Cholesterol, statins, and the LDL hypothesis

A critical synthesis of biology, evidence, and dispute (1953–2026)

Commissioned by: Anthony Booth, Dubai Compiled by: Multi-agent research orchestration with citation audit Date: May 2026 Edition: 1.0 Format: Markdown with footnoted Vancouver-style citations and companion HTML reading view Stance: Steelmanned both sides, landing at a calibrated position

Abstract

This document is a post-graduate-level critical synthesis of the cholesterol hypothesis of atherosclerotic cardiovascular disease and of HMG-CoA reductase inhibitors ("statins"), commissioned to equip a non-clinician with the substantive evidence base needed to debate the topic at depth with an industry-adjacent interlocutor. It walks the biology end-to-end (mevalonate pathway, lipoprotein particles, atherogenesis, the failed HDL hypothesis), the literature arc from the diet-heart era through the 2024–2026 frontier (STAREE, PREVENTABLE, ORION-4, the Lp(a) RNA therapeutics), and the structure of contemporary dispute (consensus camp, heterodox camp, methodological middle, Mendelian-randomisation adjudication, industry conflicts of interest). It then aggregates the harms ledger (SAMS and the nocebo question, diabetes signal, cognition, hepatic/renal/rhabdo, asymmetric AE capture), works through special populations where the evidence is genuinely population-specific (primary vs secondary prevention, women, the over-75s, familial hypercholesterolaemia, diabetes/CKD/post-ACS), adjudicates the eight most common myths in either direction, and lands a calibrated position with explicit residual uncertainty.

The headline finding: in secondary prevention and familial hypercholesterolaemia, the evidence is overwhelming and the dispute is closed; in high-risk primary prevention the evidence is strong; in low-risk primary prevention, in women without prior events, and in the over-75s pending STAREE and PREVENTABLE, genuine uncertainty remains and reasonable people can hold different views on absolute benefit-to-harm ratio. Industry capture has demonstrably distorted specific trials and analyses, but the core LDL-causality finding survives the conflict-of-interest critique because it is confirmed by Mendelian randomisation, investigator-initiated trials, post-patent generic data, and four mechanistically distinct drug classes (statins, ezetimibe, PCSK9 inhibitors, bempedoic acid).

Reader's note

This document is not medical advice. It is a research synthesis intended to support informed conversation. Clinical decisions belong with a clinician who knows the patient. Where this document expresses a calibrated position, it does so transparently and with declared evidence weight; readers are invited to weight differently.

Evidence hierarchy (declared at the outset)

1. Individual-patient-data (IPD) meta-analyses of pre-registered RCTs — top tier, with the caveat that the CTT IPD remains unavailable for independent reanalysis, which is a legitimate epistemic complaint.
2. Pre-registered RCTs with publicly accessible protocols and statistical analysis plans.
3. Mendelian randomisation for causality questions (LDL-on-CHD, PCSK9 LoF, HMGCR variant studies).
4. Prospective cohort studies for risk-marker mapping and rare-harm signal detection.
5. Mechanistic and laboratory work for biological plausibility.
6. Expert opinion and guidelines — informative but lowest standalone weight.

Industry-sponsored trials with restricted IPD access are down-weighted relative to investigator-initiated or post-patent generic trials of the same molecule. Evidence is up-weighted where MR and RCT converge on the same direction and approximate magnitude.

Falsification criteria (stated before reviewing evidence, not after)

• The LDL-causal hypothesis would be substantially weakened by: (a) a well-powered Mendelian randomisation showing no causal effect of lifelong genetic LDL variants on CHD; (b) a properly-controlled non-statin LDL-lowering intervention reducing LDL but not events; (c) consistent failure of distinct drug classes hitting the same target. None of these has occurred. ENHANCE failed a surrogate endpoint and was followed by IMPROVE-IT which succeeded on a clinical one. The CETP class failed not because LDL lowering doesn't work but because the molecules had off-target effects.
• The statin-benefit hypothesis would be substantially weakened by: (a) independent reanalysis of CTT IPD showing major-vascular-event reductions are not robust to differing trial selection; (b) STAREE and PREVENTABLE reading out negative in their primary populations; (c) blinded N-of-1 trials confirming high-prevalence specific muscle harm. SAMSON found the opposite for harm; STAREE/PREVENTABLE are pending.
• The heterodox position would be substantially weakened by: (a) MR-RCT convergence on additional drug classes (it has happened with ezetimibe, PCSK9 inhibitors, bempedoic acid); (b) blinded re-challenge of self-reported intolerant patients showing specific drug harm rather than nocebo (SAMSON went the other way). Both have occurred.

Table of contents

Part I — Biology and pathways 1. The mevalonate pathway, statin pharmacology, and the isoprenoid side-branch 2. Lipoprotein particles, the LDL-receptor cycle, PCSK9, and Lp(a) 3. Atherogenesis and the inflammation-vs-lipid adjudication 4. Reverse cholesterol transport and the death of the HDL hypothesis

Part II — The literature arc (1953–2026) 5. The diet-heart hypothesis era 6. The first statin era (1994–1998): 4S, WOSCOPS, CARE, LIPID, AFCAPS/TexCAPS 7. The mega-trial era (2002–2008): HPS, ASCOT-LLA, PROSPER, PROVE-IT, TNT, JUPITER 8. The CTT Collaboration and the meta-analytic consensus 9. Non-statin LDL lowering: ENHANCE, SHARP, IMPROVE-IT 10. PCSK9 monoclonal antibodies: FOURIER, ODYSSEY OUTCOMES, FOURIER-OLE 11. Inclisiran and the siRNA era: ORION-9/10/11, ORION-4, VICTORION 12. Bempedoic acid and the statin-intolerant niche: CLEAR Outcomes and 2024–2025 follow-ons 13. The Lp(a) frontier: pelacarsen, olpasiran, lepodisiran 14. The 2024–2026 frontier: STAREE, PREVENTABLE, AHA PREVENT, Byrne et al.

Part III — The disputes 15. The consensus camp — Baigent, Collins, Sabatine, Ridker, Nissen, Stone, Grundy, Catapano, Mach, Kastelein, Ference 16. The heterodox camp — Ravnskov, de Lorgeril, DuBroff, Diamond, Malhotra, Kendrick, Sultan 17. Methodological critics inside the mainstream — Ioannidis, Redberg, Demasi 18. Mendelian randomisation as adjudicator 19. Industry funding, ghost-writing, and the conflict-of-interest landscape 20. Roche's actual lipid footprint — context for the conversation

Part IV — The harms ledger 21. Statin-associated muscle symptoms (SAMS), SAMSON, and the nocebo phenomenon 22. New-onset diabetes — JUPITER, Sattar, HMGCR-MR 23. Cognitive complaints — FDA label, FOURIER-OLE, mechanism debate 24. Hepatic, renal, and rhabdomyolysis signals 25. Asymmetric adverse-event capture in pivotal trials

Part V — Special populations and common myths 26. Primary vs secondary prevention — the NNT chasm 27. Women — the underrepresentation legacy 28. The over-75s — STAREE, PREVENTABLE, NLA/AGS 2024 consensus 29. Familial hypercholesterolaemia — the strongest case 30. Diabetes, CKD, post-ACS — uncontested ground 31. Eight common myths adjudicated

Part VI — Adjudication and calibrated position 32. Bradford-Hill applied 33. NNT and NNH by risk stratum — the practical table 34. The calibrated landing position 35. What would change my mind — pre-committed updating

Back matter - Master trial table (all trials referenced, cross-indexed) - Numbered bibliography (Vancouver style) - Glossary - Author/agent attribution and methods note

Part I — Biology & Pathways

A critical synthesis for the cholesterol / statin debate. Steelmanned on both sides. Footnotes are Vancouver-style and resolvable to PubMed, DOI, or society URLs. Where a claim is widely-held but I could not verify a specific primary source within the time budget, the footnote is marked [CITATION NEEDED] so the auditor can adjudicate before the document goes external.

Chapter 1 — The mevalonate pathway, end-to-end

1.1 Why this pathway is the right place to start a cholesterol argument

Almost every substantive disagreement about statins — whether they cause myopathy, whether they cause cognitive impairment, whether their "pleiotropic" anti-inflammatory effects are a benefit or a hidden harm, whether CoQ10 supplementation is justified — collapses, in the end, onto a single biochemical fact: the enzyme statins inhibit, HMG-CoA reductase, sits upstream of far more than cholesterol. Block it hard enough and you don't just lose cholesterol; you lose ubiquinone, dolichol, isoprenylated signalling proteins, haem A, and a handful of other end-products that human cells need to function. The proponents say the dose at which statins are clinically used spares those branches; the critics say it doesn't, or at least not reliably, and that the muscle and cognitive complaints are the audible signal of that biochemistry. To adjudicate that argument you must first see the pathway.

The mevalonate pathway is one of the most ancient and conserved biosynthetic routes in eukaryotic biology. Plants run a variant of it; archaea run a variant; yeast and humans run essentially the same scheme. It is the single source in human cells of isoprenoid carbon — the C5 building blocks that, by serial condensation, become everything from the cholesterol in cell membranes to the prenyl tail on Ras that lets it dock to the inner leaflet of the plasma membrane and switch a cell into proliferation. It is, in other words, structural chemistry and signalling chemistry at once, and that double role is why the argument over statins is fundamentally about more than LDL.

1.2 Step-by-step: acetyl-CoA to cholesterol

I lay the pathway out below in canonical form, with the clinically relevant enzymes named at each step. Two enzymes are bolded because they are the points where pharmacological intervention sits in 2026: HMG-CoA reductase (statins) and ATP-citrate lyase (bempedoic acid, one step upstream of the statin target).¹

 Glucose / fatty acid β-oxidation
            │
            ▼
       Acetyl-CoA  (cytosolic pool)
            │
            │   ATP-citrate lyase  ◄─── bempedoic acid acts here[^2]
            ▼
       Acetoacetyl-CoA
            │
            │   HMG-CoA synthase (HMGCS1, cytosolic)
            ▼
       3-hydroxy-3-methylglutaryl-CoA  (HMG-CoA)
            │
            │   ★ HMG-CoA REDUCTASE (HMGCR)  ◄─── statins act here[^3]
            │   rate-limiting step; NADPH-consuming;
            │   regulated by SREBP-2, feedback by sterols & non-sterol isoprenoids
            ▼
       Mevalonate  (C6)
            │
            │   Mevalonate kinase (MVK)        [defect → mevalonate kinase deficiency / HIDS]
            ▼
       Mevalonate-5-phosphate
            │
            │   Phosphomevalonate kinase (PMVK)
            ▼
       Mevalonate-5-pyrophosphate
            │
            │   Mevalonate diphosphate decarboxylase (MVD)
            ▼
       Isopentenyl pyrophosphate (IPP, C5)  ◄══►  Dimethylallyl pyrophosphate (DMAPP, C5)
                                  │   (IPP isomerase, IDI1)
                                  │
   ┌──────────────────────────────┤
   │                              ▼
   │                       Geranyl pyrophosphate (GPP, C10)
   │                              │
   │                              │   Farnesyl pyrophosphate synthase (FDPS)
   │                              ▼
   │                       ░░░ Farnesyl pyrophosphate (FPP, C15) ░░░   ◄── branch point
   │                              │
   │   ┌──────────────┬───────────┼───────────────┬─────────────────────┐
   │   ▼              ▼           ▼               ▼                     ▼
   │  Squalene     Geranyl-       Heme A         Prenylated           Dolichol
   │  synthase    geranyl-PP      (cytochromes   proteins             (N-glycosylation
   │  (FDFT1)     (C20, via       a/a3 in        — Ras (farnesyl)     of glycoproteins;
   │   │          GGPS1)          mitochondria;  — Rho/Rac/Cdc42      ER membrane)
   │   ▼          via FPP +       requires        (geranylgeranyl)    [^4]
   │  Squalene    IPP             FPP-derived    — nuclear lamins
   │   │                          farnesyl       (farnesyl)
   │   │                          tail of haem
   │   ▼          ▼
   │  Squalene   Ubiquinone (CoQ10)
   │  mono-      (mitochondrial ETC,
   │  oxygenase  complex I→III electron
   │  (SQLE)     carrier; the carbon
   │   │         tail derives from
   │   ▼         FPP/IPP via trans-prenyltransferase)[^5]
   │  Lanosterol (C30)
   │   │
   │   │   ~19 enzymatic steps
   │   │   (Bloch / Kandutsch–Russell branches; CYP51, DHCR24, DHCR7…)
   │   ▼
   │  Cholesterol (C27)
   │
   └─► (re-feeds: bile acids, steroid hormones, vitamin D, oxysterols → LXR ligands,
        membrane lipid raft architecture, hedgehog protein modification, etc.)

