Reviewing Dave Feldman’s “Lipid Triad Model” & Claims

In All Articles, Blog Posts by Alan FlanaganLeave a Comment

A previous episode of Sigma Nutrition Radio (episode 321) provided an opportunity to have a real-time discussion about lipids and risk, and in particular to discuss the concept of the ‘lipid triad’ proposed by the show guest, Dave Feldman.

Dave has become one of the most prominent voices in the low-carbohydrate, high-fat (LCHF) community, and Dave is particularly interested in the risk (or lack of) of elevated low-density lipoprotein (LDL), when it is observed in the context of high levels of high-density lipoprotein (HDL), and low triglycerides (TGs). This combination of high LDL, high HDL, and low TGs, is what Dave has termed the ‘lipid triad’.

This combination is commonly reported anecdotally from the LCHF community, as a response to going on an LCHF diet. The conversation with Dave in SNR episode #321 provided a lot of food for thought, and in the following written segment we have attempted to take Dave’s case at its highest based on the points made in the podcast, and address key aspects of the argument by reference to the current literature. Previous 'Sigma Statements' on lipids, diet & CVD (available here) give a more general overview of our position. The goal of this article is to zero in on some specific points made in the podcast episode with Dave.

So rather than go into every individual point made in that discussion, this article will focus in on points Dave made in relation to four concepts:

  1. The ‘Lipid Profile-Centric Model'
  2. Atherogenic Dyslipidemia and the ‘Lipid Triad’
  3. Remnant Lipoproteins
  4. Inflammation

The ‘Lipid Profile-Centric Model'

Dave proposes what he calls the "lipid profile-centric model" as an alternative the widely-accpepted consensus viewpoint in lipidology and cardiovascular sciences (which Dave labels a "lipoprotein-centric viewpoint"). The proposed model was mentioned in the podcast episode:

Dave Feldman [10:37]: "I think conventional medicine, especially in lipidology has a kind of lipoprotein centric viewpoint which is to say that lipoproteins are themselves pathogenic or in particular at least Apo B containing lipoproteins, LDL particles especially that they themselves at a certain concentration level will drive atherosclerosis. And what I sort of want to put on the table is something that I would call a kind of lipid profile centric model, which is to say that I want to distinguish how much it is that a lipid profile drives the disease especially of atherosclerosis or whether or not it's the disease that drives the lipid profile..."

The first key point to note here is Dave’s suggestion that conventional medicine has a "viewpoint" that ApoB-containing lipoproteins (especially LDL) “at a certain concentration will drive atherosclerosis." This phrasing suggests that the causal role of ApoB (and LDL) in driving atherosclerosis is simply a competing hypothesis, with as much claim to be correct as others. But this is just not the case. Within the lipidology and cardiovascular sciences community, that LDL-C and ApoB-containing lipoproteins drive atherosclerosis is not considered a mere “viewpoint”, but rather an established fact with sufficient proof to deem LDL-C causal.

In looking at the issue of causality in a scientific context, Dave's hypothesis of "whether the disease drives the lipid profile…” (i.e., reverse causality) can be addressed through three of the specific criteria for establishing causality in science (the Bradford-Hill criteria):

The temporal relationship between high blood cholesterol levels and cardiovascular disease (CVD) is well established in prospective cohort studies indicating that in disease-free participants at baseline, exposure to increasing LDL-C concentrations precedes increased risk of atherosclerotic cardiovascular disease events.

The Prospective Studies Collaboration, (which included 892,337 participants from 61 prospective cohort studies free from cardiovascular disease (CVD) at baseline, a collective 11.6-million person-years at risk during which 33,744 ischaemic heart disease deaths occurred) found a pronounced and linear increase in ischaemic heart disease deaths mortality over time. From a temporality perspective, the association increased with age at-risk, such that the effect of exposure to 1.0mmol/L lower LDL-C in the 40-49yo age bracket was associated with 56% (HR 0.44, 95% CI, 0.42-0.48) lower risk for ischaemic heart disease mortality. 

This almost mirrors the evidence from Mendelian randomisation studies on genetic exposure to lifelong lower LDL-C levels, where there is a 54% relative risk reduction of coronary heart disease (95% CI, 0.41-0.52) per each 1.0mmol/L lower long-term exposure to LDL-C was.

