostasis is the clotting cascade. Somewhere between secondary hemostasis and full atheroma is a controlled, resolvable plaque state where macrophages manage a site of vascular injury, accumulate LDL particles as part of the innate immune response, and mostly resolve the situation if the damage is not too severe. His belief: macrophages are pros at this resolution, and we are going to find more evidence in time of plaque mobility and macrophage-driven clearance — processes the current literature tends to underemphasise.

• Nick V’s response: the proponents of the lipid heart hypothesis would say the whole process is pathology. Dave’s position is more nuanced: the process starts as physiological response to injury and switches to pathology when the injury is too severe or too sustained for the innate immune system to manage. This is consistent with the response-to-injury hypothesis, which Dave considers well established. His argument is not that the response-to-retention hypothesis is wrong, but that it is a confounder that is extremely hard to eliminate given how difficult it is to produce atherosclerosis in animals without species-inappropriate diets, transgenic modifications, or direct mechanical injury to the vessel.

• The key implication for LMHRs: they are metabolically healthy, not hyperinflamed, and not on a species-inappropriate diet. They happen to have very high LDL. If atherosclerosis in this population proceeds at a meaningfully different rate than in metabolically compromised populations with similar LDL, that would help separate the response-to-injury and response-to-retention hypotheses in a way that is currently impossible to do with available data.

Statistical adjustments, Bradford Hill, and why Dave is not a correlation-is-not-causation absolutist [1:41:27]

• Nick V picks up a thread from a prior conversation: he thinks the “correlation is not causation” refrain is sometimes taken too far. His observation: people apply this dismissal selectively. When data associates LDL with cardiovascular disease, correlation is not causation gets invoked. When data associates blood pressure, insulin resistance, or weight with cardiovascular disease, those same people do not apply the same critique. He finds that inconsistency telling.

• His actual position: correlation is not causation, and that is true. But statistics has developed tools over decades that can control for variables and shift a correlational finding toward something more like a causal hypothesis — not proof, but meaningful information. Each controlled variable increases the likelihood you are pointing in the right direction.

• Dave’s Bradford Hill framework: he is a Bradford Hill stan. The key criteria are strength, consistency, and temporality. Bradford Hill himself, in his initial speech outlining these criteria, noted that for smoking and cardiovascular disease or all-cause mortality, the hazard ratio was barely two — which he did not consider very strong. Modern nutrition epidemiology regularly makes causal claims from hazard ratios far below two. Dave’s heuristic: he does not give much weight to findings below a hazard ratio of two. Not dismissal, but also not the basis for categorical claims.

• Where they converge: Dave is not dismissing sub-two associations as worthless. He calls them hypothesis-building. The frustration he articulates is that nobody in the nutrition space seems to give that answer. It is either the data are trash (correlation is not causation, full stop) or the data prove X causes Y. The intermediate position — here is a correlation of moderate strength, here is the confidence level warranted, here is what additional evidence would be needed to move toward causality — is almost never offered. Nick V agrees and calls this the nuance problem.

The WHO classification of red meat and the arrogance problem [1:48:20]

• Dave uses the WHO’s classification of red meat as a Class 1A probable carcinogen as an example of the certainty problem. Nick V’s reaction: the word “probable” is doing a lot of work, but it gets lost. People who want to reduce red meat consumption run with “WHO says it causes cancer.” People who want to dismiss it run with “correlation is not causation.” Almost nobody occupies the middle: this organisation thinks there is enough evidence of an association to classify it at this level of probability, here is what probable means here, here is the strength of the effect size, here is the context.

• Nick V’s broader diagnosis of what he calls the arrogance problem: it exists on both sides. On the low-carb side, some people conclude from the LMHR data that LDL simply does not matter and all prior evidence can be dismissed. On the conventional lipidology side, some reduce cardiovascular risk to ApoB or LDL as the only variable worth discussing and resist any conversation about exceptions or subpopulations. Both postures prevent the conversation that is actually needed.

• His defence of clinicians and conventional medicine as a counterweight: clinicians have to be ultra-conservative. They cannot assume an exception applies to their patient without substantial data. The ship turns slowly, and it turns on large amounts of data. That is not arrogance; it is appropriate caution given the responsibility of making treatment decisions for individual patients. What he pushes back on is the refusal to engage with the question of whether exceptions exist and whether subpopulations might be identifiable.

Dave’s challenge: find me a model that predicts LMHR outcomes before we publish the data [1:53:47]

• Dave makes a pointed request directed at people who use statistical models to adjust away metabolic health confounders and conclude that LDL drives atherosclerosis independently of metabolic context: give me your model now, before our longitudinal data is published. Tell me what it predicts for this cohort. Do not come back after the data are out and ad hoc adjust your model to accommodate it.