The pathway has three features that matter for the rest of this document.

First, HMG-CoA reductase is the rate-limiting step, and it is under spectacularly tight feedback control. Brown and Goldstein characterised the regulatory architecture across the 1970s and 1980s and showed it operates at three levels: transcription (via the SREBP-2 / SCAP / Insig system, which senses sterol abundance in the ER membrane); translation (via non-sterol isoprenoid signals); and post-translational degradation (sterol-induced ubiquitination of HMGCR by gp78 and Insig).⁶⁷ When statins inhibit HMGCR, this feedback machinery responds by up-regulating both HMGCR (so the body partially compensates) and, critically, the LDL receptor. The LDL-receptor upregulation is the entire therapeutic mechanism — statins lower plasma LDL primarily by making the liver more avid for circulating LDL particles, not by stopping cholesterol biosynthesis per se.⁸

Second, the pathway is branched at FPP. This is the single most contested fact in the statin debate. The same C15 farnesyl-pyrophosphate molecule is the precursor for: squalene (and therefore cholesterol); GGPP (and therefore geranylgeranylated small GTPases); the prenyl tail of haem A (and therefore mitochondrial cytochromes a/a3 in complex IV); the prenyl side-chain of ubiquinone (CoQ10) in mitochondrial electron-transport complexes I/II/III; and dolichol (and therefore N-linked glycosylation of essentially all secreted and membrane glycoproteins). Block flux at HMGCR and you risk reducing the supply to all downstream branches, not only the cholesterol branch.

Third, the pathway is not just a metabolic conveyor belt: it is a signalling pathway. Farnesyl-PP and geranylgeranyl-PP are the obligatory lipid anchors for some of the most consequential signalling proteins in the human cell. Ras, Rho, Rac, Cdc42, Rheb, the Rab family, the nuclear lamins — none of them traffic to their target membrane without their prenyl tail, and they don't get that tail without the FPP/GGPP supply that the mevalonate pathway alone provides. This is the mechanistic substrate of the "pleiotropic effects" literature, and as we'll see, the same mechanism that the statin sceptics invoke to argue harm is invoked by the statin enthusiasts to argue extra benefit beyond LDL-lowering. Both can be true at different doses, in different tissues, in different patients. That tension is unresolved.

1.3 The history (compressed): Bloch, Endo, Brown & Goldstein

The pathway as drawn above was reconstructed across roughly forty years. Konrad Bloch's lab in the 1950s and 1960s, using radio-labelled acetate, traced the carbon flow from acetate through squalene to cholesterol and won the 1964 Nobel Prize for it.⁹ Feodor Lynen, sharing that Nobel, characterised acetyl-CoA itself. Brown and Goldstein at UT Southwestern then spent the 1970s working out the receptor-mediated uptake of LDL and the feedback regulation of HMGCR, work for which they shared the 1985 Nobel in Physiology or Medicine.¹⁰ The pharmacological coup, separately, belongs to Akira Endo at Sankyo, who in 1972–1976 screened over 6,000 fungal extracts for HMGCR inhibitors and isolated compactin (ML-236B, later mevastatin) from Penicillium citrinum; his 1976 papers are the moment the statin era begins.¹¹ Merck's lovastatin (a Penicillium / Aspergillus terreus product structurally close to compactin) reached market in 1987 as the first FDA-approved statin.

1.4 What statins actually do at the molecule

Statins are competitive inhibitors of HMG-CoA reductase. Their bulky, decalin-ring-and-acid pharmacophore mimics the HMG portion of HMG-CoA and binds the active site with affinities (Ki) in the low nanomolar range — roughly four orders of magnitude tighter than the natural substrate.¹² Different statins differ in their lipophilicity (simvastatin and lovastatin are highly lipophilic and cross membranes by passive diffusion; rosuvastatin and pravastatin are hydrophilic and depend on OATP1B1-mediated hepatic uptake), in their pharmacokinetic half-life (rosuvastatin and atorvastatin long; simvastatin and lovastatin short), and in their susceptibility to CYP3A4 versus CYP2C9 metabolism (which determines drug-interaction profile, especially with macrolides, azoles, and grapefruit furanocoumarins). These differences matter clinically — they explain why some patients tolerate one statin but not another, and why rhabdomyolysis with cerivastatin (Baycol, withdrawn 2001) was so much more frequent than with the surviving statins.¹³

The downstream consequence of HMGCR inhibition, as Brown and Goldstein worked out, is a fall in intracellular sterol pool → SREBP-2 cleavage and nuclear translocation → up-regulation of both HMGCR (partial compensation) and LDLR (the therapeutic effect). The liver, the dominant site of LDL clearance, then pulls more LDL from plasma. Net plasma LDL falls 30–55% depending on statin and dose. Importantly, statins also reduce VLDL secretion modestly and increase LDL particle catabolism, and through mechanisms still debated they reduce hsCRP — the "pleiotropic" effect that Ridker built the JUPITER programme around.¹⁴

1.5 The sceptic case, steelmanned: CoQ10 depletion and prenylation collapse

This is where the document has to slow down. The case that statin biochemistry necessarily produces harm at the FPP branches is not a fringe argument; it has been made in respectable journals by serious investigators, and dismissing it as quackery (as some establishment voices have) is bad scholarship. Let me lay it out as strongly as I can before adjudicating.

The CoQ10 argument. Ubiquinone (coenzyme Q10) is the lipid-soluble electron carrier of the mitochondrial inner membrane. It accepts electrons from complex I and complex II and shuttles them to complex III. Its tail is a polyprenyl chain — ten isoprene units in humans (hence Q10) — assembled by trans-prenyltransferase from FPP plus successive IPP additions. If you block HMGCR you reduce mevalonate, and therefore IPP/DMAPP, and therefore the tail-synthesis substrate for CoQ10. Folkers, Langsjoen and colleagues reported in 1990 in PNAS that lovastatin reduced blood CoQ10 levels in humans and that the depletion was clinically meaningful in a small cohort with cardiomyopathy.¹⁵ Langsjoen's 2003 review in BioFactors pulled together two decades of animal and human data showing reproducible CoQ10 depletion across statins.¹⁶ Beatrice Golomb at UCSD took this further: her 2008 American Journal of Cardiovascular Drugs review of statin adverse effects compiled nearly 900 studies and explicitly proposed a mitochondrial mechanism — CoQ10 depletion plus impaired mitochondrial respiration — as the unifying biology for muscle, cognitive, peripheral neuropathy, and energy-related complaints.¹⁷ In 2011, Langsjoen reported that statin-associated cardiomyopathy responded to statin withdrawal and CoQ10 supplementation in observational case series.¹⁸

The mechanism is biologically plausible. CoQ10 deficiency, when severe (the inherited primary deficiencies caused by mutations in COQ2, COQ4, COQ6, etc.), produces a spectrum that includes encephalomyopathy, nephrotic syndrome, ataxia, and exercise intolerance — a phenotype that overlaps non-trivially with the patient-reported statin adverse-effect cluster. And the muscle biopsies of statin-myopathy patients have, in several series, shown ragged-red fibres, lipid accumulation, and reduced respiratory chain enzyme activity, consistent with a mitochondrial lesion.¹⁹

The prenylation argument. GGPP (and to a lesser extent FPP) is the obligatory lipid anchor for the small GTPases Rho, Rac, Cdc42, and Rheb. Statins, by depleting GGPP, can reduce membrane localisation of these proteins. In skeletal muscle, this disrupts the Rho/ROCK pathway that maintains cytoskeletal architecture and Rac-dependent mitophagy. In macrophages, it disrupts Rac-NADPH-oxidase coupling. In vitro and in some animal studies, the muscle-toxicity phenotype of statins can be partially rescued by adding back GGPP or mevalonate, suggesting prenylation collapse is causally implicated.²⁰ In the cognition literature, Ras and Rho GTPase activity is required for synaptic plasticity, and there is at least a plausible biological route by which severe prenylation deficit could produce the cognitive complaints that some patients report.²¹

The sceptic, then, has a coherent biochemical story: HMGCR inhibition depletes IPP/DMAPP/FPP/GGPP; CoQ10 supply falls; mitochondrial respiration suffers; prenylated GTPases fail to traffic; muscle fibres and neurons (the most ATP-demanding cells) are first to show symptoms. On this story, statin "side effects" are not idiosyncratic — they are the expected pharmacology, audible in any patient sensitive enough to notice.

1.6 Adjudication: where the sceptic case is right, where it is overreached

Now the harder work. The sceptic case is partly right and partly wrong, and the document has to say which is which.

What is well-supported. That statins reduce circulating CoQ10 is essentially undisputed — multiple independent studies show 16–54% reductions in plasma CoQ10 with statin therapy, and the mechanism (reduced isoprenoid supply plus the fact that plasma CoQ10 travels on LDL, which itself falls) is not in question.²² That a subset of patients on statins experience muscle symptoms is also undisputed — though the frequency and the question of nocebo contribution is the subject of vigorous argument (Part IV deals with this in detail).

What is weakly-supported. That muscle tissue CoQ10 falls in proportion to plasma CoQ10 is contested. Several muscle-biopsy studies have failed to find significant intramuscular CoQ10 depletion in statin-treated patients, even when plasma levels were down.²³ This is the key empirical gap: plasma CoQ10 is largely an LDL hitchhiker, so a fall in plasma CoQ10 partly reflects the fall in LDL particles rather than a fall in intracellular CoQ10 supply in the tissues that need it.

What has been falsified at the trial level. That CoQ10 supplementation prevents statin-associated muscle symptoms in randomised, blinded trials. The 2015 Cochrane-style meta-analysis by Banach et al. and the 2018 trial by Taylor et al. both failed to find a clinically meaningful effect of CoQ10 supplementation on statin myalgia in blinded comparison with placebo.²⁴ If the CoQ10 mechanism were the dominant cause of statin myalgia, replacement should help. It doesn't, robustly. This is a real problem for the strong version of the CoQ10 hypothesis.