Finally, the temporal relationship between exposure to higher LDL-C and atherosclerosis is confirmed by the evidence from Familial Hypercholesterolaemia (FH), which irrespective of the underlying mechanism (i.e., different genetic defects in LDL-receptor function), results in profoundly elevated LDL-C concentrations, early and rapid development of atherosclerosis, and early mortality from coronary artery disease if left untreated.

Biological Gradient
biological gradient, or dose-response, is evident in every line of evidence. The prospective studies have indicated a strong relationship between increasing CVD risk with increasing exposure to LDL-C. This relationship is confirmed by meta-analysis of statin intervention studies, where the relative risk per unit increase in LDL-C increases in linear fashion. What the temporal relationship displays is the effect of a cumulative exposure over time; if this were true, then reversibility should be evident, i.e., does reducing the exposure change the incidence of disease.

This change in disease incidence by reducing the exposure is precisely what is shown, irrespective of the method of reduction. So looking at drug targets, we can consider:

  1. HMGCR: The target of statins
  2. NPC1L1: The target of ezetimibe
  3. PCSK9: The target of PCSK9-inhibitors

Statins reduce the risk of CVD by approximately the same amount per unit reduction in LDL-C that would be predicted from MR studies. The same is true for variants in the NPC1L1 gene, which is the target of the drug, ezetimibe. Per unit reduction in LDL-C, variants in both NPC1L1 and HMGCR have a near identical effect on lower CVD risk. Intervention shows what would be anticipated from genetic predictions; IMPROVE-IT demonstrated that the 20% reduction in risk for vascular events after 6yrs from ezetimibe and the achieved reduction in LDL-C 0.33mml/L (12.5mg/dL) is exactly what would be predicted from the same duration of treatment and same level of LDL-C reduction from statin interventions (18%).

Effect of PCSK9 variants on reduced risk is the exact same as both HMGCR and NPC1L1; again, the implication is that PCSK9-inhibitors and statins should have the same effect on risk of CVD events per unit reduction in LDL-C. This is exactly what interventions show; FOURIER used a PCSK9-inhibitor, reduced LDL-C by 54mg/dL and resulted in a 17% reduction in CVD events after 2-years. This effect size in this duration is exactly what would be expected from a statin intervention achieving the same level in the same duration. Both of these effects have been predicted precisely through MR.

In the above graph from Ference et al., there is a stark difference in the slopes of the two lines: the top line representing genetically lower LDL-C, and the bottom line representing trials in which LDL-C was lowered pharmacologically. The difference in the slopes of the lines showing that the relative risk reduction at a given LDL-C reduction is not as pronounced for drug interventions as it is for the genetic variants causing the lower LDL-C. This is explained by considering the length of exposure to lower LDL-C:

  1. In drug intervention trials the mean age at the outset is 63 years old; so previous to this, LDL-C was much higher.
  2. Whilst the genetic variants mean lifelong exposure to lower levels.

Therefore, genetic exposure to lifelong low LDL-C levels is associated with a 3-fold greater reduction in risk than treatment with a statin later in life. Hence why the effect is greater. The clinical benefit of lowering LDL-C seems to be determined by both the absolute reduction in LDL-C and the total duration of exposure to lower LDL-C.

Thus, convincing evidence exists that lowering the exposure reduces the incidence of CVD. In particular, there is no evidence of a threshold to benefit in reducing LDL-C, with levels of <1.29mmol/L (<50mg/dL) demonstrating the most pronounced reductions in risk from statin interventions compared to each incrementally higher category.

The recent FOURIER trial pre-specified for analysis of progressively lower LDL-C levels and clinical outcomes; there was a linear relationship between progressively lower LDL-C and CVD outcomes, with the lowest pre-specified subgroup median of 0.36mmol/L (13.8mg/dL). A quarter of this group achieved on-treatment levels on 0.18mmol/L (6.9mg/dL), with no evidence of adverse effects.

A ‘look under the hood’ in relation to reversibility was provided by the ASTEROID trial, in which high-intensity statin therapy reduced LDL-C from 3.3mmol/L (130mg/dL) to achieved levels of LDL-C 1.5mmol/L (60mg/dL), and reduced atheroma by >50%. Image below compares cross-section at baseline (left) and after 24 months of treatment (right).

