# HSEF Review: LDL Particle Concentration as a Causal Driver of Atherosclerotic Cardiovascular Disease

**HSEF v5.4 structured hypothesis review**  
**Review date:** 2026-09-29 (authoring run) · **Red-team synthesis:** 2026-09-30  
**Method packet SHA256:** `7E097D3D50D2068F68BDB1CD022DF4A045DAC46A9F25E51262EC7568E9DB11EB`  
**Provenance:** 39 sources · 28 claims · 28 verified · 0 unverified · 0 contradicted

> **Adversarial review:** Generated by Sol 6.1; adversarially reviewed by Claude Opus 5.5 (Anthropic) on 2026-10-14. 1 correction applied (see ADVERSARIAL REVIEW AND APPROVAL). Status: **Approved as revised.**

The supplied packet is used as review methodology, not as scientific evidence. The submitted sentence is preserved and then restated as measurable antecedents before grading. LDL particle number (LDL-P), apolipoprotein B (apoB) concentration, LDL cholesterol (LDL-C) and non-HDL cholesterol (non-HDL-C) are kept distinct throughout. This edition is the published synthesis of a cross-model red-team audit of the 2026-09-29 review. What changed, and why, is summarised in the "Red-team synthesis note" near the end and documented in full in the companion audit logs.

## LAY SUMMARY

**Overall grade: Ζ (zeta) — Multiple independent causal lines converge. Confidence: Moderate.** The grade needs a careful reading, because two questions sit inside one sentence. The first question is whether LDL-type particles in the blood cause atherosclerotic heart and artery disease. The evidence for that is strong. People who inherit lifelong lower LDL have much less coronary disease, several unrelated drug classes that lower LDL reduce heart attacks and related events, and plaque imaging moves in the predicted direction [2-5,7-16,38]. On their own, those parts of the claim grade at Η or Θ.

The second question is what "particle concentration" adds: whether the **number** of particles, rather than the cholesterol they carry, is the causal quantity. Here the evidence is real but unsettled. Genetic studies, large cohorts, and drug trials in which LDL cholesterol and particle count moved apart all agree that LDL cholesterol alone is not the whole story [17-19,21,25-27,30-33]. But independent genetic analyses disagree about whether particle number beats total cholesterol content [20]. They also show that particles of different types carry different risk per particle: remnant particles and lipoprotein(a) are more harmful per particle than LDL [35,36]. Because the review's rules, fixed before scoring, require every part of the claim to hold, the unsettled particle-number part sets the overall grade.

In plain terms, LDL-class particles are a primary cause of atherosclerotic disease. What is not settled is whether counting particles captures that cause better than measuring the cholesterol they carry. LDL is also not the only cause: blood pressure, smoking, diabetes, inflammation, plaque rupture and clotting, lipoprotein(a), and other illnesses all contribute [39].

## AT A GLANCE

| Field | Adjudication |
|---|---|
| Submitted hypothesis | Elevated LDL particle concentration is a causal driver of atherosclerotic cardiovascular disease |
| Gradeable restatement | Higher cumulative exposure to circulating LDL-class apoB particles increases ASCVD through arterial entry and retention; lowering exposure should reduce atherosclerotic events; and particle burden, rather than cholesterol mass, is the operative quantity when the two diverge (preregistered SC4) |
| Overall HSEF | **Ζ — Multiple independent causal lines converge** |
| Confidence | **Moderate** overall; **High** for LDL-class causality (SC2, SC3, SC5); **Moderate** for particle-number specificity (SC4) and the retention step (SC1) |
| Role | **Primary driver** of ASCVD through atherogenesis, at the level of LDL-class apoB particles (derivation step 3) |
| Best model | Broader apoB-particle retention model (M2), itself qualified by unequal per-particle atherogenicity |
| Nominal-model rank | 2 of 5 |
| Deciding experiment | A replicated human discriminator that moves particle number and total cholesterol cargo in opposite directions and tracks hard ASCVD outcomes |
| Aggregation sensitivity | Weakest-link (preregistered) gives Ζ; scope-matched would give Η; primary-subclaim gives Ζ if SC4 is primary, Η if SC3 is |
| Momentum / Stability / Divergence | Mechanistic only / Stable / Mild overstatement (implied Η, assessed Ζ, +1) |
| Calibration | Not a calibration entry; no sealed target supplied |
| Evidence access | 39 sources: 14 read in complete full text across the authoring and red-team passes; the rest at PubMed abstract/metadata depth, labeled per row |

### Subclaim matrix

| ID | Subclaim | Type | Scope | Level | Confidence |
|---|---|---|---|---|---|
| SC1 | Arterial entry and retention initiate and propagate lesions | mechanism | arterial atherogenesis | Η | Moderate |
| SC2 | Lifelong inherited exposure changes human coronary risk | causality | human lifelong LDL-class exposure | Θ | High |
| SC3 | Lowering through independent targets changes plaque and hard events | reversibility | prespecified ASCVD outcomes | Η | High |
| SC4 | When measures diverge, risk tracks particle burden more than cholesterol mass | measurement-specific mechanism | LDL-P/apoB versus LDL-C and non-HDL-C | Ζ | Moderate |
| SC5 | The effect generalizes across major ASCVD settings with mapped limits | generalization | primary prevention, established ASCVD, CKD | Η | High |

**Aggregation:** SC1=Η, SC2=Θ, SC3=Η, SC4=Ζ, SC5=Η. The preregistered weakest-link rule across SC1–SC4 returns **Ζ**, and SC5 does not lower it. Under the other rules, scope-matched aggregation would return Η (SC5 matches the human-ASCVD scope), and primary-subclaim aggregation returns Ζ with SC4 designated primary or Η with SC3. The one-rung spread is caused entirely by the particle-number conjunct.

## COMPETING MECHANISTIC MODELS

1. **Broader apoB-particle retention model (M2) — Η.** On this model, cumulative exposure to all arterial-retained apoB particles (LDL, triglyceride-rich remnants and lipoprotein(a)) is the causal lipid burden. It explains why lipolysis-pathway (LPL) and LDL-receptor-pathway variants both change coronary risk when scaled to apoB [17]. It explains why benefit from CETP-type LDL-C lowering follows the smaller apoB change [25]. And it explains why apoB outranks other lipid measures in multivariable MR and in combined cohort models [18,19,32]. It is graded Η rather than Θ because of two unresolved exceptions. Per-particle atherogenicity is not uniform: remnant-rich and Lp(a) particles carry more risk per apoB than LDL [35,36]. And one MR analysis favors total cholesterol cargo over particle count [20].

2. **Nominal LDL-particle model (M1) — Ζ.** On this model, the number of circulating LDL particles determines how many enter and are retained in the arterial wall. It is supported by LDL retention biology [1], LDL-P discordance in MESA and Framingham [30,31], and apoB-prioritizing genetics [17-19]. It ranks second for two reasons. ApoB is broader than LDL-P, and the evidence that non-LDL apoB particles are causal [17,35,36] is what separates M2 from M1. And the particle-over-cargo question that gives this model its distinctive content is contested [20,35].

3. **Cholesterol-cargo model (M3) — Ζ.** On this model, the cholesterol delivered by retained particles, best represented by non-HDL-C, is the operative exposure. It is directly supported by Helgadottir et al. [20]. It is indirectly supported by the higher per-particle risk of cholesterol-rich remnants [35] and by the proportional relation between LDL-C lowering and benefit [15]. The **LDL-C-only** version is disfavored by the CETP genetics and trials [25-27] and by discordance cohorts in which LDL-C loses its association once particle or non-HDL-C measures are considered [21,30-33]. The **non-HDL-C** version remains a live competitor.