Where the prenylation case lands. The prenylation biology is real, but the dose-response in clinical practice appears to leave most prenylation largely intact. In vitro studies typically use statin concentrations 10–100× higher than steady-state plasma levels in patients. The cell does not deplete its prenylated-protein pool at therapeutic statin exposures — it would have manifested as catastrophic, not as muscle aches in 5–15% of users. The strongest version of the prenylation case is therefore biochemically plausible but quantitatively wrong at therapeutic doses for most patients. The weaker, individually-variable version — that some patients have polymorphisms in OATP1B1, CYP3A4, GGPS1, or COQ-pathway enzymes that make them outliers — is genuinely supported (the SLCO1B1*5 variant predicts simvastatin myopathy risk).²⁵

So the steelman survives in part: there is a biologically real, mechanistically coherent harm pathway from HMGCR inhibition through to mitochondrial and prenylation disturbance, and it explains a real fraction of statin intolerance. But the strong, universalist claim — that every patient on a statin is sustaining quiet mitochondrial damage and that CoQ10 should be co-prescribed — is not supported by the blinded trial data on CoQ10 replacement. The truth lives in pharmacogenomic outliers and in dose-dependence, not in the population mean.

I'll return to this in Part IV when I do the harms numbers. For Chapter 1 the takeaway is structural: the pathway is branched at FPP, the branches are clinically real, and any honest discussion of statins has to start with the biochemistry rather than the marketing materials.

Chapter 2 — Lipoprotein particles, apolipoproteins, and the LDL-receptor cycle

2.1 Why we have lipoproteins at all

Cholesterol and triglyceride are hydrophobic. Plasma is aqueous. Some packaging is therefore necessary, and evolution settled on micelle-like particles built around a phospholipid monolayer with a hydrophobic core (cholesteryl ester, triglyceride) and a surface decorated with apolipoproteins that act as the address labels and enzymatic cofactors. The system has five functional classes — chylomicrons, VLDL, IDL, LDL, HDL — separable historically by density on ultracentrifugation (which is where the names "low density" and "high density" come from; it has nothing to do with cholesterol "goodness").

The system has three jobs: it moves dietary fat from gut to peripheral tissues (chylomicrons); it moves endogenously-synthesised fat and cholesterol from liver to peripheral tissues (VLDL → IDL → LDL); and it moves excess cholesterol from peripheral tissues back to the liver for disposal (HDL → reverse cholesterol transport). Each particle class has a different apolipoprotein signature, a different size, and a different fate. Understanding atherogenesis requires understanding all of them, because the proximal cause of atherosclerotic lesions is the retention of certain of these particles in the arterial wall.

2.2 The particle classes

Chylomicrons are the largest and least dense (75–1,200 nm, density <0.95 g/mL). They are assembled in enterocytes after a fatty meal, packaged in the ER around the truncated apolipoprotein ApoB-48 (a unique gut-specific isoform generated by mRNA editing of the ApoB-100 transcript). They acquire ApoC-II and ApoE in plasma from HDL particles; ApoC-II is the activator of lipoprotein lipase (LPL) on capillary endothelium, which hydrolyses their triglyceride load and releases fatty acids for tissue uptake. The chylomicron remnants — now smaller, denser, depleted of triglyceride, enriched in cholesteryl ester — are cleared by the liver via the LDL-receptor-related protein (LRP1) and the LDL receptor itself, both recognising ApoE.²⁶ Chylomicron remnants are atherogenic; this is one reason post-prandial triglyceride matters.

VLDL (very-low-density lipoproteins; 30–80 nm, density 0.95–1.006 g/mL) are the hepatic equivalent of chylomicrons. They carry endogenously-synthesised triglyceride (largely from de novo lipogenesis and free fatty-acid recycling) plus cholesteryl ester, built around one molecule of ApoB-100 per particle. ApoB-100 is enormous — 4,536 amino acids, the largest single-chain protein in human plasma — and a single VLDL particle, throughout its life cycle, retains the same ApoB-100 molecule. This fact, that one particle equals one ApoB, is the conceptual hinge that makes ApoB measurement equivalent to particle counting.²⁷

IDL (intermediate-density lipoproteins; density 1.006–1.019 g/mL) are the transient product of partial VLDL lipolysis. They retain ApoB-100, some ApoE and ApoC. About half are cleared by the liver via the LDLR (recognising ApoE); the other half are further hydrolysed (largely by hepatic lipase) to LDL.

LDL (low-density lipoproteins; ~22 nm, density 1.019–1.063 g/mL) are the end-stage product of VLDL → IDL → LDL processing. They are cholesterol-enriched (about 50% cholesteryl ester by mass), triglyceride-depleted, and carry ApoB-100 as their sole apolipoprotein. ApoB-100 is the ligand that lets the LDL receptor bind LDL. The plasma half-life of LDL is 2–3 days, much longer than VLDL or IDL, which is why LDL is the dominant ApoB-bearing particle in fasted plasma.

LDL is not a uniform population. It exists as a continuum of sizes and densities, conventionally separated into "pattern A" (large, buoyant) and "pattern B" (small, dense). Pattern-B LDL is more atherogenic per particle, for reasons we'll get to: it has higher affinity for arterial proteoglycans, lower affinity for the hepatic LDLR, and longer plasma residence.²⁸

HDL (high-density lipoproteins; 5–17 nm, density 1.063–1.21 g/mL) are the smallest and most protein-dense particles. They are nucleated by ApoA-I secreted from liver and gut as "discoidal" pre-β-HDL, which then acquire phospholipid and free cholesterol from peripheral tissues via the ATP-binding cassette transporters ABCA1 (initial efflux to lipid-poor ApoA-I) and ABCG1 (efflux to mature HDL). LCAT (lecithin-cholesterol acyltransferase) esterifies the free cholesterol in the HDL core. Mature HDL then either returns cholesterol to the liver directly via the SR-BI receptor (selective uptake, particle retained) or hands cholesterol off indirectly via CETP-mediated exchange with ApoB-bearing particles. This is "reverse cholesterol transport," and it is the basis of the now largely-failed HDL-raising drug programme that Chapter 4 will dissect.

2.3 The apolipoproteins, briefly

The apolipoprotein signature is the address label of the particle. A short summary:

• ApoB-100 — one per VLDL/IDL/LDL/Lp(a). Liver origin. Ligand for LDLR. Non-exchangeable between particles (because of its hydrophobic anchoring). This is what makes ApoB count = particle count.²⁹
• ApoB-48 — gut-derived chylomicron form, lacking the LDLR-binding domain (the C-terminal half of ApoB is missing due to a stop codon introduced by the apobec-1 mRNA-editing enzyme).
• ApoA-I — main HDL protein. Liver/gut origin. LCAT activator. SR-BI ligand. ABCA1 acceptor.
• ApoA-II — second HDL protein; function debated, possibly anti-LCAT.
• ApoC-II — LPL activator. Deficiency produces familial chylomicronaemia syndrome.
• ApoC-III — LPL inhibitor. Genetic loss-of-function variants are associated with lower triglycerides and lower CHD risk — the basis for the volanesorsen and olezarsen ASO programmes.³⁰
• ApoE — three common isoforms (E2, E3, E4); ligand for LDLR and LRP1. ApoE2/E2 homozygosity causes type III dysbetalipoproteinaemia. ApoE4 is the major genetic risk factor for late-onset Alzheimer's, an inconvenient fact for the lower-is-always-better school.
• Apo(a) — the kringle-rich glycoprotein covalently attached by a disulfide bond to ApoB-100 in Lp(a). See section 2.6.

2.4 The LDL-receptor cycle: Brown & Goldstein's machine

Joseph Goldstein and Michael Brown received the 1985 Nobel for the LDL-receptor pathway. The story is, briefly, this. They were studying patients with familial hypercholesterolaemia (FH) — children with cholesterol levels of 600–1,000 mg/dL and aggressive premature atherosclerosis. They cultured the patients' fibroblasts and discovered that the cells failed to take up LDL from media. The defect localised to a cell-surface receptor, the LDL receptor, encoded by the LDLR gene on chromosome 19. Homozygous FH patients had two non-functional alleles and essentially no hepatic LDL clearance; heterozygotes had half the normal complement and double-normal LDL.³¹³²

The receptor cycle works as follows. The LDLR is a 839-amino-acid single-pass transmembrane glycoprotein. It is synthesised in the ER, glycosylated in the Golgi, and trafficked to the cell surface where it concentrates in clathrin-coated pits (the discovery of clathrin-mediated endocytosis began with this receptor). At neutral plasma pH, the receptor's ligand-binding domain binds ApoB-100 (or ApoE on remnant particles). The receptor–ligand complex is internalised into endosomes; the falling pH of the endosome triggers ligand dissociation; the receptor recycles to the plasma membrane (lifetime: ~20 hours, 150 cycles); the LDL particle is delivered to the lysosome where cholesteryl ester is hydrolysed to free cholesterol; the free cholesterol exits the lysosome via NPC1, enters the regulatory pool, suppresses SREBP-2, downregulates HMGCR and LDLR transcription, and the homeostatic loop closes.³³

Statins act by opening this loop: they reduce intracellular cholesterol biosynthesis, the SREBP-2 sensor responds as if the cell were cholesterol-starved, and LDLR is upregulated. More LDLR on hepatocytes = more LDL clearance from plasma. The drug's plasma-LDL-lowering effect is downstream of receptor upregulation, not directly downstream of biosynthesis inhibition. This is why statins fail to lower LDL in homozygous FH patients (no functional LDLR to upregulate) and partly succeed in heterozygous FH (one allele can be upregulated).

2.5 PCSK9: the discovery that re-shaped the LDLR cycle

In 2003, Marianne Abifadel and Catherine Boileau in Paris reported gain-of-function mutations in a previously poorly-characterised gene, PCSK9 (proprotein convertase subtilisin/kexin type 9), in French families with autosomal-dominant hypercholesterolaemia who were LDLR- and APOB-wild-type. The gain-of-function mutation produced a severe FH-like phenotype.³⁴ The mechanism, worked out subsequently, was elegant: PCSK9 is a secreted protein that binds the LDLR on the hepatocyte surface, and when the receptor–PCSK9 complex internalises, PCSK9 prevents receptor recycling. The receptor goes to the lysosome and is degraded rather than returned to the surface. PCSK9 is, in effect, a built-in negative regulator of the LDLR cycle.