• His reasoning: lipid dynamics are far more fluid than most models assume. The things that change LDL — dietary fat, carbohydrate restriction, body composition, metabolic state — also change everything else in a metabolically interconnected system. A model that does not account for this interdependence cannot be trusted to isolate LDL as the causal variable. He is the sceptic for data that is not there: he does not assume the model is wrong, but he wants to see it tested prospectively rather than validated post hoc.

• Nick V’s response: this is completely fair. He adds that the ethical considerations that lead trial designers to power studies to events — stopping the trial as soon as statistical significance is reached rather than continuing for a predetermined time period — create their own problems. You cannot know if the effect would have persisted, reversed, or been modified by longer follow-up. The stock market analogy Dave uses: Microsoft being above Apple at one moment does not predict the long-term trajectory.

Publication reform: pre-registration, signed peer review, and multi-journal submission [1:53:00]

• Dave proposes a reform to scientific publishing he has been thinking about: studies should be done in two phases. In phase one, you submit your study design to a journal, negotiate the methodology, and receive a commitment that the journal will publish the results regardless of direction. The peer review happens at the design stage, not after the data are in. In phase two, you execute the study and submit the results.

• The benefits he sees: p-hacking becomes largely pointless because the journal has already committed. Motivated post-hoc restructuring is eliminated. The peer review is productive because it shapes the design before any incentive exists to shade results. Multi-journal submission should be permitted at the design stage — you submit to Nature, JAMA, Lancet simultaneously, whoever accepts first gets it, and the fact of the pre-commitment is on record.

• He adds two additional reform wishes: signed peer review (reviewers should put their names to their critiques, which he finds strange they currently do not have to do) and compensated peer review (reviewers’ time has value, and journals extract enormous amounts of it for free while charging institutions and readers access fees).

• Nick V’s main concern about the pre-registration model: the scooping problem. If peer reviewers at the design stage are researchers in the same small field, they gain advance knowledge of what you are planning. Dave’s answer: names in the open solves this. If reviewers are identified and their role is on record, the incentive to leak or scoop is dramatically reduced. He also notes that LMHR research is somewhat protected from this problem given that he runs the largest LMHR Facebook group — it would be very difficult for a competing team to recruit an equivalent cohort quickly even if they knew the design.

• ClinicalTrials.gov: Nick V praises its existence as a partial solution to the publication bias problem. One of the markers of a good meta-analysis, he notes, is whether it includes ClinicalTrials.gov data — because registered trials are required to report results, creating a database of null findings that would otherwise never be published. Dave agrees and sees this as a step in the right direction but not a complete solution.

The run-in period problem in statin trials, and Dave’s additional complaints about RCT design [2:31:24]

• Dave raises the run-in period, which he considers one of the most underappreciated structural problems in the statin literature. Before a trial officially begins and patients are randomised, the entire population takes the study drug for two to six weeks. Anyone who experiences side effects during this pre-randomisation period can leave the study. They typically do. The result: by the time randomisation occurs, the intervention group is disproportionately composed of people who have already demonstrated they tolerate the drug. Side effects that would have shown up in the intervention group have been pre-selected away.

• This, Dave argues, is one explanation for the discrepancy between the side effect rates reported in statin RCTs and the side effect rates clinicians observe in practice. The surveillance data from actual patients shows substantially higher rates of side effects than the trials. The run-in period is probably filtering them out at the design stage.

• His second complaint: he would prefer time-based RCTs to event-powered RCTs. Stopping a trial the moment statistical significance is achieved means you do not know what would have happened if the trial continued. The effect might have persisted, plateaued, or reversed. The comparison to watching stock prices is apt: Microsoft being above Apple on a given day tells you nothing definitive about the long-term trajectory. He wants to see all prior time-based studies and whether any that achieved statistical significance subsequently saw that significance erode or reverse.

• The post-2004 statin-in-control-group problem: before 2004 some trials had true placebo groups. After 2004, even new cholesterol-lowering drug trials typically give the control group a statin as standard of care. This makes it structurally impossible to study whether statins provide benefit relative to no treatment in contemporary populations. The default has been baked into trial design in a way that makes the alternative invisible.

• Nick V’s ethical counterpoint: there are legitimate IRB-level reasons for these design choices. If existing evidence supports a treatment, it is considered unethical to withhold it from a control group. Dave acknowledges this but notes the circularity: the existing evidence was itself developed in trials that did not yet have this constraint, and the extrapolation from those trials to current populations is where the motivated reasoning tends to enter.