4. **Plaque-instability and thrombosis model (M5) — Ζ for the event transition, as assessed on this ledger.** Inflammation, plaque rupture or erosion, and thrombosis govern when an existing plaque causes a clinical event. Interleukin-1β inhibition reduced recurrent events without lowering lipids [39], and residual events persist after intensive LDL lowering [9-13]. Antithrombotic trials were outside this review's ledger. The model complements LDL-driven plaque formation rather than replacing it.

5. **Inflammation or endothelial-injury-first model (M4) — Ε as a complete alternative.** Inflammation is causally operative in the recurrent-event setting [39]. But the claim that LDL is merely permissive fails against consistent lifelong genetic effects and cross-target lipid-lowering trials that did not target inflammation [2-16,38]. This model falls under the immunology rubric (05) for its own evidence.

The tournament informs the role of mechanism and the biological wording. It does not set the HSEF letter. The study that most directly separates rank 1 from rank 2 is Ference et al. [17]: the LPL and LDLR scores gave nearly identical risk per unit apoB, which shows that non-LDL apoB particles are causal. Björnson et al. [35,36] agree that non-LDL particles are causal but dispute the equal-per-particle interpretation.

## HYPOTHESIS STATE EVALUATION FRAMEWORK

<!-- consensus-consistency-2026-10-09 -->
### Consensus consistency validation — 9 October 2026

This is a transparently unblinded consistency check, not a new peer review or an independent blinded audit. The review's frozen cutoff remains **2026-09-29**, and its grade and scorecard counts are preserved.

**Grade:** Ζ / Moderate (unchanged). The review's central wording adds a stronger particle-concentration proposition than LDL-class causality alone. SC4 is the binding subclaim because evidence comparing particle number with cholesterol cargo is heterogeneous and directionally inconsistent. The discovery search's first-ranked European Atherosclerosis Society consensus statement is secondary synthesis and is not treated as primary evidence.

**Frozen scorecard inventory:** 39 citation-ledger sources, of which 33 are scored evidence rows; 28 claims. The Consensus keyword searches and fetched records are retained as raw JSON in this folder. Search output is discovery metadata; primary claims below are checked against the review's frozen source records and recorded PubMed abstract/metadata inspection.

**Primary comparisons; all already in the original review inventory:**

| PMID / DOI | Source type and ledger role | Original access recorded | Finding and reconciliation |
|---|---|---|---|
| 12066187 / 10.1038/nature00804 | Primary mouse retention-mechanism experiment; SC1 | PubMed abstract and metadata | Altering apoB proteoglycan binding changed atherosclerosis in vivo. Supports a retention mechanism in mice, not human event effect size or the unique causal role of LDL particle count. |
| 16554528 / 10.1056/NEJMoa054013 | Human PCSK9 genetic natural experiment; SC2/SC5 | PubMed abstract and metadata | PCSK9 variants associated with lifelong lower LDL-C and lower CHD risk. Strongly supports LDL-C causality but indexes LDL-C and correlated particle traits together; it cannot adjudicate particle number versus cholesterol cargo. |
| 7968073 / DOI not recorded in the frozen ledger | 4S secondary-prevention randomized trial; SC3/SC5 | PubMed abstract and metadata | Simvastatin reduced major coronary events and mortality in established CHD. Supports LDL-lowering benefit; a statin trial does not separate LDL-P from LDL-C or exclude every drug-specific pathway. |
| 28304224 / 10.1056/NEJMoa1615664 | FOURIER randomized evolocumab outcome trial; SC3/SC5 | PubMed abstract and final metadata | Lower LDL-C with PCSK9 inhibition reduced cardiovascular events. LDL-C, apoB, and LDL-P move together, so the trial does not isolate particle number from cholesterol content. |

These source boundaries reproduce the frozen review's distinction between established LDL-class causality and the more specific number-versus-cargo claim. The fresh Consensus output supplies additional discovery pointers but no independently source-verified finding that resolves the binding discordance question. The score stays Ζ; original sources/claims stay at 39/28 (33 of the 39 sources are scored evidence rows).



This dated supplement adds source reconciliation to the original review. Its original narrative, registered scope, source inventory and grades remain historical records. No earlier adversarial approval has been verified for this edition at the supplement date. This provenance supplement has no separate approval and is not an independent or blinded adjudication. New sources described here are supplemental and are not counted in the original inventory.
<!-- end-consensus-consistency -->


### 1. Claim parsing and falsifiable scope

The hypothesis is not read as claiming that LDL is sufficient, necessary in every individual, or the sole determinant of an acute event. Its testable form is cumulative and upstream. Greater LDL-class particle exposure should increase arterial flux and retention. Inherited exposure differences should change human disease. Lowering through independent targets should alter plaque and atherosclerotic events.

The entry says "particle concentration" rather than "LDL cholesterol". The preregistration (locked 2026-09-29, before outcome adjudication) therefore added a particle-number conjunct, SC4, and applied weakest-link aggregation across SC1–SC4. This edition keeps that rule. The distinctive content of this entry, and the strong form its proponents defend, is that particle number rather than cholesterol content is the causal quantity. Read more loosely, as "LDL-class particles cause ASCVD", the claim would be graded by the scope-matched rule, which returns Η. That alternative is reported and not substituted.

The qualifiers are outcome-independent. "LDL-class" is defined by direct LDL-P or by prespecified apoB/LDL-C contrasts. "Cumulative" means concentration over time or a lifelong genetic proxy. "ASCVD" means prespecified coronary or major atherosclerotic endpoints, with nonatherosclerotic cardiovascular death separated.

### 2. Rubric routing

All seven supplied rubrics were checked. None covers atherosclerosis or lipidology. The metabolism rubric (06) is the closest fit for the nominal lipid-exposure claim, because it explicitly names "Mendelian randomization on the exposure itself, multiple pharmacological classes acting on the same node with concordant endpoint effects, dose-response across independent populations, and natural experiments of lifelong exposure" as upper-ladder equivalents. Its lifespan/healthspan language is adapted here to hard ASCVD events and plaque trajectory. That adaptation does not allow biomarker-only evidence to cross the causal boundary. The immunology rubric (05) governs the inflammation-first competitor (M4) and the inflammatory part of M5. Its caps bind inflammation claims and do not lower the nominal lipid-exposure claim under the cross-field lowest-ceiling rule. The hypothesis index has no entry for this claim.

### 3. Arterial mechanism

Skålén et al. changed the proteoglycan-binding capacity of apoB-containing LDL. Mice carrying binding-defective particles developed significantly less atherosclerosis [1]. This is a direct intervention on retention rather than a serum correlation. Its limits are that it is one perturbation in one species, read here at abstract depth, and that it measures lesions rather than clinical events.

GLAGOV supplies a human trajectory bridge. Adding evolocumab to statin therapy changed percent atheroma volume from a **0.05% increase** with placebo to a **0.95% decrease**, and plaque regression occurred in **64.3% versus 47.3%** of patients [16]. Imaging is a surrogate, and the trial shares the PCSK9 class with the outcome trials. It therefore supports "lowering changes lesions" but does not test the retention step.

SC1 is held at Η under the rubric's convergence rule. Human genetics and trials establish that LDL-class exposure drives lesions, and the mouse perturbation supplies mechanistic specificity. Confidence is **Moderate**, because the retention step itself rests on one cited perturbation.

### 4. Lifelong human causality

PCSK9 loss-of-function provides a clean natural experiment. In Black ARIC participants, nonsense variants associated with **28% lower LDL-C** were associated with an **88% lower CHD risk** (HR 0.11). In White participants, a variant associated with **15% lower LDL-C** was associated with a **47% lower risk** (HR 0.50) [2]. A synthesis of nine variants in six genes (312,321 participants) found a **54.5% lower CHD risk per 1 mmol/L lifelong lower LDL-C**, about three times the per-unit reduction seen with statins started later in life [3].