The therapeutic logic followed within a few years and is one of the cleanest stories in modern drug development. Helen Hobbs and Jonathan Cohen at UT Southwestern found, in 2005–2006, that loss-of-function PCSK9 variants are common in African-American populations and confer a 28% reduction in LDL-C with an 88% reduction in CHD risk over the followed cohort; in white populations a different variant gave 15% LDL reduction and 47% CHD risk reduction.³⁵ This was a Mendelian-randomisation demonstration that lifelong moderate LDL reduction produces large CHD risk reduction — and it was a demonstration that knocking out PCSK9 was both safe (the homozygous LOF carrier had cholesterol of 14 mg/dL and was healthy) and powerfully cardioprotective. The monoclonal-antibody programme followed: alirocumab (Sanofi/Regeneron) and evolocumab (Amgen) entered phase 3 within a decade, both showing ~60% LDL reductions on top of statins. FOURIER (2017, evolocumab) and ODYSSEY OUTCOMES (2018, alirocumab) confirmed cardiovascular event reduction.³⁶ The FOURIER open-label extension reported in 2022–2023 confirms a durable benefit out to a median 5 years of OLE with no cognitive safety signal.³⁷

Inclisiran (Novartis/Alnylam, marketed as Leqvio) is the second generation: a small interfering RNA conjugated to GalNAc for hepatocyte targeting, dosed twice yearly, knocking down hepatic PCSK9 mRNA. LDL reduction ~50%, sustained between doses. The outcomes trial ORION-4 reads out July 2026 and will tell us whether the LDL reduction translates to MACE benefit in a primary-prevention-adjacent population.³⁸

2.6 Lp(a): the genetic LDL-like particle the statins barely touch

Lipoprotein(a), or "L-P-little-a," is the genetically determined nuisance of modern lipidology. Discovered by Kåre Berg in 1963 in Oslo (originally as a serum antigen variant detectable by immunodiffusion), Lp(a) is structurally an LDL particle — ApoB-100, cholesteryl-ester core, phospholipid surface — with one additional component: a covalently bound molecule of apolipoprotein(a), a kringle-rich glycoprotein encoded by the LPA gene on chromosome 6q26-27.³⁹

Apo(a) is evolutionarily a derivative of plasminogen (the fibrinolytic zymogen), with which it shares the signature kringle domains. Where plasminogen has five distinct kringles, apo(a) has lost most of them and amplified one — kringle IV type 2 (KIV-2) — by a variable-copy-number repeat. Different individuals have anywhere from 3 to >40 copies of KIV-2, producing apo(a) isoforms ranging from ~300 to >800 kDa. The number of KIV-2 repeats is inversely correlated with plasma Lp(a) concentration: shorter isoforms (fewer KIV-2 copies) are secreted faster and accumulate to higher plasma levels. Roughly 70–90% of the variance in plasma Lp(a) is genetic, set at birth, essentially fixed for life.⁴⁰

Why Lp(a) matters: large epidemiology (Copenhagen General Population Study, Reykjavik, UK Biobank) and Mendelian randomisation (variants in LPA affect Lp(a) levels and predict CHD risk in proportion) converge on Lp(a) as an independent causal risk factor for atherosclerotic cardiovascular disease, myocardial infarction, ischaemic stroke, and — distinctively — calcific aortic valve stenosis.⁴¹ The risk threshold conventionally cited is ~50 mg/dL (or ~125 nmol/L), above which CV risk rises proportionately. About 20% of the global population, and substantially more of South Asian and African-ancestry populations, exceed that threshold.

The mechanism by which Lp(a) is atherogenic combines its LDL-like component (delivery of cholesterol into the intima, oxidation, foam cell formation) with two distinctive features: it preferentially carries oxidised phospholipids on the apo(a) tail (making it pro-inflammatory), and it competes with plasminogen at the endothelial surface (making it weakly pro-thrombotic). The aortic-valve association is thought to derive from the same oxidised-phospholipid carriage, driving calcific valve remodelling.

Why standard statins fail Lp(a). Statins lower LDL by upregulating the LDLR. But Lp(a) is cleared predominantly through other pathways — plasminogen receptors, scavenger receptors, and to a lesser extent the LDLR — and statin-driven LDLR upregulation does not translate efficiently to Lp(a) clearance. In fact, several studies have reported small increases in Lp(a) on statin therapy, possibly because reduced ApoB-100 competition frees up apo(a) assembly.⁴² Lp(a) is genetically fixed and, until very recently, pharmacologically untouchable. PCSK9 inhibitors lower it by ~25%; niacin by ~25%; lipoprotein apheresis by ~70% acutely. None of these is a satisfactory therapeutic answer.

The 2024–2026 frontier. Three RNA-targeted therapies aim directly at hepatic apo(a) synthesis. Pelacarsen (Novartis/Akcea, an antisense oligonucleotide) reduces Lp(a) by ~80% in phase 2; the Lp(a)HORIZON outcomes trial reads out mid-2025.⁴³ Olpasiran (Amgen, an siRNA) reduces Lp(a) by ~94% with quarterly dosing; OCEAN(a)-Outcomes reads out 2027.⁴⁴ Lepodisiran (Eli Lilly, a longer-acting siRNA) reduces Lp(a) by ~94% out to 360 days after a single dose; ACCLAIM-Lp(a) phase 3 is enrolling. Whether reducing Lp(a) by 80–94% translates to CV event reduction — and how much — is the most consequential open question in lipidology in 2026.

2.7 The discordance literature: why ApoB may matter more than LDL-C

Here is the conceptual problem that drives a real and ongoing argument in clinical lipidology. The conventional risk metric is LDL-cholesterol (LDL-C), measured in mg/dL or mmol/L, and it represents the mass of cholesterol carried in LDL particles per unit volume of plasma. But the atherogenic event in the arterial wall is not driven directly by cholesterol mass — it is driven by particle entry and retention. One LDL particle is one LDL particle, whether it carries 1,500 cholesterol molecules or 2,500. The particle is what crosses the endothelium and binds the subendothelial proteoglycan matrix. Particle count, not particle cargo, is the atherogenic unit.

Because ApoB-100 is non-exchangeable and one ApoB-100 = one ApoB-bearing particle, the plasma ApoB concentration is the ApoB-particle count. And because LDL accounts for ~90% of ApoB-bearing particles in fasted plasma, plasma ApoB is approximately the LDL particle count plus a small contribution from VLDL and Lp(a). Plasma ApoB and plasma LDL-C are well-correlated in most patients — but not in all. The minority in whom they discord are clinically meaningful: typically patients with metabolic syndrome, insulin resistance, or type 2 diabetes, whose LDL particles are small and dense and cholesterol-poor. In these patients, LDL-C under-estimates the true particle count and therefore under-estimates atherogenic burden. The patients look "fine" on a standard lipid panel and develop premature CHD anyway.⁴⁵

Allan Sniderman, working out of McGill, has been the central figure pushing this. His 2019 JAMA Cardiology meta-analysis and subsequent UK Biobank work demonstrated that when LDL-C and ApoB discord, cardiovascular risk tracks with ApoB, not with LDL-C.⁴⁶ The 2024 European Heart Journal analysis of discordance, with a 2025 JAMA Cardiology follow-up, showed that for any given LDL-C value, ApoB spans a wide range (~80–110 mg/dL at LDL-C of 130 mg/dL), and the high-ApoB-low-LDL-C subgroup carries excess CV risk that LDL-C measurement misses entirely.⁴⁷

The ESC/EAS guidelines now endorse ApoB as a reasonable primary risk metric, and the 2018 ACC/AHA US guidelines acknowledged it as a secondary measurement in selected patients. Routine clinical practice in the US, however, still leans on LDL-C, partly because it's cheaper, partly because the trial literature targets it, and partly because endocrinology and lipidology services remain in disagreement about how aggressively to switch. The discordance argument is one of the genuinely live methodological debates in 2026 and feeds directly into Part III's "what should the target be" question.

2.8 Familial hypercholesterolaemia as the human knockout experiment

Before leaving the lipoprotein chapter, a note on FH because it is, conceptually, the natural experiment that resolves much of the LDL-causality argument. Heterozygous FH (loss-of-function in one LDLR allele, or in APOB, or gain-of-function in PCSK9) occurs in ~1:250 globally — far more common than once thought. Affected individuals have LDL-C of 200–400 mg/dL from birth and develop premature CHD: MI in the 30s and 40s untreated, severe coronary disease by 50. Homozygous FH (~1:300,000) produces LDL-C of 600–1,000 mg/dL and MI in the second decade of life if untreated.⁴⁸

The FH phenotype is the cleanest demonstration available that lifelong elevated LDL causes ASCVD. It is not a confounded epidemiological association. It is a genetic dose-response: more LDLR dysfunction → higher LDL → more atherosclerosis → earlier MI. The treatment story is correspondingly clean: statins extend life expectancy in heterozygous FH from ~60s into normal range; PCSK9 inhibitors and (for homozygotes) LDL apheresis and lomitapide further extend it. Any account of cholesterol that does not deal with FH is incomplete; any account that dismisses LDL as a non-causal bystander has to explain why FH patients die young of MI, and the standard sceptic narrative does not, in my view, succeed in that.

Chapter 3 — Atherogenesis: lipid retention, oxidation, inflammation, plaque

3.1 The lesion, in stages

A coronary atheroma is not an instantaneous event. It is the result of a multi-decade process that begins in childhood (fatty streaks are detectable in the aortas of adolescents at autopsy after accidental death) and progresses, over years or decades, through stages that the American Heart Association working group categorised in 1995 (Stary classification):

• Type I — isolated intimal macrophages with intracellular lipid.
• Type II — fatty streak: multiple foam cells, extracellular lipid in small pools.
• Type III — pre-atheroma: extracellular lipid pools coalesce.
• Type IV — atheroma: a defined lipid core forms in the intima.
• Type V — fibroatheroma: a fibrous cap of smooth-muscle-cell-derived collagen overlies the lipid core.
• Type VI — complicated lesion: rupture, ulceration, haemorrhage, thrombosis.⁴⁹

The clinical event — myocardial infarction, ischaemic stroke, sudden cardiac death — almost always corresponds to a Type VI event: plaque rupture or erosion with overlying thrombosis. The dominant insight of the last twenty years, driven largely by intravascular imaging (IVUS, OCT) and pathology series from Renu Virmani and colleagues, is that the dangerous plaque is not necessarily the biggest one. A high-grade stenosis (>70%) is usually a stable, fibrocalcific plaque with a thick cap and a small lipid core. The dangerous plaque is the vulnerable plaque: a moderate stenosis with a large necrotic lipid core, a thin fibrous cap, abundant macrophages, and intra-plaque inflammation. These plaques rupture, expose tissue factor to blood, and thrombose acutely.⁵⁰

3.2 The response-to-retention hypothesis

The dominant conceptual model of atherogenesis since the mid-1990s is the response-to-retention hypothesis, articulated by Kevin Jon Williams and Ira Tabas in a 1995 Arteriosclerosis, Thrombosis, and Vascular Biology paper that has been cited tens of thousands of times.⁵¹ The argument is that subendothelial retention of ApoB-containing lipoproteins is the necessary first event in atherogenesis, and that essentially everything else in the plaque — the inflammation, the foam cells, the smooth-muscle migration, the cap formation, the necrotic core — is a downstream response to that retention.