Copyright © 2006, American Medical Association. Nissen et al., JAMA. 2006;295(13):1556-1565

The GLAGOV trial similarly demonstrated no threshold to benefit in profound reductions in LDL-C, directly related to the magnitude of coronary atheroma regression. The most important point regarding reversibility is that to achieve regression of coronary atheroma, on the basis of the current evidence, a LDL-C of <1.8mmol/L (<70mgd/dL) is required.

To re-cap:

  • Temporality: exposure to LDL-C precedes incidence of CVD.
  • Biological gradient: increasing levels of the exposure increases incidence of CVD in a dose-dependent manner.
  • Reversibility: Lowering LDL-C reduces risk for CVD morality and events in a dose-dependent manner, with no evidence of ceiling thresholds to benefit of lower levels.
  • Feldman's Claim: The disease drives the lipid profile (i.e., reverse causality).
  • This claim is refuted by fact that the totality of evidence supports that the atherogenic lipoproteins, in particular LDL, are in the causal pathway driving atherosclerosis.
  • This is clearly demonstrated by looking at the Bradford-Hill criteria of temporality, biological gradient and reversibility. Doing so, one finds that:
    a) exposure to LDL-C precedes incidence of CVD.
    b) increasing levels of the exposure increases incidence of CVD in a dose-dependent manner.
    c) lowering LDL-C reduces risk for CVD morality and events in a dose-dependent manner.

Atherogenic Dyslipidemia and the ‘Lipid Triad’

The ‘atherogenic lipoprotein phenotype’ (ALP) is a well-characterised clustering of cardiometabolic risk factors, in particular low-HDL/high-triglycerides (TGs), and a prevalence of small, dense, LDL particles. The contention Feldman seems to offer in support of the "Lipid Triad" is that, as the ALP is known to be atherogenic, if the reverse is the case (i.e. high-HDL and low-TGs), then perhaps this is protective or at least a much less risk:

Dave Feldman [08:54]: “Atherogenic dyslipidemia is typically characterized by having low levels of HDL cholesterol, high levels of triglycerides which is a measure of fat in the blood, and typically a high preponderance of small LDL particles. Now, what's fascinating is that we find within the low-carb community many people who go on a low-carb diet see an almost exact reverse of that, they tend to see that their HDL cholesterol will go up, their triglycerides will go down, and separately their LDL may also increase…”

The first aspect to clarify is the difference between independent causal risk factors and systems biomarkers. This is fundamental to any analytical process which views a number of biomarkers together:

  • Independent risk factor: biomarker in a causal pathway between the exposure and outcome.
  • Systems biomarker: biomarker which provides indications of underlying cardio-metabolic processes, but are not causal independently.

Both HDL and triglycerides are appropriately classed as systems biomarkers; neither are causally associated with CVD in themselves. Neither intervention studies or Mendelian randomisation studies suggest a causal role for HDL with regard to CVD, although epidemiological data consistently suggests a protective effect of higher HDL. Triglycerides are a particularly important systems biomarker; high-TG can provide an indicator of increasing remnant cholesterol levels, increased secretion of VLDL, and precipitate remodelling of HDL and LDL. However, independent associations with TGs are no longer evident once adjusted for non-HDL-C (note: non-HDL-C is the cholesterol content of ApoB containing lipoproteins, namely  VLDL, VLDL remnants, IDL, Lp(a) and LDL). This was confirmed in a recent analysis of genetic variants in TGs which found that once TGs were adjusted for ApoB, the independent relationship with TGs was no longer evident. Collectively, this demonstrates that it is the levels of atherogenic lipoproteins within the measure of TGs which relates to risk.

While observational studies suggest a protective effect of high-HDL/low-TGs, this effect appears to be most pronounced when LDL-C remains below <3.3mmol/L (<130mg/dL). Feldman's view of the "lipid triad" suggests that in the context of high-HDL/low-TGs, the larger LDL particle size means LDL has a lower atherogenic potential. However, the predominance of LDL-I and LDL-II (i.e., the largest LDL particles) in individuals with low plasma TGs results in rapid clearance of LDL-C from circulation, and is therefore characterised by low LDL-C levels (~2.5mmol/L or 100mg/dL). This is lower by orders of magnitude compared to the levels of LDL-C often anecdotally reported from the LCHF community. The only available data of very high LDL-C concentrations in the context of low-TGs is that in homozygous FH, in which both HDL and TGs are in normal range: the accelerated development of atherosclerosis and coronary artery disease relates to LDL-C. This again indicates the difference between a causal, independent risk factor that is a direct target for treatment (LDL-C) and correlated biomarkers that may report on cardio-metabolic function, but are not independent causal factors themselves.