NPC1L1 inactivating-mutation carriers had about **12 mg/dL lower LDL-C** and an **OR of 0.47** for CHD, which matches the target of ezetimibe [4]. PCSK9 and HMGCR genetic scores each gave an **OR of 0.81 per 10 mg/dL lower LDL-C**, with independent, additive effects [5]. Sequencing of FH genes in 26,025 participants showed that, within any stratum of observed LDL-C, mutation carriers had higher coronary risk than non-carriers. Carriers with LDL-C ≥190 mg/dL had an **OR of 22.3** against a reference group with LDL-C <130 mg/dL, compared with **6.0** for non-carriers with LDL-C ≥190 mg/dL, and carriers had greater cumulative LDL-C exposure [38]. An earlier population study reported an OR of 13.2 for coronary disease with clinically diagnosed FH [6]. That estimate is inflated by design, because the Dutch Lipid Clinic criteria used for diagnosis include a personal history of premature coronary disease. It is retained as supportive only.

The consistency across genes, ancestries, exposure durations and drug targets makes reverse causation or single-target pleiotropy an inadequate global explanation. SC2 reaches Θ for LDL-related cumulative exposure. It does not settle whether LDL-P, apoB or cholesterol cargo is the operative scale.

### 5. Pharmacologic reversibility and hard outcomes

The evidence is counted as independent by target class, not by paper. Statins reduced events in primary and secondary prevention. In WOSCOPS, LDL-C fell by **26%** and definite coronary events by **31%**. In 4S, simvastatin reduced all-cause mortality (**RR 0.70**), coronary death (**RR 0.58**) and major coronary events (**RR 0.66**) [7,8]. These trials form one HMGCR line.

NPC1L1 inhibition supplied a different mechanism. In IMPROVE-IT, ezetimibe lowered median time-weighted LDL-C from **69.5 to 53.7 mg/dL** and reduced the seven-year event rate from **34.7% to 32.7%** (**HR 0.936**) [9]. PCSK9 antibodies added another class. FOURIER and ODYSSEY OUTCOMES each reported **HR 0.85** in established ASCVD and recent acute coronary syndrome [10,11]. VESALIUS-CV reported **HR 0.75** for three-point MACE in patients with atherosclerosis or diabetes who had no previous myocardial infarction or stroke [13]. The three trials count as one PCSK9 line.

CLEAR Outcomes extended concordance to ATP-citrate-lyase inhibition in statin-intolerant patients. LDL-C fell **29.2 mg/dL** more than with placebo (a difference of **21.1 percentage points** in percent reduction). Primary MACE had **HR 0.87**, myocardial infarction **HR 0.77** and revascularization **HR 0.81**, with no significant reduction in cardiovascular or all-cause death [12]. SHARP reduced major atherosclerotic events from **13.4% to 11.3%** (**RR 0.83**) in chronic kidney disease [14]. The CTT individual-participant synthesis found a proportional reduction of about one-fifth in major vascular events per 1 mmol/L LDL-C reduction, with no threshold in the range studied. It adds precision rather than another perturbation [15].

**The "any mechanism" test: CETP inhibition.** CETP inhibitors are the main apparent exception to "LDL lowering by any mechanism reduces events". Dalcetrapib had minimal effect on LDL-C and no effect on events (**HR 1.04**) [28]. It is therefore a negative control for raising HDL-C, not a test of LDL lowering. Evacetrapib lowered LDL-C by **31.1%** at 3 months, but the trial stopped for futility after a median of 26 months (**HR 1.01**) [26]. Anacetrapib lowered non-HDL-C by **17 mg/dL (−18%)** and reduced major coronary events over a median of 4.1 years (**rate ratio 0.91**) [27]. Genetic scores that mimic CETP inhibition added to statin action lowered LDL-C without a proportional fall in apoB, and the corresponding risk reduction followed the smaller apoB change, not the LDL-C change [25]. Indexed by LDL-C, evacetrapib is a genuine exception. Indexed by apoB or non-HDL-C, the CETP results are broadly concordant. Evacetrapib's own apoB change was not verified in this review, and its short duration is an alternative explanation.

This is the canonical convergent pattern the rubric describes: human genetics, several pharmacologic classes, and dose-duration evidence. SC3 reaches Η with High confidence.

### 6. The particle-number discriminator (SC4)

The narrowest phrase in the hypothesis is also the least settled. Two comparisons need to be kept apart. One is particle burden against **LDL-C**, the cholesterol mass per LDL particle. The other is particle burden against **total cholesterol cargo** in all apoB particles, which non-HDL-C measures.

**Against LDL-C, the evidence converges.** Ference et al. compared LPL and LDLR genetic pathways. They found nearly identical CHD effects per 10 mg/dL lower apoB (**OR 0.771 and 0.773**), and after multivariable adjustment the LDL-C and triglyceride estimates were null while apoB remained (**OR 0.761**) [17]. Richardson et al. found that apoB retained a robust effect in multivariable MR (**OR 1.92 per SD**) while the LDL-C estimate reversed and was nonsignificant (**OR 0.85**). They warned that lipoprotein traits are highly correlated [18]. An agnostic Bayesian model-averaging analysis of 30 lipoprotein measures ranked an apoB-only model first and selected apoB in every sensitivity analysis [19]. In CETP-plus-HMGCR genetics, risk followed apoB, and per unit of LDL-C the discordant score was significantly weaker (**OR 0.916 versus 0.831 per unit LDL-C; P < .001**) [25]. The observational discordance cohorts agree. In MESA, among participants whose LDL-C and LDL-P were discordant, only LDL-P was associated with incident cardiovascular disease (**HR 1.45** versus **1.07** for LDL-C) [30]. In Framingham Offspring, LDL-P was more strongly related to future events than LDL-C or non-HDL-C [31]. In statin-treated Copenhagen participants, high apoB and non-HDL-C, but not high LDL-C, were associated with myocardial infarction [33].

**Against total cholesterol cargo, the evidence is split.** On the particle side:
- The MR analyses cited above favor apoB [17-19].
- In UK Biobank and in two outcome-trial populations, only apoB remained associated with myocardial infarction when assessed together with non-HDL-C and triglycerides (**aHR 1.27 per SD**) [32].
- In Framingham Offspring, LDL-P was related more strongly to future events than non-HDL-C [31].
- PROMINENT lowered triglycerides by **26.2%** and remnant cholesterol by **25.6%**, while apoB rose by **4.8%**. There was no reduction in events (**HR 1.03**) [29]. Remnant cholesterol fell without particle number falling, and there was no benefit. A particle model predicts that result, and a remnant-cholesterol-specific model does not. The trial's non-HDL-C change was not verified here, so it does not by itself separate particle number from total cholesterol cargo.

On the other side, the evidence is not consistent with particle count alone:
- Helgadottir et al. used 235 variants, 35% of which had discordant effects on non-HDL-C and apoB. Adding non-HDL-C to a model containing apoB improved the genetically predicted CAD effects (**P = 3.9 × 10⁻⁵**). Adding apoB to a model containing non-HDL-C did not (**P = 0.69**) [20].
- Björnson et al., an independent group, found that the CHD risk per SD of apoB was **1.76** in a lipolysis-pathway (remnant-rich) SNP cluster but **1.33** in a receptor-pathway (LDL-rich) cluster. The result replicated in CARDIoGRAMplusC4D, and remnant cholesterol remained associated with CHD after adjustment for apoB [35].
- The same group estimated that Lp(a) is about **6.6-fold** (95% CI 5.1–8.8) more atherogenic than LDL per particle [36]. In mass terms, an earlier MR estimated that about **101.5 mg/dL** of Lp(a) change carries the coronary risk of a 1 mmol/L LDL-C change [37]. Because Lp(a) particles are relatively few in most people [36], large percentage reductions in Lp(a) correspond to small absolute changes in exposure [37].
- The observational comparisons are also mixed. In the Women's Health Study, discordantly high non-HDL-C (**age-adjusted HR 2.92**) was at least as informative as discordant apoB (**2.48**) or LDL-P (**2.32**), and the authors state that their study was not structured to separate the three [21]. In UK Biobank, apoB, direct LDL-C and non-HDL-C had similar associations (**HR 1.23, 1.20, 1.21 per SD**), and apoB added no discrimination once total and HDL cholesterol were known [34]. In Copenhagen, discordantly high apoB with low non-HDL-C was not associated with myocardial infarction (**HR 0.93**) [33].