The argument runs as follows. LDL particles continually transit the arterial endothelium, especially at sites of disturbed flow (branch points, the lateral walls of bifurcations, the inner curve of the aortic arch — exactly where atherosclerosis preferentially develops). In a healthy artery, most of the LDL that enters the intima exits unmolested. In an at-risk artery, the subendothelial extracellular matrix — rich in proteoglycans (decorin, biglycan, versican) and glycosaminoglycans — binds the basic residues on ApoB-100 and retains the particle. Once retained, the particle is no longer flushing through — it sits in the intima and, over time, undergoes modifications: oxidation (by myeloperoxidase, NADPH oxidases, lipoxygenases, reactive nitrogen species), aggregation, enzymatic digestion. Oxidised LDL is no longer recognised by the LDLR (which would have cleared it normally); it is recognised instead by scavenger receptors (SR-A, CD36, LOX-1) on infiltrating monocyte-derived macrophages. Scavenger receptors are not feedback-regulated by intracellular cholesterol, so the macrophage continues to ingest oxidised LDL indefinitely, accumulating cholesteryl ester droplets until it becomes the histological foam cell.⁵² The foam cell secretes cytokines (TNF-α, IL-1β, IL-6, MCP-1) that recruit more monocytes; the cycle accelerates.

The strength of the response-to-retention model is that it explains why the plaque forms preferentially at certain anatomic sites (the local proteoglycan composition differs), why it correlates with circulating ApoB-particle concentration (more particles → more retention), and why genetic and pharmacological lowering of ApoB-particles (statins, PCSK9 inhibitors, ezetimibe, bempedoic acid, FH reversal) all reduce the rate of atherogenesis in proportion to the reduction in ApoB-particle entry. It is also why the LDL hypothesis and the inflammation hypothesis are not competitors: lipid retention initiates the lesion, inflammation amplifies and perpetuates it.

3.3 Smooth muscle, the fibrous cap, and the necrotic core

The intermediate stages of plaque growth involve the migration of medial smooth-muscle cells into the intima — a phenotypic switch from contractile to synthetic, regulated by PDGF, TGF-β, and (per recent lineage-tracing work by Owens and others) far more cellular plasticity than previously thought. These cells deposit collagen and form the fibrous cap that overlies the lipid core. A thick cap is a stable plaque. A thin cap (defined as <65 μm by OCT) over a large lipid core is the canonical vulnerable plaque.⁵³

The necrotic core itself forms by macrophage apoptosis under prolonged ER stress (cholesteryl crystals, oxidised LDL, oxysterols all promote apoptosis) and defective efferocytosis — the failure of neighbouring macrophages to clear the apoptotic ones. Dead cells lyse, release their cholesterol load extracellularly, and the lipid core enlarges. Cholesterol crystallisation in the necrotic core appears to be one of the activators of the NLRP3 inflammasome and the IL-1β cascade that links lipid lesion to systemic inflammation.⁵⁴

3.4 The inflammation-versus-lipid debate

This is the place where the document has to make space for genuine scientific disagreement. Three positions are defensible.

The pure-lipid position (held by, e.g., Sergio Fazio, Daniel Steinberg in his later writings, and most of the EAS lipid-trialist mainstream) holds that LDL is necessary and sufficient for atherogenesis given enough time and concentration, and that everything else is modulation. Evidence: FH; Mendelian randomisation across the genome (variants in LDLR, PCSK9, NPC1L1, HMGCR, APOB all predict CHD risk per unit LDL lowered, regardless of mechanism); the dose-response of statin and PCSK9 outcome trials; the linear LDL-vs-CHD relationship in the IMPROVE-IT meta-analysis and beyond.

The pure-inflammation position (held in its strongest form by Uffe Ravnskov, Malcolm Kendrick, and the THINCS group) holds that LDL is essentially a bystander or even protective, and that atherosclerosis is driven by endothelial injury and inflammation — infection, smoking, oxidative stress, autoimmunity. Evidence cited: the (selective) finding that older patients with higher LDL sometimes have lower mortality (the so-called cholesterol paradox in heart failure and very old populations); CANTOS as proof that anti-inflammatories work independently of lipids; the failure of pure HDL-raising drugs.⁵⁵ This is the position I'll deal with at length in Part IV, where I think the strong form fails but the weak form survives.

The both-matter position (held by Paul Ridker explicitly and probably by the modern mainstream implicitly) holds that LDL retention is the substrate and inflammation is the amplifier, that lowering either reduces events, and that the maximum benefit comes from lowering both. Evidence: CANTOS (canakinumab, an IL-1β-targeting antibody) reduced recurrent CV events by 15% at the 150 mg dose in patients with previous MI and hsCRP >2 mg/L, with no effect on LDL.⁵⁶ LoDoCo2 (colchicine 0.5 mg daily in chronic CAD) reduced the primary endpoint by 31% on top of statin therapy.⁵⁷ COLCOT (colchicine post-MI) gave similar results. Methotrexate, by contrast, in CIRT (Ridker's other anti-inflammatory trial) failed to reduce events — and the failure is informative because CIRT enrolled patients with diabetes/metabolic syndrome but without elevated hsCRP, and the methotrexate dose used did not lower hsCRP either; the trial therefore argues for residual IL-1/IL-6 axis-specific inflammation as the relevant target, not inflammation in general.⁵⁸

The current consensus position, defensible from the trial literature, is roughly: ApoB-particle retention is the necessary upstream event; once a plaque exists, inflammatory amplification of that plaque is an independent and modifiable contributor to event rates; both can and should be addressed in high-risk patients. CANTOS and LoDoCo2 are not refutations of the lipid hypothesis — they are the demonstration that even after maximal lipid-lowering there is residual inflammatory risk that further intervention reduces. That is a "yes-and," not an "either-or."

3.5 What the sceptic argument gets right, and what it overreaches

The sceptic argument that inflammation, not cholesterol, is primary is weakest when it is presented in pure form — i.e., when the claim is that LDL is a bystander. This claim has to deal with FH, with the MR data, and with the proportionality of risk reduction to ApoB reduction across drug classes operating on entirely different molecular targets. None of those facts is consistent with LDL being a bystander.

But the sceptic argument has a stronger version that survives scrutiny: that focusing exclusively on LDL is incomplete. Residual risk after maximal LDL lowering is large; some of that residual risk is Lp(a)-driven, some is triglyceride-remnant-driven, some is inflammation-driven (CANTOS/colchicine), some is haemodynamic, some is metabolic. The single-minded targeting of LDL-C as if it were the only modifier of atherosclerotic risk is a flatter version of the science than the data support. On this softer version, the sceptic is correctly identifying a real gap in mainstream practice — though their proposed remedy (abandon statins) is the wrong inference from the right observation.

I'll return to this in Part II when I look at how the trials are read, and in Part IV when I look at residual risk specifically.

Chapter 4 — Reverse cholesterol transport and the death of the HDL hypothesis

4.1 The HDL story, as it used to be told

For roughly two decades, the dominant clinical-lipidology framing was LDL = bad cholesterol; HDL = good cholesterol; raise the latter, lower the former, win. The framing was driven by consistent and reproducible epidemiology: across many large cohorts (Framingham, the MRFIT registry, PROCAM, MESA, Reykjavik), low HDL-C was an independent predictor of CHD risk. Per ~5 mg/dL decrement in HDL-C, CHD risk rose ~10%, independent of LDL-C and triglyceride.⁵⁹ The biological story to go with the epidemiology was reverse cholesterol transport: HDL collects excess cholesterol from peripheral tissues and returns it to the liver; therefore more HDL = more cholesterol drained out of the arterial wall = less plaque.

The mechanism, as I described in section 2.2, is genuinely real. HDL nucleation by ApoA-I, efflux via ABCA1 and ABCG1, LCAT esterification, return to the liver via SR-BI or via CETP-mediated exchange to ApoB-bearing particles — this all happens. It is genuinely how cells dispose of excess cholesterol. The question was whether pharmacologically raising HDL — by any mechanism — would translate to reduced atherosclerotic events. The answer, after twenty years of trials, is essentially no.

4.2 The CETP-inhibitor programme

Cholesteryl ester transfer protein (CETP) is a plasma enzyme that exchanges cholesteryl ester (from HDL) for triglyceride (from VLDL/LDL). Inhibit CETP and HDL-C rises (cholesterol doesn't leave HDL), LDL-C falls (cholesterol doesn't enter LDL). The pharmacological logic for CETP inhibition was therefore double: raise the good and lower the bad. Genetic loss-of-function in CETP (common in Japanese populations) is associated with very high HDL-C, and although the cardiovascular consequences in carriers were mixed, the framing was that CETP inhibition should be cardioprotective. Four major CETP-inhibitor programmes were taken to phase 3 trials. All failed, though the failures vary in interpretive complexity.

Torcetrapib / ILLUMINATE (Pfizer, 2007). 15,067 high-risk patients, randomised to torcetrapib + atorvastatin versus atorvastatin alone. HDL-C rose 72%, LDL-C fell 25%, and cardiovascular events rose (HR 1.25) along with all-cause mortality (HR 1.58). The trial was terminated early. The post-mortem identified an off-target effect: torcetrapib activated CYP11B2 (aldosterone synthase) and produced systolic BP elevation ~5 mmHg and increased aldosterone — a hypertensive and mineralocorticoid mechanism that drove the excess events.⁶⁰ On one reading, ILLUMINATE killed the drug but not the class; on another, ILLUMINATE killed Pfizer's $800 million programme and put a chill on the whole CETP class that took years to thaw.

Dalcetrapib / dal-OUTCOMES (Roche, 2012). This is the trial that matters for your conversation with the Roche friend. Dalcetrapib was Roche's CETP inhibitor — a more modest HDL-raiser (~30%) without significant LDL effect. The dal-OUTCOMES trial enrolled 15,871 patients with recent acute coronary syndrome, randomised dalcetrapib versus placebo, and at three years showed no benefit on the primary CV endpoint (HR 1.04, 95% CI 0.93–1.16, p=0.52).⁶¹ The trial was terminated for futility. Roche took the loss. Subsequent pharmacogenomic re-analysis (dal-GenE) tried to rescue the programme by identifying a ADCY9 genotype-defined responder subgroup, but the prospective dal-GenE trial in that subgroup, reported in 2022, also failed.⁶² Dalcetrapib is, for clinical purposes, dead. The COI implication for the conversation: when Anthony's Roche-employed friend defends "the industry," it is worth knowing that Roche's own CETP programme failed and that this should bias Roche staff toward humility about HDL pharmacology, not toward defending it.

Evacetrapib / ACCELERATE (Eli Lilly, 2017). 12,092 high-risk patients, evacetrapib versus placebo on background statin. HDL-C rose 132%, LDL-C fell 37%. Trial terminated early for futility — no difference in primary endpoint (12.9% vs 12.8%).⁶³ At this point the CETP class is hanging by a thread.

Anacetrapib / REVEAL (Merck, 2017). 30,449 patients with vascular disease on intensive atorvastatin, anacetrapib versus placebo, median 4.1 years. HDL-C rose 104%, non-HDL-C fell 17 mg/dL (18%). Major coronary events fell 9% (1640 vs 1803 events, p=0.004). Statistically positive — but Merck withdrew the drug anyway, citing the modest effect and accumulation of the drug in adipose tissue (anacetrapib has a >2-year tissue half-life). The withdrawal was, in effect, a commercial verdict that a 9% RR reduction with a drug that persists in body fat for years was not a viable product even though it technically worked.⁶⁴

Obicetrapib (NewAmsterdam Pharma, 2024–2026). A late-arriving CETP inhibitor; phase 3 BROADWAY (2024) and BROOKLYN reported substantial LDL reductions in HoFH and HeFH patients. The 2025 PREVAIL outcomes trial in patients with ASCVD is the still-running test of whether the class can finally deliver event reduction. As of mid-2026, results are pending; if positive, obicetrapib would resurrect the class.⁶⁵ If negative, the class is finished.