Thus, while LDL particles do possess varying degrees of atherogenicity, the fact remains that all LDL particles are atherogenic. To argue that in the context of high-HDL/low-TGs, LDL-C is not atherogenic, implies a causal role for HDL and TGs, which is not established on the basis of the current evidence. It also assumes that LDL-C causality is circumstantial to a specific phenotype, when in fact the evidence for LDL-C as the causal pathway is independent of other, correlated biomarkers. 

In this context, for the ‘lipid triad’ to be true, it must demonstrate that high LDL-C alone is not a risk factor for CVD, independent of its relationship to other correlated biomarkers. Establishing this would mean providing sufficient evidence to rebut the body of evidence supporting LDL-C causality itself, including:

  • over 200 prospective cohort studies
  • randomised controlled clinical intervention trials with a collective two-million participants
  • 20-million person-years of follow-up during which over 150,000 CVD events occurred
  • Mendelian randomisation studies on genetic predispositions to elevated or reduced lifelong exposure to LDL-C
  • Mechanistic research supporting the underlying pathology

  • Feldman's Claim: In the context of high-HDL/low-TGs, LDL-C is not atherogenic.
  • This claim implies a causal role for HDL and TGs. A causal role for either HDL and/or TGs is not established on the basis of the current evidence.
  • The evidence for LDL-C as the causal pathway is independent of other correlated biomarkers, thus undermining the claim that LDL-C causality is circumstantial to a specific phenotype.

Remnant Lipoproteins

The context of the ‘Lipid Triad’ acknowledges that the ALP presents a risk for atherosclerosis. Indeed, Feldman directly states:

Dave Feldman [19:58]: “... you probably have a higher preponderance of VLDLs, and that does tend to be associated with atherogenic dyslipidemia... And remnants are, as a profile, definitely highly associated with cardiovascular disease…

However, this appears to be impossible to reconcile with the contentions supporting the 'Lipid Triad' that high LDL-C is not a risk for atherosclerosis; the broad functional properties of these lipoproteins, diameter size and presence of ApoB, means that to acknowledge that VLDL and/or remnants are atherogenic implicitly acknowledges LDL as such. In fact, recent evidence indicates that the risk is almost identical per unit increase in cholesterol:

What this indicates is that the cholesterol content of ApoB-containing lipoproteins is the common issue, and increasing the cholesterol content of remnants mirrors the risk related to increasing the cholesterol content of LDL on a per-unit basis. In the JUPITER trial, participants achieved a median LDL-C of 1.4mmol/L (55 mg/dL), with low triglyceride of 1.1mmol/L (102mg/dL). However, in the context of very low LDL-C levels, residual risk was driven by the cholesterol content of VLDL, not the TG content. In fact, no association was observed for TG measures, and changing TGs was not associated with risk. Cumulatively, the data is consistent with the cholesterol content of atherogenic lipoproteins being the primary concern for atherosclerosis. 

  • Feldman's Claim: Remnant lipoproteins are "definitely highly associated with cardiovascular disease" but LDL-C is not a risk factor.
  • It is impossible to acknowledge remnant cholesterol as a factor in CVD risk without accepting that LDL-C is a risk factor.


Dave Feldman [104:40]: “I've really seen a much stronger acknowledgement towards the process of inflammation and its association back with atherosclerosis. And again with full acknowledgment, as I did earlier, that we haven't yet been able to fully elucidate how much LDL particles initiate and progress the disease, but how much they are involved, how much inflammation for that matter and the inflammatory response is involved.”

We do not propose to address the statement that, "we haven't yet been able to fully elucidate how much LDL particles initiate and progress the disease…”, because this is largely addressed above: LDL is in the causal pathway initiating the process of atherosclerosis, and progressing the disease. The progression of disease, and the risk for disease, is strongly mediated by inflammation. High-sensitivity C-reactive protein (hsCRP) is an important systems biomarker, and a strong predictor of CVD risk. It is also a mediator of the clinical benefit associated with LDL-C lowering, and a number of excellent recent trials have elucidated the relationship between magnitude of risk reduction, and achieved levels of LDL-C and hsCRP.