These analyses do not cancel by vote. They use correlated traits, overlapping datasets and different instruments. The Björnson and Ference analyses also disagree about whether LDL-receptor variants are clean LDL-specific instruments [17,35]. ApoB counts non-LDL particles, so apoB results do not isolate LDL-P. Three things can be said with confidence. LDL-C as the sole causal quantity is disfavored. Particle number versus total cholesterol cargo is narrowed but not settled. And per-particle risk differs by particle class.

The quoted thresholds give the grade. Rubric 06 Ζ requires "Reproducible intervention effects; Mechanistic narrowing; Evidence not reducible to general sickness, caloric reduction, or developmental artifacts". That is met: genetic natural experiments [17,25], a pharmacological discordance pair [26,27] and PROMINENT [29] narrow the mechanism away from LDL-C and remnant cholesterol as the unique quantities. Rubric 06 Η requires "Robust cross-context intervention evidence; Mechanistic specificity; Strong alignment of … outcomes". The framework's Η definition adds "no major unresolved upstream/downstream contradiction within the claimed scope". Those are not met. No intervention separates particle number from total cholesterol cargo, and two independent groups report contradictions within SC4's own scope [20,35,36]. **SC4 therefore meets Ζ and fails Η, with Moderate confidence.** Under the preregistered rule, this is the binding reason the overall letter is Ζ.

### 7. Context boundaries and negative controls

Two advanced-hemodialysis trials are important boundary evidence. AURORA lowered LDL-C by **43%** but did not significantly reduce its composite of cardiovascular death, nonfatal myocardial infarction or nonfatal stroke (**HR 0.96**), and it had no effect on the individual components [22]. 4D lowered LDL-C by **42%** with a nonsignificant **RR 0.92** for its primary composite. It found a nominally significant reduction in all cardiac events combined (**RR 0.82**) and an excess of fatal stroke (**RR 2.03**) [23]. Both records carry errata, which are flagged in the ledger.

These trials do not reverse LDL causality, but they are more than endpoint artifacts. Much of each composite lies outside the ASCVD scope used here: sudden and other nonatherosclerotic cardiac deaths, and hemorrhagic stroke. AURORA's null for nonfatal myocardial infarction, however, is a within-scope exception. SHARP, which prespecified major atherosclerotic events, found no good evidence that the proportional effect differed between dialysis and non-dialysis patients [14]. The dialysis results map a late-disease boundary for prediction. Advanced dialysis lies outside SC3's stated scope, but inside SC5's CKD scope, so AURORA's in-scope null keeps SC5 at Η rather than Θ.

Residual risk in positive trials has the same implication. Lowering LDL changes the probability and timing of events. It does not remove smoking, blood pressure, diabetes, inflammation [39], plaque rupture or erosion, thrombosis, lipoprotein(a) or remnant particles as causes.

### 8. Cap and Β/Γ/Δ boundary adjudication

No association-only or biomarker-only cap binds the convergent evidence. The field-valid-equivalents text applied is:

> "Mendelian randomization on the exposure itself, multiple pharmacological classes acting on the same node with concordant endpoint effects, dose-response across independent populations, and natural experiments of lifelong exposure."

Caps were applied to single evidence classes and not re-applied to convergent sets (rubric 06: "Convergence is not re-capped"; pitfall 21):
- **SC1.** The rubric-06 Γ hard rule "Single-species or single-strain evidence" binds the retention perturbation [1] in isolation. It does not bind SC1, because human genetics, imaging and trials supply independent phenotype-relevant lines.
- **SC4.** Universal cap 2 ("Pure association evidence — however large, however replicated — cannot exceed Β") binds the discordance cohorts [21,30-34] in isolation. It does not bind SC4 as a whole, which also rests on MR [17-20,25,35,36] and on perturbation evidence [26,27,29].
- **SC3.** The Γ hard rule "Biomarker-only improvement without organismal benefit" is not triggered, because the trials measured hard events.

The Β/Γ/Δ boundary is crossed for every subclaim by a perturbation with a positive, phenotype-relevant result. For SC4 that perturbation is the discordant-pathway genetics [17,25].

### 9. Threshold testing upward

Η requires **"Robust cross-context intervention evidence," "Mechanistic specificity,"** and **"Strong alignment of lifespan/healthspan outcomes,"** here read as ASCVD outcomes. SC1, SC3 and SC5 meet Η. SC2 reaches Θ, because lifelong genetic direction and the duration gradient are highly reproducible with few exceptions. SC4 meets the Ζ requirements but not Η (§6).

Θ requires **"high reproducibility with minimal contextual contradiction."** SC3 fails Θ because the evacetrapib null [26] is an unresolved exception within its scope. The particle-based explanation for it is plausible, but it is not demonstrated in that trial's own data. SC5 fails Θ because AURORA's null for nonfatal myocardial infarction [22] falls within its CKD scope. SC1 fails Θ because the retention step rests on one perturbation. Ω fails for the claim as a whole, because LDL is neither necessary nor sufficient for every plaque or event.

### 10. Independence audit

For LDL-class causality, seven lines remain after grouping by shared mechanism and dataset:
1. Human lifelong genetics [2-6,38].
2. HMGCR trials [7,8].
3. NPC1L1 inhibition [9].
4. PCSK9-antibody trials [10,11,13].
5. ATP-citrate-lyase inhibition [12].
6. Direct retention and plaque trajectory [1,16].
7. CETP inhibition [25-28], added in this edition as a separate lipid-transfer mechanism.

For SC4, three lines are recorded:
- **(a) Multivariable and pathway-contrast MR** [17-20,25,35,36]. Shared failure modes are instrument choice, pleiotropy and collinearity. This line is internally contradictory.
- **(b) Observational discordance cohorts** [21,30-34]. The shared failure mode is confounding by insulin resistance and adiposity, which is documented in [21] and [30].
- **(c) Pharmacological discordance** [26,27,29]. The shared failure mode is drug-specific off-target effects.

One systematic error could not produce all three lines. They converge against LDL-C as the unique quantity, but only line (a) addresses particle count versus total cargo directly, and it disagrees internally. The score does not count every trial, every MR paper, the CTT meta-analysis [15] or the consensus statement [24] as new causal lines.

### 11. Falsification and pitfall audit

Each rubric-06 falsification pattern was evaluated. None is triggered within the scope of the subclaim it would apply to, so no falsification downgrade is applied.
- "Late-life intervention fails despite strong early-life effects." The advanced-dialysis nulls lie outside SC3's stated scope. For SC5 they are a mapped boundary (pitfall 16), not a falsification.
- "Necessity fails when the proposed pathway is removed." Residual events test necessity, which the claim does not assert. This pattern blocks Ω only.
- "Pathway manipulation changes biomarkers but not lifespan/healthspan." This does not apply to LDL lowering in general, because hard events fell. It is echoed by evacetrapib only under LDL-C indexing [26].

The SC4 contradiction is handled at the Η threshold (§6) rather than as a falsification pattern. Rubric 06 lists no pattern for it.