4.3 What the CETP failures teach

The 17 years of CETP trials are a near-perfect natural experiment in the pharmacological-HDL-raising hypothesis. Across four to five drugs operating on the same target, in tens of thousands of patients, raising HDL-C by 30–130% produces essentially no consistent CV event reduction. Anacetrapib's modest 9% benefit appears to be attributable to its substantial non-HDL-C lowering (i.e., ApoB-particle reduction), not to its HDL raising — consistent with the lipid-hypothesis prediction.

The teaching has been internalised by the field: HDL-C is a risk marker, not a risk factor. The same plasma HDL-C concentration can correspond to functionally very different HDL particles. HDL function (cholesterol efflux capacity, anti-oxidant capacity, anti-inflammatory capacity) does not track HDL mass concentration linearly. Pharmacological elevation of HDL-C without corresponding improvement in HDL function does not appear to do anything useful. This is a humbling lesson for surrogate-endpoint medicine: a biomarker that is causally informative in epidemiology is not necessarily causally responsive in pharmacology. The arrow runs from atherosclerosis to low HDL (probably via reduced ABCA1 efflux, increased CETP activity, and inflammatory remodelling of HDL particles) more strongly than from low HDL to atherosclerosis.⁶⁶

4.4 Niacin, CETP, ApoA-I infusion: the broader HDL graveyard

Niacin raises HDL-C ~20–25%, lowers LDL-C ~15%, lowers Lp(a) ~25%. It was a standard part of intensive lipid therapy for decades. Then two large RCTs of niacin added to statin therapy — AIM-HIGH (2011) and HPS2-THRIVE (2014) — both failed to show event reduction.⁶⁷⁶⁸ AIM-HIGH was terminated for futility; HPS2-THRIVE, with 25,000 patients, showed no benefit and substantial harms (diabetes, infection, bleeding). Niacin disappeared from guideline-recommended therapy.

The ApoA-I infusion programmes (recombinant ApoA-I, ApoA-I Milano, CER-001) followed a similar trajectory. Small early imaging trials (using IVUS plaque-regression endpoints) showed promising signals; the larger outcome trials and confirmatory imaging trials (AEGIS-II with CSL112, ApoA-I infusion post-MI) reported in 2024 with no benefit on MACE at 90 days or 1 year.⁶⁹

CETP inhibition aside, every major HDL-raising or HDL-mimicking approach taken to phase 3 has failed. The HDL-as-drug-target hypothesis is, for practical purposes, dead — at least in its mass-concentration form. Researchers continue to work on HDL function (efflux capacity, ApoA-I phenotype assays, dysfunctional HDL in inflammation), but the era in which "raising HDL" was a meaningful therapeutic goal is over.

4.5 What survives: reverse cholesterol transport as biology, not as a drug target

It is worth being precise about what has and has not failed. The biology of reverse cholesterol transport is real and important. ABCA1 and ABCG1 deficiency cause severe and informative phenotypes (Tangier disease and related). LCAT deficiency causes fish-eye disease and corneal opacification. HDL functional assays (cholesterol efflux capacity, Khera et al. 2011) predict CV risk better than HDL-C concentration does.⁷⁰ The pathway is biologically central to lipid homeostasis.

What has failed is the pharmacological strategy of raising the HDL-C number. That is a strategy, not a biology. The pathway is intact; the drug-development thesis was wrong. This distinction matters because in the debate with your Roche friend, the right framing is not "HDL doesn't matter" — it does, biologically — but "raising HDL-C as a therapeutic surrogate doesn't work, and the trial graveyard proves it." The lesson is generalisable: surrogate-endpoint medicine has to be validated against outcomes, every time, no exceptions. HDL-C looked too good to be wrong in the epidemiology; in pharmacology, it was. The same caution applies — though probably less strongly — to LDL-C, where the surrogate has been validated across far more drug classes and far more outcome trials.

4.6 Why this matters for Part II

The CETP and HDL story sets up the comparative reading we'll do in Part II of the statin trial era. The point is methodological: surrogate-endpoint inference is only as good as the validation of the surrogate. LDL-C has been validated, in the sense that drug-induced LDL-C lowering across statins, PCSK9 inhibitors, ezetimibe, and bempedoic acid produces predictable, proportional MACE reduction (the Cholesterol Treatment Trialists meta-analyses; the IMPROVE-IT, FOURIER, ODYSSEY, CLEAR Outcomes results). HDL-C has not been validated — drug-induced HDL-C raising does not produce predictable MACE reduction. The LDL surrogate works; the HDL surrogate does not. The asymmetry is large, and the debate about statins, properly understood, has to start from there.

Coda — what Part I has established

Part I has laid the structural pieces that the rest of the document will build on.

1. The mevalonate pathway is branched at FPP. The branches matter. The sceptic case for harms via CoQ10 and prenylation is biochemically coherent but quantitatively oversold in its strong form; it survives in pharmacogenomic outliers and at very high doses. The blinded trial evidence on CoQ10 supplementation argues against routine co-prescription.

2. Lipoprotein particles are ApoB-counted, not cholesterol-mass-measured. ApoB and LDL-C correlate in most patients and discord in a minority — and in that minority, ApoB is the more accurate atherogenicity metric. The Brown-Goldstein LDL-receptor cycle, and the PCSK9 modulation of it, supply the mechanistic story that lets us interpret all the modern drug classes.

3. Lp(a) is the genetically determined ApoB-bearing particle that standard statins barely touch. The 2024–2026 RNA-targeted therapies (pelacarsen, olpasiran, lepodisiran) are the substantive frontier. Whether they translate Lp(a) reduction into MACE reduction is the most consequential open question in 2026 lipidology.

4. Atherogenesis is best modelled as response-to-retention: ApoB-particle entry and subendothelial retention is the necessary first event; oxidation, foam-cell formation, smooth-muscle migration, and inflammation are downstream amplifications. The inflammation-versus-lipid debate is best framed as yes-and: CANTOS and LoDoCo2 establish that residual inflammation matters; FH, MR, and the trial dose-response establish that LDL retention is the necessary upstream substrate.

5. The HDL hypothesis, as a pharmacological strategy, has failed across CETP inhibitors, niacin add-on, and ApoA-I infusion. Reverse-cholesterol-transport biology is real; pharmacological HDL-C elevation as a therapeutic target is dead. Roche's dalcetrapib (dal-OUTCOMES, 2012) is the company's own contribution to this graveyard — a useful COI anchor for the conversation.

The rest of the document builds on this. Part II reads the trial era (4S, WOSCOPS, HPS, JUPITER, the CTT meta-analyses, IMPROVE-IT, FOURIER, ODYSSEY, CLEAR Outcomes, and the elderly-primary-prevention trials STAREE and PREVENTABLE due in 2025–2026). Part III handles the disputes (the JAMA Internal Medicine "curb our enthusiasm" critique, ARR-vs-RRR framing, women, very old patients, primary versus secondary prevention). Part IV handles the harms — myalgia, diabetes, cognitive, hepatic, the SAMSON n-of-1 nocebo data. Part V deals with industry and guideline politics. Part VI is the synthesis: what should a reasonable person actually do.

Footnotes

Part II-A — The Historical Trial Era (1953–2010)

A steelmanned narrative of how the cholesterol hypothesis was tested, contested, and consolidated, from Ancel Keys' pre-RCT epidemiology through the Cholesterol Treatment Trialists' meta-analytic era.

Reader's note. This chapter is the historical spine of the review. It traces the literature from the diet-heart hypothesis through the foundational statin trials, the mega-trial era, and the meta-analytic consolidation that produced today's consensus. The objective is not to "win" the cholesterol argument either way: it is to present, for each trial, the design, what was found, who paid, and what critics — heterodox and orthodox — have said about it. The companion Part II-B (chapters 9–12) carries the story forward into the ezetimibe, PCSK9, bempedoic acid and elderly-prevention eras that the 2024–2026 grounding file¹ tracks.

Chapter 5 — The Diet-Heart Hypothesis Era

5.1 The post-war puzzle

Coronary heart disease (CHD) was, in 1950, an epidemiological emergency without a settled cause. In the United States, age-adjusted CHD mortality had roughly tripled between 1920 and 1950 and was still rising; by 1955 it accounted for more than a third of all deaths in men aged 45–64.² President Eisenhower's myocardial infarction in September 1955 dramatised the problem politically — his cardiologist Paul Dudley White famously told the press the next day that the President's recovery depended on diet, weight and rest.³ The intellectual question was: why an epidemic of a disease that had been rare a generation earlier in apparently the same population?

Three rival explanations contended through the 1950s. The first was stress and Type-A behaviour, associated with Friedman and Rosenman in San Francisco. The second was tobacco, championed by Doll and Hill in Britain on the back of their 1950 lung-cancer paper.⁴ The third — the one that would eventually prevail — was diet, specifically saturated fat and serum cholesterol. The diet-heart hypothesis is most strongly associated with Ancel Keys, then Professor of Physiological Hygiene at the University of Minnesota, whose Minnesota Starvation Experiment of 1944–45 had already given him an international reputation for human-experimental nutrition science.⁵

5.2 Keys' Six-Country Graph and the Seven Countries Study

In 1953, Keys delivered a lecture at Mt Sinai Hospital in New York in which he plotted age-adjusted CHD mortality against percentage of dietary calories from fat for six countries: Japan, Italy, England-and-Wales, Australia, Canada and the United States.⁶ The points formed a near-perfect upward curve, and the graph became one of the most reproduced images in nutrition science.

This six-country chart is the founding artefact of the diet-heart hypothesis, and it is also the locus of the most-cited historical critique of Keys: that the Food and Agriculture Organization (FAO) had food-balance data for 22 countries in 1951, and that Keys selected the six that fit his hypothesis. In 1957, Jacob Yerushalmy (Berkeley biostatistician) and Herman Hilleboe (New York State Health Commissioner) published "Fat in the Diet and Mortality from Heart Disease: A Methodologic Note" in the New York State Journal of Medicine.⁷ Their re-plot using all 22 countries showed a weaker, though still positive, association between dietary fat and CHD, and showed that animal protein actually correlated more strongly with CHD mortality than fat did. They argued — and this is the methodologically substantive point that survives — that an ecological correlation across countries cannot distinguish a specific dietary cause from confounders such as overall economic development, sugar intake, smoking, or physical activity.

Keys responded to this criticism not rhetorically but logistically: from 1956 he organised what became the Seven Countries Study (SCS), a prospective cohort of about 12,763 men aged 40–59 across 16 cohorts in the United States, Finland, the Netherlands, Italy, the former Yugoslavia, Greece and Japan, with standardised baseline examinations and follow-up.⁸ The first major SCS publication in 1970 (American Heart Association monograph #29) reported a strong cross-cohort correlation between mean cohort saturated-fat intake, mean serum cholesterol, and 5-year and 10-year CHD incidence.⁹ At 25-year follow-up the across-cohort correlation between mean saturated-fat intake and CHD mortality was approximately r = 0.88; the cohort with the highest CHD rate (East Finland) had a mean serum cholesterol of about 7.1 mmol/L, and the cohort with the lowest (Crete) about 5.1 mmol/L.¹⁰

5.3 Steelmanning the Keys critique

The heterodox literature on Keys is now sizeable, much of it tracing back to Gary Taubes' Good Calories, Bad Calories (2007), Nina Teicholz's The Big Fat Surprise (2014), and more recent re-readings by Zoe Harcombe, Tim Noakes, and Aseem Malhotra.¹¹¹² The strongest version of the critique runs as follows.