Tribalism in nutrition, why Nick V does not disclose his diet, and the ADA plate problem [2:39:10]

• Nick V has never publicly disclosed what he eats, and this is a deliberate decision. He has watched other science-based creators share “what I eat in a day” videos and then be defined by them for years. People will dismiss a creator’s entire body of scientific analysis because of one food they eat on camera. He wants to be known only as someone who follows the data. He has a video titled “I am a hypocrite” making exactly this point: what he does is not always what the science would optimally prescribe, because he does not do this to be maximally optimal — he does it because he loves the science.

• Dave’s version of the origin of low-carb tribalism: he spent decades being told fat was fattening, carbs were harmless, and that the phrase “empty calories” was how you should think about sugar. Then he found forums where people were saying the opposite, tried it, and his A1C went down. The conventional institutions were wrong in a way that was affecting his health. That personal experience of being failed by the conventional advice is, he thinks, the origin of the emotional intensity in the low-carb community. They are not entirely wrong to be angry. The system did let them down.

• The ADA plate incident that hit Dave personally: just before the summer of the year prior to this recording, the ADA released new material including a plate graphic dividing dinner into half non-starchy vegetables, one quarter lean protein, and one quarter “healthy carbs” — featuring potatoes or rice. Dave had an emotional reaction. He has family members with severe insulin resistance who follow ADA guidance and are comforted by doctors who point to their low LDL as evidence they are practically heart-attack-proof — when they are metabolically unwell. The plate graphic showing rice as a recommended food for diabetics struck him as exactly the wrong prescriptive advice for the people closest to him.

• He is also explicit about something he considers important: he does not like how much people in the nutrition space comment on the physical appearance of people they disagree with. He saw it with the ADA video and finds it across the board. Individual health circumstances are not visible from appearance. You do not know what else is going on in someone’s life. He tries not to do it and does not tolerate it when he sees it.

• Nick V’s framing on why dietary tribalism is so intense: people’s dietary choices are tied to their personal health stories, their identity, and their sense of having figured something out that the system missed. When you change your diet and feel better, you do not just update a belief — you build a narrative around it. Challenging that narrative feels personal. He notices that almost no influential creator ever says publicly: “You know, I’ve been championing this diet for years, but honestly I’ve been getting tired of it and I’m just happier eating differently.” That level of honesty would be disarming. It almost never happens.

COI, bridging the gap, and why the conference attracts a low-carb bent [2:56:34]

• Nick V raises the COI structural challenge: the conference naturally attracts a low-carb-leaning audience because that is who follows Dave and who is most interested in LMHR research. That creates a selection effect that makes it harder to have genuinely diverse scientific dialogue.

• Dave’s approach: a lot of earnest effort over time. He describes how some attendees who might be expected to decline — people with more conventional lipid views — have been approached respectfully, told the full speaker lineup in advance, and have come anyway. Bill Cromwell came to the previous COI. Dave tried to get Gil Carvalho (Nutrition Made Simple), Tom Rifai, and Mario Kratz (Nourish by Science) for the current one. He wants these people in the room because the conversation changes when people are physically present with each other.

• Nick V’s observation about why in-person dialogue is different: on social media you only see the extremes. When people with different views are actually in the same space having a conversation, the audience sees how much overlap there actually is — and the areas of genuine disagreement become clearer too, without the tribal amplification. At the COI panel he attended, the one common theme that cut across all disagreements was that lifestyle intervention should come before pharmacological intervention. That “shouldn’t be revolutionary,” he says — but saying it together in the same room, on record, means something.

• The charisma confounding factor Nick V raises: scientists who are careful and nuanced often come across as dry compared to people who deliver bumper-sticker takes with confidence. The careful person has to explain reverse causality, context-dependence, and the limits of the study design. The confident person just says “lower ApoB means no heart disease.” People are drawn to the latter even when the former is more accurate. Dave’s response: nuance and charisma are not mutually exclusive, and some of his strongest convictions came from failing to defend weaker positions against good interlocutors. The challenge for nuanced communicators is to find ways to be compelling without sacrificing the precision.

• Nick V’s closing prediction for the longitudinal data: he firmly hypothesises no rapid progression of atherosclerosis comparable to a high-risk population. He believes metabolic health is a hugely relevant factor. He would be surprised if it were not. His position: LMHR are at lower risk than their LDL alone would suggest, but the question of whether they are at the same risk as someone who has everything optimised including low LDL remains open. That is what the data will tell them. He is genuinely excited to find out.

Share

Leave a comment