Pitfalls surfaced:
- LDL-C treated as proof of particle number.
- apoB silently equated with LDL-P.
- Equal per-particle atherogenicity assumed across apoB classes.
- Reversibility mislabeled as mechanism.
- Drug classes over-counted.
- Overlapping MR datasets treated as independent.
- Observational discordance cited as causal discrimination (pitfall 2).
- Mixed cardiovascular endpoints treated as purely atherosclerotic.
- "Primary driver" inflated into sole necessity or sufficiency.

### 12. Steelman

**One rung above (Η).** Almost every piece of evidence in this review points the same way. Particle-indexed measures consistently outperform LDL-C. Four MR analyses, three cohort studies and a CETP genetics-plus-trial pair agree on that point. PROMINENT shows that lowering remnant cholesterol without lowering particle number does not reduce events. And the LDL-class causal subclaims sit at Η or Θ. On this view the remaining dispute is a refinement: which weighting of apoB particles is best.

**One rung below (Ε).** The distinctive claim, particle number over total cholesterol, has no clean intervention. The MR evidence for it is contradicted by two independent groups. The observational evidence cannot separate apoB from non-HDL-C. And particle type, not only particle count, changes per-particle risk.

**Which is rejected, and the blocker.** Η is rejected. The blocker is the absence of any intervention or replicated natural experiment that separates particle number from **total** cholesterol cargo, together with contradictions inside SC4's scope from independent groups [20,35,36]. That is the framework's "major unresolved … contradiction", and rubric 06's missing "Mechanistic specificity". Ε is also rejected. The alternative models are genuinely narrowed: LDL-C as the unique quantity is excluded by independent genetic, pharmacological and observational lines [17-19,25-27,30-33], and PROMINENT is an intervention result that a remnant-cholesterol-specific model does not predict [29]. This is multi-line convergence beyond a single context.

### 13. Final adjudication

**Overall: Ζ — Multiple independent causal lines converge. Confidence: Moderate.** Elevated cumulative exposure to LDL-class apoB particles is a primary upstream cause of human ASCVD, and on their own the subclaims establishing that grade at Η (SC1, SC3, SC5) and Θ (SC2). The submitted wording adds that particle **concentration**, meaning number, is the causal quantity. Against LDL-C that is well supported. Against total cholesterol cargo it is contested, and per-particle risk varies by particle class. The preregistered weakest-link rule therefore returns Ζ. Scope-matched aggregation would return Η. Primary-subclaim aggregation returns Ζ if SC4 is designated primary and Η if SC3 is.

### 14. Role of mechanism

The role is **primary driver**, and the derivation matched at **step 3**. Step 1 does not match: disease does not arise or progress independently of LDL-class exposure, so it is not a downstream consequence. Step 2 does not match: causal dependency is demonstrated, so it is not an epiphenomenon. Step 3 matches on all three conditions:
- **Removal prevents or delays.** Lowering exposure prevents or delays hard atherosclerotic outcomes [7-14].
- **Engagement initiates.** Lifelong elevation from FH mutations increases risk at any observed LDL-C level, with greater cumulative exposure [38], and lifelong lowering protects [2-5].
- **Rate-limiting.** Effects scale with absolute exposure and duration [3,15].

The role applies to ASCVD through atherogenesis, at the level of LDL-class apoB particles. It does not certify particle number rather than cholesterol cargo as the rate-limiting property. Inflammation [39], plaque disruption, thrombosis and other risk factors help determine which plaque becomes a clinical event, and when. The tournament was not used as a tiebreaker.

### 15. Momentum, stability, and divergence

**Momentum — Mechanistic only.** Two outcome trials are pending, and neither is designed to separate particle number from total cholesterol cargo. [PREVAIL (NCT05202509)](https://clinicaltrials.gov/study/NCT05202509) tests the CETP inhibitor obicetrapib. It is active, not recruiting, with 9,541 participants and estimated primary completion in November 2026. It may help separate apoB from LDL-C if achieved apoB and non-HDL-C are reported. [ORION-4 (NCT03705234)](https://clinicaltrials.gov/study/NCT03705234) tests inclisiran. It is active, not recruiting, with 16,124 participants and estimated primary completion in October 2026, and it will add a lowering modality.

**Stability — Stable.** No named pending result is expected to cross SC4's particle-versus-total-cargo boundary. [Lp(a)HORIZON (NCT04023552)](https://clinicaltrials.gov/study/NCT04023552) is completed, and the sponsor's topline announcement (2026-09-04) says the primary endpoint was not met. That announcement is not peer reviewed and is not used in scoring. Its full report bears on the broader apoB model (M2) rather than on the nominal wording. PREVAIL could shift SC4 confidence in either direction.

**Divergence — Mild overstatement.** Read as the field usually presents it, the sentence asserts an established causal driver without claiming necessity or sufficiency, which implies Η. The assessed level is Ζ, a gap of +1 rung. The overstatement lies entirely in the particle-number reading. Read as "LDL-class particles cause ASCVD", the wording is aligned with the evidence.

## PROPOSED DECISIVE EXPERIMENTS

1. **Orthogonal particle-versus-cargo discriminator.** Use genetic instruments or a randomized therapy combination that produces prespecified, opposite changes in apoB (or LDL-P) and non-HDL cholesterol. Measure both contrasts, and follow hard ASCVD events and plaque imaging. If risk consistently tracks cholesterol cargo once particle number is controlled, particle primacy is falsified. If risk tracks particle number once cargo is controlled, it is supported. Particle class (LDL, remnant, Lp(a)) should be recorded as a covariate, because per-particle risk differs by class [35,36].

2. **Human arterial flux and retention mediation.** Use labeled autologous particle tracers and validated arterial-wall imaging before and after selective lowering, partitioning LDL, remnants and lipoprotein(a). Test whether the change in arterial particle flux and retention mediates the plaque response independently of cholesterol cargo.

3. **Prespecified model comparison in pending CETP-inhibitor outcomes.** Before unblinding, register the event reduction that LDL-C, apoB and non-HDL-C models each predict from the achieved changes, then compare with the observed result. Replicate in a second trial or cohort.

## KNOWLEDGE GAPS

- No hard-outcome intervention cleanly separates particle number from total cholesterol cargo.
- Per-particle atherogenicity differs by class (remnant and Lp(a) versus LDL) [35,36], but the mechanism and the correct weighting are unresolved. Particle count alone is therefore an incomplete exposure metric.
- MR analyses disagree partly because of instrument choice. LDL-receptor variants also affect remnant clearance [35], and genome-wide sets carry pleiotropy [20].
- The apoB change behind the evacetrapib null was not established here, and its short duration remains an alternative explanation [26].
- A single plasma concentration is an imperfect proxy for cumulative arterial flux, residence, modification and retention.
- The modifiable fraction of atherosclerotic risk in advanced dialysis remains difficult to predict [14,22,23].
- How LDL burden interacts quantitatively with plaque inflammation [39], rupture or erosion, and thrombosis remains incompletely modelled.

## LIMITATIONS OF THIS REVIEW

No cardiovascular-specific rubric was supplied. The rubric-06 adaptation and the field-valid-equivalents quotation are disclosed. Fourteen of 39 records were read in complete full text across the authoring and red-team passes. The remainder were checked at PubMed abstract or metadata depth, and no claim that needs text beyond the abstract is attributed to them. Three records have PubMed-listed errata, which are flagged. None of them is the sole support for any score. Several MR studies share UK Biobank or CARDIoGRAMplusC4D data. The sponsor topline for Lp(a)HORIZON is named for currency only and is not scored. Citation-context (Scite) counts were not used.

## CALIBRATION STATUS

This is **not** a calibration entry. No independently authored sealed expected band was supplied. The hypothesis, evidence ledger and score were derived without a calibration target.