First, country selection was not random and could not have been: Keys chose countries where his collaborators could plausibly mount field work, where dietary data were defensible, and where cholesterol could actually be measured by his Auto-Analyzer protocol.¹³ This is a real form of selection. France, for instance, with high saturated-fat intake and relatively low CHD ("the French paradox"), was not in the SCS.

Second, within-country, within-cohort analyses are weaker than the across-cohort headline finding. The Crete cohort, with low CHD, had high olive-oil (monounsaturated fat) intake but at the within-cohort level individual fat intake was not a strong predictor of individual CHD outcome — the strong association is between cohort means, not within-cohort individual differences.¹⁴ This ecological/individual distinction is real, and Keys' team did discuss it in the original SCS monographs, but it has not been adequately preserved in the popular summary of the study.

Third, the McGovern Senate Select Committee Dietary Goals for the United States (1977) translated the SCS findings into a public-health prescription — eat less saturated fat — that was politically pre-decided. The Committee staff, led by Nick Mottern (a journalist, not a nutrition scientist), drafted the goals partly in advance of the scientific testimony that was supposed to support them, and the Committee heard substantial dissent (notably from Robert Olson and George Mann) that did not survive into the final document.¹⁵ The 1980 Dietary Guidelines for Americans and the 1985 NIH Consensus Conference on Lowering Blood Cholesterol then locked the saturated-fat-causes-CHD hypothesis into public health doctrine.¹⁶

Fourth, and most pointedly, the Minnesota Coronary Experiment (MCE; 1968–73; 9,423 institutionalised patients in mental hospitals and a nursing home, randomised to a vegetable-oil-rich diet vs control) was published in summary form in 1989 but the full data — including the finding that the vegetable-oil arm lowered cholesterol but did not reduce mortality — was only fully recovered and analysed in 2016 by Christopher Ramsden's group at NIH.¹⁷ Ramsden et al found that for every 30 mg/dL fall in serum cholesterol on the vegetable-oil diet, there was a 22% higher risk of death — a striking inversion of what the hypothesis would predict. Importantly Keys was a co-investigator on the MCE. The 2016 reanalysis is now used by both camps: heterodox critics as evidence that Keys knew about a non-confirmatory trial; consensus defenders as evidence that the MCE was underpowered, used a now-discredited oil (very high linoleic acid, possibly oxidised), and concerned an institutionalised population whose churn made follow-up unreliable.

5.4 Adjudication of the Keys era

The honest reading is that Keys was directionally right but his evidence base was weaker than later canonised. The 25-year SCS finding that across-cohort saturated-fat intake correlates strongly with across-cohort CHD rates is real and has not been overturned. The 1957 Yerushalmy-Hilleboe critique is also real, and never satisfactorily refuted at the methodological level — ecological correlations cannot prove individual causality. The MCE reanalysis genuinely complicates the simple "lower cholesterol via vegetable oils → lower mortality" story.

What rescues the Keys-era hypothesis, retrospectively, is not the 1950s–70s evidence base; it is everything that came after: the 1984 LRC-CPPT (5.5 below), the 1994 4S trial (chapter 6), the FH genetics literature, and the Mendelian randomisation work covered in the grounding file.¹ Without those later proofs of cholesterol → CHD causality, Keys' diet-heart hypothesis would be a contested ecological hypothesis. With them, the directional claim — that high serum cholesterol is a cause of CHD — is now about as well-supported as any claim in chronic-disease epidemiology. The question of whether dietary saturated fat specifically is the cholesterol-raising lever, and the question of how much dietary intervention helps an individual, remain considerably less settled — see chapter 16 of this review.

5.5 Framingham — the within-population proof

While Keys was building his between-country case, the Framingham Heart Study was building the within-population case. Initiated in 1948 under the auspices of the US Public Health Service, Framingham enrolled 5,209 men and women aged 30–62 from the town of Framingham, Massachusetts, and followed them at biennial examinations. Thomas Royle Dawber assumed direction in 1950.¹⁸

The 1961 Annals of Internal Medicine paper "Factors of risk in the development of coronary heart disease — six-year follow-up experience" introduced the phrase "risk factor" to medicine and identified serum cholesterol, blood pressure and electrocardiographic abnormality as independent predictors of CHD.¹⁹ By the 1971 6-year follow-up, total cholesterol was associated with a graded increase in CHD risk: men with cholesterol >260 mg/dL had roughly 3–5× the CHD rate of men with cholesterol <200 mg/dL.²⁰ The Offspring Cohort, initiated in 1971, replicated and extended these findings, and the Framingham Risk Score (Wilson 1998) became the canonical clinical risk-prediction tool for two decades.²¹

Framingham was not an RCT and could not establish causality; what it did establish is the graded, continuous, within-population relationship between cholesterol and CHD risk in men below age 50. The findings were weaker in older adults and weaker in women, and an awkward post-1987 Framingham paper actually found that falling cholesterol predicted higher mortality in the elderly — a finding the heterodox literature has used heavily.²²

5.6 The Coronary Drug Project (CDP)

The Coronary Drug Project, sponsored by the NHLBI, randomised 8,341 men aged 30–64 with prior myocardial infarction (MI) to one of five active arms — clofibrate, niacin (nicotinic acid 3g/day), low-dose oestrogen, high-dose oestrogen, dextrothyroxine — or placebo, between 1966 and 1975.²³ Three of the five arms were stopped early for harm: high-dose oestrogen for increased non-fatal cardiovascular events, low-dose oestrogen for increased cancer, dextrothyroxine for increased mortality. Clofibrate failed to reduce mortality and was effectively discontinued for that indication. The fifth arm — niacin — showed no significant overall mortality benefit at the end of the in-trial period in 1975, but a 26% reduction in non-fatal MI (p<0.005).²⁴

The trial would have been a footnote had the investigators not, in 1986, published a 15-year post-trial mortality follow-up — nine years after the in-trial period ended and study drugs were discontinued. Canner et al. found that mortality in the niacin arm was 11% lower than in the placebo arm at 15 years (52.0% vs 58.2%; p = 0.0004).²⁵ This is one of the cleanest demonstrations in cardiovascular pharmacology of a "legacy effect" — a benefit that persists long after the intervention itself is stopped. It is also a striking vindication of the cholesterol hypothesis: the only one of five active CDP arms that worked was the one that lowered LDL-C (and raised HDL-C), and the benefit accumulated over time. The relevance of the CDP niacin signal to modern practice is contested — later niacin trials (AIM-HIGH 2011, HPS2-THRIVE 2014) failed to show benefit on a background of statins (chapter 9 in Part II-B) — but as a 1980s anchor for the cholesterol hypothesis, the CDP niacin result is genuinely strong.

5.7 The Lipid Research Clinics Coronary Primary Prevention Trial (LRC-CPPT)

If one trial anchored the cholesterol consensus prior to statins, it is the LRC-CPPT. Funded by the NHLBI, the trial randomised 3,806 asymptomatic middle-aged men with type II hyperlipoproteinemia (LDL-C ≥190 mg/dL after diet) to cholestyramine 24g/day vs placebo. Mean follow-up was 7.4 years. The primary endpoint was CHD death or definite non-fatal MI.²⁶

Results: cholestyramine reduced LDL-C by 20.3% vs 12.6% on placebo (a between-group LDL-C difference of about 7.7%). The primary endpoint occurred in 7.0% of the cholestyramine arm vs 8.6% of the placebo arm — a 19% relative reduction (p<0.05). Hazard ratio for CHD death + non-fatal MI ~0.81; ARR ~1.6%; NNT ~62 over 7.4 years.²⁶ All-cause mortality showed a non-significant 7% relative reduction.

Three things made LRC-CPPT historically pivotal. First, it was the first large RCT of pharmacological cholesterol lowering with a clinically meaningful endpoint. Second, the investigators argued — controversially at the time — that the within-trial CHD reduction was roughly proportional to the LDL-C reduction achieved, generalising to the prediction that a 25% LDL-C reduction would produce a 50% CHD reduction. Third, it triggered the National Cholesterol Education Program (NCEP), launched by the NHLBI in 1985 with the first NCEP Adult Treatment Panel report in 1988 setting target LDL thresholds.²⁷

LRC-CPPT had real limitations. The drug effect was modest — a 7.7% absolute LDL-C reduction over placebo, partly because cholestyramine compliance was poor (the average dose actually taken was ~14g, not the prescribed 24g, because the resin is unpalatable). The primary endpoint p-value (0.05) was achieved only by a one-tailed test, which was unusual even by 1984 standards. The trial enrolled only men. And — the steelmanned critique — the LRC investigators arguably overstated the trial's implications by linearly extrapolating its results to all patients and all LDL-C reductions. The 1985 NIH Consensus Conference on Lowering Blood Cholesterol, which cemented LDL-C as a treatment target, has been characterised by skeptics (notably Paul Rosch and Uffe Ravnskov) as having pre-determined conclusions and as having relied on a generalisation from LRC-CPPT that the trial itself could not strictly support.²⁸ On the orthodox side, defenders point out that subsequent statin trials decisively confirmed the LRC-CPPT signal, and that the LRC investigators' generalisation turned out to be approximately right. We will return to this in the CTT chapter.

5.8 What chapter 5 establishes

By the late 1980s, the cholesterol hypothesis rested on: - Ecological/cross-country data (Keys, Seven Countries) — directionally supportive, methodologically weak. - Within-population graded risk data (Framingham, MRFIT screen population²⁹) — strong but observational. - Genetic data on familial hypercholesterolemia (Brown and Goldstein's LDL receptor work, Nobel 1985)³⁰ — strong causal evidence for the high end of the distribution. - One large positive RCT (LRC-CPPT) — modest effect, with bile-acid sequestrant, in highly selected men. - One striking legacy-effect RCT (CDP niacin 15-year follow-up).

The intellectual case for cholesterol-as-causal-of-CHD was probable, not settled. What it lacked was a drug effective and tolerable enough to test the hypothesis decisively. That drug was the statin.

Chapter 6 — The First Statin Era (1994–1998)

6.1 The arrival of the HMG-CoA reductase inhibitors

Akira Endo isolated mevastatin (ML-236B, compactin) from Penicillium citrinum in 1973 at Sankyo, and Alfred Alberts at Merck independently isolated lovastatin from Aspergillus terreus in 1978.³¹ Lovastatin (Mevacor) was the first statin approved, in 1987, on the strength of LDL-C reduction (~30%) without a clinical-endpoint trial. Simvastatin (Zocor) followed in 1991, pravastatin (Pravachol) in 1991, fluvastatin in 1993, atorvastatin (Lipitor) in 1996, cerivastatin (Baycol) in 1997 (withdrawn 2001), rosuvastatin (Crestor) in 2003, and pitavastatin (Livalo) in 2009.