In the SHARP trial, reductions in LDL-C reduced CVD risk independent of whether baseline levels of hsCRP were >3mg/L or <3mg/L. In FOURIER, treatment with a PCSK9-inhibitor reduced CVD risk independent of baseline hsCRP, in participants stratified by baseline hsCRP <1mg/L, 1-3mg/L, and >3mg/L, respectively. However, the greatest clinical benefit was observed in those participants with highest baseline hsCRP levels, and the greatest reduction in risk was observed in patients with both the lowest LDL-C and hsCRP levels. JUPITER recruited participants with baseline LDL-C of <3.3mmol/L (<130mg/dL) and hsCRP 2mg/L, however, although there was some degree of improvement in inflammatory markers, the magnitude of CVD risk reduction related to achieving LDL-C levels of <1.8mmol/L (<70mg/dL). This suggests a threshold effect by which residual risk from either inflammation or LDL-C may persist.

This is precisely what was shown in both PROVE-IT and IMPROVE-IT, where the greatest magnitude of effect in reducing CVD risk was derived from achieving both an LDL-C <1.8mmol/L (<70mg/dL) and hsCRP <2mg/L. Of critical importance is the fact that there is no evidence that elevated LDL-C with low hsCRP is not a risk, i.e., that (as suggested by many in the LCHF community) LDL-C is only potentially atherogenic in the context of high inflammation. This is clearly not the case in the evidence from well-conducted, large, intervention studies. Where LDL-C remains >70mg/dL but hsCRP is <2mg/L, there remains ‘residual cholesterol risk’ that is equivalent to if <70mg/dL but hsCRP is <2mg/L.

Copyright © 2019, Wolters Kluwer Health.
From: Ridker, Circulation Research. 2019;124:437–450

An important factor in the trials mentioned above - SHARP, FOURIER, JUPITER, PROVE-IT, IMPROVE-IT - is that these were not direct tests of the effects of lowering inflammation; hsCRP was an additional measure in the context of lipid-lowering therapy. The first trial to specifically address inflammation reductions with change in lipoproteins was the recent CANTOS trial, where patients in secondary prevention on lipid-lowering therapy and achieved LDL-C in the 1.8mmol/L (70mg/dL) range, were assigned to an anti-inflammatory drug or placebo. Profound reductions in hsCRP, without altering any lipid parameter, led to a significant 25% reduction in CVD risk; the effect mirrors that observed in FOURIER, which used a PSCK9-inhibitor to achieve further reductions in LDL-C. 

Cumulatively, these data indicate that the risk associated with LDL-C persists independent of inflammation. However, inflammation is a strong mediating factor in the progression of atherosclerosis, and hsCRP is an important systems biomarker, particularly for identifying residual risk where LDL-C is low (<1.8mmol/L (<70mg/dL)), but hsCRP remains elevated >2mg/L. However, where hsCRP is <2mg/L but LDL-C remains elevated, the risk relates to the concretely established causal role of LDL-C in driving atherosclerosis.

  • Feldman's Claim: In the context of low inflammation, elevated LDL may not infer increased ASCVD risk.
  • Risk associated with LDL-C persists independently of inflammation.
  • Inflammation is a strong mediating (but not causal) factor in atherosclerosis development. LDL is independently causal.

Overall Conclusions

  1. On review of some of the claims made in SNR #321, and subsequent follow-up review of the literature, we find no compelling evidence to support that in the context of high-HDL/low-TGs, very high LDL-C is not a risk factor.
  2. We are in agreement with the European Atherosclerosis Society, that the scientific criteria for established causality has been convincingly discharged in the relation to LDL-C, which is in the causal of pathway of atherosclerosis.
  3. In this context, LDL-C is an independent causal risk factor that is a target for treatment.
  4. This is in contrast with TGs and hsCRP, which are important systems biomarkers, but not causal factors.
  5. In particular, both TGs and hsCRP may be additional targets for treatment in the context of residual risk after LDL-C lowering to current target levels.
  6. While we acknowledge that high HDL appears to be protective in epidemiological data, the absence of evidence from intervention studies and lack of causal relationship evident in Mendelian randomisation studies do not support emphasis on independent effects of HDL in itself. Indeed, over-reliance on high HDL-C in the context of high LDL-C may be a misnomer for risk management.

Leave a Comment