## VALIDATION NOTE

The companion validator report records the HSEF structural checks for this edition (validator 2.2). A pass confirms canonical level and descriptor use, aggregation arithmetic, source and claim links, access depth, status flags, independence structure and provenance counts. It does not constitute independent scientific replication, inter-reviewer calibration or external endorsement. Nor does it confirm that identifiers resolve to the cited works. That was checked separately in the red-team audit.

## RED-TEAM SYNTHESIS NOTE

This edition is the synthesis of a cross-model red-team pass on 2026-09-30. The review was authored by a different model family. Full details are in `redteam/REDTEAM_REPORT.md` (findings RT-01 to RT-32) and `redteam/SYNTHESIS_CHANGE_LOG.md`.

- **Grade: Η → Ζ; confidence High → Moderate.** The preregistration makes the particle-number subclaim (SC4) a required conjunct. The original graded it Η while conceding that no intervention isolates particle number and that its evidence conflicts. Those conditions fail the Η definition and rubric text quoted in §6. SC4 now meets Ζ. The LDL-class causal subclaims are unchanged (SC1, SC3, SC5 at Η; SC2 at Θ), and scope-matched aggregation would return Η.
- **Evidence added.** The review now includes CETP-inhibitor genetics and trials [25-28], PROMINENT [29], the MESA, Framingham, UK Biobank and Copenhagen discordance cohorts [30-34], and per-particle heterogeneity for remnants and Lp(a) [35-37], which is a second, independent contradiction to particle count as the whole causal quantity. It also adds FH sequencing [38] and CANTOS [39].
- **Corrections.** The Women's Health Study discordance result [21] is no longer cited as favoring particles over cholesterol, because its non-HDL-C result was at least as strong. Reference 13's first author is corrected. The CLEAR Outcomes and VESALIUS-CV descriptions are corrected. The FH odds ratio of 13.2 [6] is flagged as inflated by its diagnostic criteria. A one-sided affiliation remark is removed. The tournament levels, divergence, aggregation alternatives and falsification-pattern coding are re-derived.
- **Citations.** All 24 original identifiers resolve to the stated works, and one first author was corrected. Fifteen verified references were added.

## ADVERSARIAL REVIEW AND APPROVAL

**Authoring model:** Sol 6.1 (this edition is the published synthesis of a cross-model red-team pass on 2026-09-30). **Adversarial reviewer for this round:** Claude Opus 5.5 (Anthropic). **Date:** 2026-10-14.

**Scope of checks.** PMIDs verified 39/39 against PubMed on title, first author, journal, year and DOI; no retraction or expression of concern on any row. Three records carry PubMed-listed errata (references 6, 22, 23); each is attached to its parent row rather than given a row of its own, which is what the ledger discipline requires, and none is the sole support for a score. Load-bearing numbers spot-checked digit for digit against the retrieved abstracts: 16 (the PCSK9 nonsense-variant effects in both ancestries, the nine-variant lifelong-exposure synthesis, the NPC1L1 odds ratio, the PCSK9/HMGCR score equivalence, the familial-hypercholesterolaemia sequencing odds ratios, WOSCOPS, 4S, IMPROVE-IT, FOURIER, ODYSSEY OUTCOMES, VESALIUS-CV, CLEAR Outcomes, SHARP, GLAGOV, the LPL-versus-LDLR apoB equivalence, and the one Mendelian-randomization analysis that favours cholesterol cargo). Every one matched. Internal counts recomputed from the score card: 39 citation-ledger rows, 37 Tier A, 33 scored evidence rows, 28 claims, access depths 14 complete full text / 25 abstract-metadata, and no blank status flag. Scoring logic checked against the HSEF framework and SKILL.md: the conjunctive-claim rule, the pre-registered weakest-link aggregation across SC1-SC4 and both reported alternatives, the cap audit and the Β/Γ/Δ boundary, the independence audit for the seven LDL-class lines and the three SC4 lines, the upward threshold tests at Η and Θ, the role derivation at step 3, and the M/S/D artifacts. Validator re-run: PASS, 0 fail, 0 warn, 33 pass under the v5.4 validator (30 pass under the archived v5.3 validator).

### Findings

| ID | Severity | Location | Original to corrected | Evidence |
|---|---|---|---|---|
| E1 | Major | "Consensus consistency validation — 9 October 2026" block, frozen-inventory line and conclusion; the same two sentences in `consensus-addendum.md` | "Frozen scorecard inventory: 33 evidence rows; 28 claims" and "original sources/claims stay at 33/28" corrected to name both counts: 39 citation-ledger sources, of which 33 are scored evidence rows, and 28 claims | The figure 33 is the length of the score card's `evidence_inventory`, but the sentences present it as the review's source count, which contradicts the header ("39 sources · 28 claims"), AT A GLANCE ("39 sources"), the score card's `provenance_header.sources_total: 39`, the 39-row `citation_ledger`, and the review's own limitations section ("Fourteen of 39 records"). The 9 October revision introduced the clause, so this is a regression rather than an inherited error, and it is the kind of count mismatch a reader checking artifacts against each other would hit first. Scoped rather than deleted, so both inventories stay traceable. No grade, subclaim or claim count moves. |

### Rejected findings (checked, no change)

- **Headline Ζ.** I argued both directions and could not move it. The registered sentence is conjunctive — LDL-class particles cause ASCVD, lowering exposure reduces events, **and** particle burden rather than cholesterol mass is the operative quantity when the two diverge — and the framework is explicit that a conjunctive hypothesis is scored on the conjunction, not on its easier half. Weakest-link across SC1-SC4 was pre-registered on 2026-09-29, before outcome adjudication, and returns Ζ from SC4. The Η reading is reported rather than substituted, and the arithmetic for all three rules is stated. Η for SC4 is refused on rubric text and on a named contradiction from independent groups; Ε is refused because the alternatives are genuinely narrowed. That is the correct handling, and the one-rung spread is attributed to the particle-number conjunct rather than hidden inside a single letter.
- **Confidence Moderate with High on three subclaims.** Checked against the standing concern that a Moderate band can be set from reviewer access depth rather than from the evidence. **Not the case here:** the rationale cites directionally inconsistent evidence from independent groups and heterogeneous assays (NMR LDL-P, immunoassay apoB, calculated non-HDL-C and remnant cholesterol), which is the framework's Moderate text almost verbatim, and access depth is reported separately in the limitations section. The overall band tracks the binding subclaim, which is the coherent pairing.
- **"Best model" naming a non-nominal model.** Checked against the renderer bug in which the field prints the nominal. **Absent:** the score card's first model is the broader apoB-particle retention model with `rank_in_tournament: 1` and `is_nominal: false`, the nominal LDL-particle model is ranked 2, and AT A GLANCE records "Nominal-model rank: 2 of 5". The prose names the correct rank-1 rival.
- **Momentum "Mechanistic only" with two active outcome trials named.** Rejected as a defect: the artifact names PREVAIL and ORION-4 with registry identifiers and enrolment, and states why they do not make this "Decisive work underway" — neither is designed to separate particle number from total cholesterol cargo, which is the boundary the binding subclaim turns on. That is the framework's definition applied, not boilerplate.
- **The Lp(a)HORIZON sponsor topline.** Named for currency, explicitly flagged as not peer reviewed and excluded from scoring, and attributed to the broader apoB model rather than to the graded wording. Correct handling of a dated non-peer-reviewed fact.
- **"Earlier approval" provenance clause.** The spliced block's closing sentence refers to an earlier approval. Checked before changing, because this clause is false boilerplate elsewhere in the corpus: here an earlier independent adversarial pass does exist — the 2026-09-30 cross-model red-team pass recorded in this document's own synthesis note and in `reviews/ldl-ascvd/redteam/redteam-pass-2026-09-30/`, with 25 of 32 findings accepted and the grade moved Η to Ζ. The clause is true and was left alone.
- **Reviewer-notes contradiction with the spliced Consensus block.** Checked for the systematic 9 October artifact. **Absent:** this document contains no assertion that no Consensus work was performed, and the spliced block states its own limits.
- **Collective-author reference defect.** Checked specifically. References 4, 7, 8, 24 and 27 all carry consortium or group authorship and all render with the group named and no leading comma.
- **Rendering defects.** None found; this edition carries none of the missing-space, empty-DOI or "ErratumIn" artifacts seen elsewhere in the corpus.
- **Score card `review_metadata.reviewer`.** Left as issued; corpus-wide house provenance.

### Residual limitations

Access depth is the main constraint, and the review states it accurately: 14 of 39 records were read in complete full text and 25 at abstract or metadata depth, with no claim requiring text beyond the abstract attributed to the latter. Two items could not be re-verified in this pass and are disclosed rather than corrected: the apoB change behind the evacetrapib null, which the review itself lists as an open knowledge gap and which is load-bearing for reading the CETP results as concordant; and the 2026-09-04 Lp(a)HORIZON topline, which is a sponsor announcement with no peer-reviewed report and is correctly excluded from scoring. The rubric adaptation is the review's own largest caveat: no cardiovascular or lipidology rubric was supplied, so rubric 06's lifespan and healthspan language is read as hard ASCVD events and plaque trajectory. That adaptation is disclosed and does not let biomarker-only evidence cross the causal boundary, but it is what a second reviewer is most likely to contest.

**Approval:** Following adversarial review and the correction listed above, the revised document is approved. — Claude Opus 5.5 (Anthropic), 2026-10-14.

## REFERENCES

1. Skålén et al. Subendothelial retention of atherogenic lipoproteins in early atherosclerosis. Nature. 2002;417:750-4. · [PMID 12066187](https://pubmed.ncbi.nlm.nih.gov/12066187/) · [DOI](https://doi.org/10.1038/nature00804)
2. Cohen et al. Sequence variations in PCSK9, low LDL, and protection against coronary heart disease. N Engl J Med. 2006;354:1264-72. · [PMID 16554528](https://pubmed.ncbi.nlm.nih.gov/16554528/) · [DOI](https://doi.org/10.1056/NEJMoa054013)
3. Ference et al. Effect of long-term exposure to lower low-density lipoprotein cholesterol beginning early in life on the risk of coronary heart disease: a Mendelian randomization analysis. J Am Coll Cardiol. 2012;60:2631-9. · [PMID 23083789](https://pubmed.ncbi.nlm.nih.gov/23083789/) · [DOI](https://doi.org/10.1016/j.jacc.2012.09.017)
4. Myocardial Infarction Genetics Consortium Investigators; Stitziel et al. Inactivating mutations in NPC1L1 and protection from coronary heart disease. N Engl J Med. 2014;371:2072-82. · [PMID 25390462](https://pubmed.ncbi.nlm.nih.gov/25390462/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC4335708/) · [DOI](https://doi.org/10.1056/NEJMoa1405386)
5. Ference et al. Variation in PCSK9 and HMGCR and risk of cardiovascular disease and diabetes. N Engl J Med. 2016;375:2144-53. · [PMID 27959767](https://pubmed.ncbi.nlm.nih.gov/27959767/) · [DOI](https://doi.org/10.1056/NEJMoa1604304)
6. Benn et al. Familial hypercholesterolemia in the Danish general population: prevalence, coronary artery disease, and cholesterol-lowering medication. J Clin Endocrinol Metab. 2012;97:3956-64. Erratum: J Clin Endocrinol Metab. 2014;99:4758-9. · [PMID 22893714](https://pubmed.ncbi.nlm.nih.gov/22893714/) · [DOI](https://doi.org/10.1210/jc.2012-1563)
7. Shepherd et al., West of Scotland Coronary Prevention Study Group. Prevention of coronary heart disease with pravastatin in men with hypercholesterolemia. N Engl J Med. 1995;333:1301-7. · [PMID 7566020](https://pubmed.ncbi.nlm.nih.gov/7566020/) · [DOI](https://doi.org/10.1056/NEJM199511163332001)
8. Scandinavian Simvastatin Survival Study Group. Randomised trial of cholesterol lowering in 4444 patients with coronary heart disease: the Scandinavian Simvastatin Survival Study (4S). Lancet. 1994;344:1383-9. · [PMID 7968073](https://pubmed.ncbi.nlm.nih.gov/7968073/)
9. Cannon et al. Ezetimibe added to statin therapy after acute coronary syndromes. N Engl J Med. 2015;372:2387-97. · [PMID 26039521](https://pubmed.ncbi.nlm.nih.gov/26039521/) · [DOI](https://doi.org/10.1056/NEJMoa1410489)
10. Sabatine et al. Evolocumab and clinical outcomes in patients with cardiovascular disease. N Engl J Med. 2017;376:1713-22. · [PMID 28304224](https://pubmed.ncbi.nlm.nih.gov/28304224/) · [DOI](https://doi.org/10.1056/NEJMoa1615664)
11. Schwartz et al. Alirocumab and cardiovascular outcomes after acute coronary syndrome. N Engl J Med. 2018;379:2097-107. · [PMID 30403574](https://pubmed.ncbi.nlm.nih.gov/30403574/) · [DOI](https://doi.org/10.1056/NEJMoa1801174)
12. Nissen et al. Bempedoic acid and cardiovascular outcomes in statin-intolerant patients. N Engl J Med. 2023;388:1353-64. · [PMID 36876740](https://pubmed.ncbi.nlm.nih.gov/36876740/) · [DOI](https://doi.org/10.1056/NEJMoa2215024)
13. Bohula et al. Evolocumab in patients without a previous myocardial infarction or stroke. N Engl J Med. 2026;394:117-27 (epub 2025). · [PMID 41211925](https://pubmed.ncbi.nlm.nih.gov/41211925/) · [DOI](https://doi.org/10.1056/NEJMoa2514428)
14. Baigent et al. The effects of lowering LDL cholesterol with simvastatin plus ezetimibe in patients with chronic kidney disease (Study of Heart and Renal Protection): a randomised placebo-controlled trial. Lancet. 2011;377:2181-92. · [PMID 21663949](https://pubmed.ncbi.nlm.nih.gov/21663949/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC3145073/) · [DOI](https://doi.org/10.1016/S0140-6736(11)60739-3)
15. Cholesterol Treatment Trialists' (CTT) Collaboration. Efficacy and safety of more intensive lowering of LDL cholesterol: a meta-analysis of data from 170,000 participants in 26 randomised trials. Lancet. 2010;376:1670-81. · [PMID 21067804](https://pubmed.ncbi.nlm.nih.gov/21067804/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC2988224/) · [DOI](https://doi.org/10.1016/S0140-6736(10)61350-5)
16. Nicholls et al. Effect of evolocumab on progression of coronary disease in statin-treated patients: the GLAGOV randomized clinical trial. JAMA. 2016;316:2373-84. · [PMID 27846344](https://pubmed.ncbi.nlm.nih.gov/27846344/) · [DOI](https://doi.org/10.1001/jama.2016.16951)
17. Ference et al. Association of triglyceride-lowering LPL variants and LDL-C-lowering LDLR variants with risk of coronary heart disease. JAMA. 2019;321:364-73. · [PMID 30694319](https://pubmed.ncbi.nlm.nih.gov/30694319/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC6439767/) · [DOI](https://doi.org/10.1001/jama.2018.20045)
18. Richardson et al. Evaluating the relationship between circulating lipoprotein lipids and apolipoproteins with risk of coronary heart disease: a multivariable Mendelian randomisation analysis. PLoS Med. 2020;17:e1003062. · [PMID 32203549](https://pubmed.ncbi.nlm.nih.gov/32203549/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC7089422/) · [DOI](https://doi.org/10.1371/journal.pmed.1003062)
19. Zuber et al. High-throughput multivariable Mendelian randomization analysis prioritizes apolipoprotein B as key lipid risk factor for coronary artery disease. Int J Epidemiol. 2021;50:893-901. · [PMID 33130851](https://pubmed.ncbi.nlm.nih.gov/33130851/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC8271202/) · [DOI](https://doi.org/10.1093/ije/dyaa216)
20. Helgadottir et al. Cholesterol not particle concentration mediates the atherogenic risk conferred by apolipoprotein B particles: a Mendelian randomization analysis. Eur J Prev Cardiol. 2022;29:2374-85. · [PMID 36125206](https://pubmed.ncbi.nlm.nih.gov/36125206/) · [DOI](https://doi.org/10.1093/eurjpc/zwac219)
21. Mora et al. Discordance of low-density lipoprotein (LDL) cholesterol with alternative LDL-related measures and future coronary events. Circulation. 2014;129:553-61. · [PMID 24345402](https://pubmed.ncbi.nlm.nih.gov/24345402/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC4501252/) · [DOI](https://doi.org/10.1161/CIRCULATIONAHA.113.005873)
22. Fellström et al. Rosuvastatin and cardiovascular events in patients undergoing hemodialysis. N Engl J Med. 2009;360:1395-407. Erratum: N Engl J Med. 2010;362:1450. · [PMID 19332456](https://pubmed.ncbi.nlm.nih.gov/19332456/) · [DOI](https://doi.org/10.1056/NEJMoa0810177)
23. Wanner et al. Atorvastatin in patients with type 2 diabetes mellitus undergoing hemodialysis. N Engl J Med. 2005;353:238-48. Erratum: N Engl J Med. 2005;353:1640. · [PMID 16034009](https://pubmed.ncbi.nlm.nih.gov/16034009/) · [DOI](https://doi.org/10.1056/NEJMoa043545)
24. Ference et al., EAS Consensus Panel. Low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies. A consensus statement from the European Atherosclerosis Society Consensus Panel. Eur Heart J. 2017;38:2459-72. · [PMID 28444290](https://pubmed.ncbi.nlm.nih.gov/28444290/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC5837225/) · [DOI](https://doi.org/10.1093/eurheartj/ehx144)
25. Ference et al. Association of genetic variants related to CETP inhibitors and statins with lipoprotein levels and cardiovascular risk. JAMA. 2017;318:947-56. · [PMID 28846118](https://pubmed.ncbi.nlm.nih.gov/28846118/) · [PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC5710502/) · [DOI](https://doi.org/10.1001/jama.2017.11467)
26. Lincoff et al., ACCELERATE Investigators. Evacetrapib and cardiovascular outcomes in high-risk vascular disease. N Engl J Med. 2017;376:1933-42. · [PMID 28514624](https://pubmed.ncbi.nlm.nih.gov/28514624/) · [DOI](https://doi.org/10.1056/NEJMoa1609581)
27. HPS3/TIMI55–REVEAL Collaborative Group; Bowman et al. Effects of anacetrapib in patients with atherosclerotic vascular disease. N Engl J Med. 2017;377:1217-27. · [PMID 28847206](https://pubmed.ncbi.nlm.nih.gov/28847206/) · [DOI](https://doi.org/10.1056/NEJMoa1706444)
28. Schwartz et al., dal-OUTCOMES Investigators. Effects of dalcetrapib in patients with a recent acute coronary syndrome. N Engl J Med. 2012;367:2089-99. · [PMID 23126252](https://pubmed.ncbi.nlm.nih.gov/23126252/) · [DOI](https://doi.org/10.1056/NEJMoa1206797)
29. Das Pradhan et al., PROMINENT Investigators. Triglyceride lowering with pemafibrate to reduce cardiovascular risk. N Engl J Med. 2022;387:1923-34. · [PMID 36342113](https://pubmed.ncbi.nlm.nih.gov/36342113/) · [DOI](https://doi.org/10.1056/NEJMoa2210645)
30. Otvos et al. Clinical implications of discordance between low-density lipoprotein cholesterol and particle number. J Clin Lipidol. 2011;5:105-13. · [PMID 21392724](https://pubmed.ncbi.nlm.nih.gov/21392724/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC3070150/) · [DOI](https://doi.org/10.1016/j.jacl.2011.02.001)
31. Cromwell et al. LDL particle number and risk of future cardiovascular disease in the Framingham Offspring Study — implications for LDL management. J Clin Lipidol. 2007;1:583-92. · [PMID 19657464](https://pubmed.ncbi.nlm.nih.gov/19657464/) · [DOI](https://doi.org/10.1016/j.jacl.2007.10.001)
32. Marston et al. Association of apolipoprotein B-containing lipoproteins and risk of myocardial infarction in individuals with and without atherosclerosis: distinguishing between particle concentration, type, and content. JAMA Cardiol. 2022;7:250-6. · [PMID 34773460](https://pubmed.ncbi.nlm.nih.gov/34773460/) · [PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC8590731/) · [DOI](https://doi.org/10.1001/jamacardio.2021.5083)
33. Johannesen et al. Apolipoprotein B and non-HDL cholesterol better reflect residual risk than LDL cholesterol in statin-treated patients. J Am Coll Cardiol. 2021;77:1439-50. · [PMID 33736827](https://pubmed.ncbi.nlm.nih.gov/33736827/) · [DOI](https://doi.org/10.1016/j.jacc.2021.01.027)
34. Welsh et al. Comparison of conventional lipoprotein tests and apolipoproteins in the prediction of cardiovascular disease. Circulation. 2019;140:542-52. · [PMID 31216866](https://pubmed.ncbi.nlm.nih.gov/31216866/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC6693929/) · [DOI](https://doi.org/10.1161/CIRCULATIONAHA.119.041149)
35. Björnson et al. Triglyceride-rich lipoprotein remnants, low-density lipoproteins, and risk of coronary heart disease: a UK Biobank study. Eur Heart J. 2023;44:4186-95. · [PMID 37358553](https://pubmed.ncbi.nlm.nih.gov/37358553/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC10576615/) · [DOI](https://doi.org/10.1093/eurheartj/ehad337)
36. Björnson et al. Lipoprotein(a) is markedly more atherogenic than LDL: an apolipoprotein B-based genetic analysis. J Am Coll Cardiol. 2024;83:385-95. · [PMID 38233012](https://pubmed.ncbi.nlm.nih.gov/38233012/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC7616706/) · [DOI](https://doi.org/10.1016/j.jacc.2023.10.039)
37. Burgess et al. Association of LPA variants with risk of coronary disease and the implications for lipoprotein(a)-lowering therapies: a Mendelian randomization analysis. JAMA Cardiol. 2018;3:619-27. · [PMID 29926099](https://pubmed.ncbi.nlm.nih.gov/29926099/) · [PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC6481553/) · [DOI](https://doi.org/10.1001/jamacardio.2018.1470)
38. Khera et al. Diagnostic yield and clinical utility of sequencing familial hypercholesterolemia genes in patients with severe hypercholesterolemia. J Am Coll Cardiol. 2016;67:2578-89. · [PMID 27050191](https://pubmed.ncbi.nlm.nih.gov/27050191/) · [PMC full text](https://pmc.ncbi.nlm.nih.gov/articles/PMC5405769/) · [DOI](https://doi.org/10.1016/j.jacc.2016.03.520)
39. Ridker et al. Antiinflammatory therapy with canakinumab for atherosclerotic disease. N Engl J Med. 2017;377:1119-31. · [PMID 28845751](https://pubmed.ncbi.nlm.nih.gov/28845751/) · [DOI](https://doi.org/10.1056/NEJMoa1707914)