The statins reduced LDL-C by 25–55% with reasonable tolerability — vastly more than diet or bile-acid sequestrants — and so could finally test the cholesterol hypothesis with adequate effect size. Five trials between 1994 and 1998 did that test.

6.2 The Scandinavian Simvastatin Survival Study (4S, 1994)

Design. Multicentre, double-blind, placebo-controlled RCT. 4,444 patients (men and women) aged 35–70 with history of angina pectoris or prior MI and serum cholesterol 5.5–8.0 mmol/L (212–309 mg/dL) on a lipid-lowering diet. Randomised to simvastatin 20 mg/day (titrated to 40 mg if cholesterol target not met) vs placebo. Median follow-up 5.4 years. Investigator-initiated by Terje Pedersen and the Scandinavian Simvastatin Study Group; funded by Merck.³²

Primary endpoint. All-cause mortality.

Results. Total mortality 8% (182/2,221) on simvastatin vs 12% (256/2,223) on placebo. Relative risk (RR) 0.70 (95% CI 0.58–0.85; p = 0.0003). ARR 3.3% over 5.4 years; NNT ≈ 30. Coronary mortality RR 0.58 (95% CI 0.46–0.73). Major coronary events RR 0.66 (95% CI 0.59–0.75). No increase in non-cardiovascular mortality, including cancer.³²

Funding and conflicts. Investigator-initiated; protocol developed by the Scandinavian Simvastatin Study Group (chaired by Terje Pedersen, Oslo); Merck funded the trial and supplied drug. Data analysis was performed by the Scandinavian investigators with Merck statistical support; the trial used independent data monitoring.

Controversies and heterodox readings. 4S is the trial that the consensus camp considers the strongest single piece of evidence for statin therapy, and the steelmanned heterodox case against it is correspondingly weak. The most substantive criticisms run as follows:

1. The population was secondary prevention with elevated cholesterol — 4S patients had mean baseline total cholesterol of 6.75 mmol/L (261 mg/dL) and prior coronary disease. This is the patient population most likely to benefit; the trial cannot itself be extrapolated to primary prevention or to lower-cholesterol patients (and was not, until subsequent trials did so).
2. Industry sponsorship — Merck funded the trial and analytics, although the Scandinavian Simvastatin Study Group retained academic control. By 1990s standards 4S was unusually clean; by current standards (CONSORT, ICMJE data-sharing requirements), it would be expected to share IPD, which it has not. The dataset is held by Merck and the original investigators.
3. The mortality reduction may be driven by something other than LDL-C — the heterodox camp (Ravnskov, Kendrick, de Lorgeril) has at various points argued that 4S's benefit may reflect statins' "pleiotropic" effects (anti-inflammatory, antithrombotic) rather than LDL-C lowering per se. This is plausible biologically but does not undermine the trial's clinical relevance: statins, as a class, reduce mortality in this population. Whether the mechanism is purely LDL-C lowering is a separate question, addressed by Mendelian randomisation and PCSK9 trials in later chapters.

Assessment. 4S is the foundational statin trial. Its NNT of ~30 over 5.4 years for all-cause mortality in secondary prevention is one of the largest effect sizes for any chronic-disease intervention in cardiovascular medicine. The 10-year mortality follow-up published in 2004 confirmed the effect persisted, with no excess cancer signal.³³ If the cholesterol hypothesis is wrong, 4S is unexplained. The strongest version of the heterodox position has to grant 4S and then argue about generalisability — and that is, I think, the correct version of the debate.

6.3 The West of Scotland Coronary Prevention Study (WOSCOPS, 1995)

Design. Multicentre, double-blind, placebo-controlled RCT. 6,595 men aged 45–64 from west Scotland with mean plasma cholesterol 272 ± 23 mg/dL, LDL-C ~192 mg/dL, and no prior MI (primary prevention, though high-risk by current standards — about 5% had a history of angina, claudication, or stroke). Randomised to pravastatin 40 mg/day vs placebo. Mean follow-up 4.9 years. Funded by Bristol-Myers Squibb.³⁴

Primary endpoint. Definite CHD death or definite non-fatal MI.

Results. Primary endpoint 5.5% on pravastatin vs 7.9% on placebo; RR 0.69 (95% CI 0.57–0.83; p<0.001). ARR 2.4%; NNT ≈ 42 over 5 years. All-cause mortality RR 0.78 (95% CI 0.60–0.99; p = 0.039) — a 22% relative reduction with borderline significance. No excess cancer; non-CV mortality non-significantly lower on pravastatin.³⁴

Funding and conflicts. Industry-funded (BMS), but executed by an academic group (West of Scotland Coronary Prevention Study Group, led by James Shepherd at Glasgow). The trial is widely regarded as methodologically clean.

Controversies and heterodox readings. WOSCOPS is more contested than 4S, on two main grounds:

1. Population was high-risk primary prevention, not low-risk primary prevention — the WOSCOPS cohort had a baseline 5-year CHD risk of about 8%, comparable to many secondary-prevention cohorts of the time. Extrapolating WOSCOPS to lower-risk primary prevention populations is therefore not straightforward — and the AFCAPS/TexCAPS trial (below) is needed to address that gap.
2. The all-cause mortality reduction has been re-examined — WOSCOPS' 22% all-cause mortality reduction was based on a small number of total deaths (106 on pravastatin vs 135 on placebo). The 10-year (Ford 2007 NEJM) and 15-year follow-ups in the West of Scotland have continued to show legacy benefit, with the original-pravastatin arm showing lower CV mortality even after most patients in both arms had crossed over to statins post-trial.³⁵ This is consistent with the FOURIER-OLE finding in the 2024 grounding (i.e. "earlier is better"). The heterodox reading — that WOSCOPS' mortality reduction is statistical noise — has not survived the long-term follow-up.

Assessment. WOSCOPS established that statin therapy works in primary prevention, at least in high-risk men, and produced the second clean CV mortality signal (after 4S). The 15- and 20-year follow-ups have if anything strengthened the original conclusion. The trial does not, by itself, establish that statins should be used in lower-risk primary prevention.

6.4 The Cholesterol and Recurrent Events Trial (CARE, 1996)

Design. Multicentre, double-blind, placebo-controlled RCT. 4,159 patients (3,583 men, 576 women) with prior MI 3–20 months before randomisation and total cholesterol <240 mg/dL (mean 209 mg/dL; mean LDL-C 139 mg/dL — i.e. average cholesterol for an MI population, not elevated). Randomised to pravastatin 40 mg/day vs placebo. Mean follow-up 5.0 years. Funded by Bristol-Myers Squibb.³⁶

Primary endpoint. Fatal coronary event or non-fatal MI.

Results. Primary endpoint 10.2% on pravastatin vs 13.2% on placebo; RR 0.76 (95% CI 0.64–0.91; p = 0.003). ARR 3.0%; NNT ≈ 33. CABG -26%, PCI -23%, stroke -31%. No significant reduction in all-cause mortality (RR 0.91, 95% CI 0.74–1.12). No excess cancer overall, though a numerically increased breast cancer count in the pravastatin arm (12 vs 1) attracted comment; this signal did not reproduce in subsequent trials or in the long-term CARE follow-up.³⁶³⁷

Funding. BMS-funded; executed by a Brigham-led academic group (Sacks, Pfeffer, Braunwald).

Controversies and heterodox readings. CARE matters less for its absolute results than for what it established conceptually: statins reduce events even in patients with average baseline cholesterol. This was the first chink in the "high-cholesterol-only" framing and prefigured the Heart Protection Study (chapter 7) and the eventual shift to "treat the patient, not the lipid."

The CARE breast-cancer signal was used by Ravnskov and others to argue for a statin-cancer link.³⁸ The signal did not replicate, including in the long-term CARE follow-up nor in pooled CTT analyses, and is now generally regarded as a chance finding in a small subgroup. The heterodox reading — that small adverse signals in industry trials are systematically downplayed — has more force as a meta-claim than as a specific charge against CARE.

6.5 The Long-Term Intervention with Pravastatin in Ischaemic Disease Trial (LIPID, 1998)

Design. Multicentre, double-blind, placebo-controlled RCT in Australia and New Zealand. 9,014 patients aged 31–75 with prior MI or hospitalised unstable angina and total cholesterol 155–271 mg/dL (mean ~218 mg/dL). Randomised to pravastatin 40 mg/day vs placebo. Mean follow-up 6.1 years. Funded by Bristol-Myers Squibb.³⁹

Primary endpoint. CHD mortality.

Results. CHD mortality 6.4% on pravastatin vs 8.3% on placebo; RR 0.76 (95% CI 0.65–0.88; p<0.001). All-cause mortality 11.0% on pravastatin vs 14.1% on placebo; RR 0.78 (95% CI 0.69–0.87; p<0.001). ARR for total mortality 3.1%; NNT ≈ 32. MI -29%, stroke -19%, revascularisation -20%. No excess cancer.³⁹

Funding. BMS-funded; academic execution (LIPID Study Group, chaired by Andrew Tonkin).

Controversies. LIPID is the broadest secondary-prevention trial — it included patients with average and even relatively low cholesterol, and confirmed CARE's finding that statins work across the cholesterol range in patients with established disease. It is uncontroversial in the consensus literature. The heterodox literature has less material to work with on LIPID; the main argument is the meta-claim that the pravastatin trials as a group were over-promoted relative to their effect size, which is moderate (pravastatin lowers LDL-C by 25–30%, less than higher-potency statins).

6.6 The Air Force/Texas Coronary Atherosclerosis Prevention Study (AFCAPS/TexCAPS, 1998)

Design. Multicentre, double-blind, placebo-controlled RCT. 6,605 patients (5,608 men, 997 women) with no clinically evident atherosclerotic disease, mean LDL-C 150 mg/dL (3.89 mmol/L) and below-average HDL-C. Randomised to lovastatin 20–40 mg/day vs placebo. Mean follow-up 5.2 years. Funded by Merck.⁴⁰

Primary endpoint. First acute major coronary event (fatal or non-fatal MI, unstable angina, or sudden cardiac death).

Results. Primary endpoint 3.5% on lovastatin vs 5.5% on placebo; RR 0.63 (95% CI 0.50–0.79; p<0.001). ARR 2.0%; NNT ≈ 50 over 5.2 years. No significant reduction in all-cause mortality (80 deaths on lovastatin vs 77 on placebo; RR 1.04).⁴⁰ Lovastatin reduced LDL-C by 25% to 115 mg/dL.

Funding. Industry-funded (Merck); academic execution (Downs, Clearfield et al.).

Controversies and heterodox readings. AFCAPS/TexCAPS is the most contested of the five "classical" statin trials. The criticisms:

1. No mortality benefit. In a low-risk population (10-year predicted CHD risk ~10%), AFCAPS/TexCAPS found no all-cause mortality reduction despite a clear reduction in coronary events. The heterodox reading — that primary prevention with statins shifts events from coronary to non-coronary causes without prolonging life — is most strongly supported by AFCAPS/TexCAPS taken in isolation.
2. Effect size is modest in absolute terms. ARR of 2% over 5 years (NNT 50) for a composite endpoint that included unstable angina is meaningfully smaller than the secondary-prevention trials. For an asymptomatic 50-year-old with LDL-C 150 mg/dL, the marginal benefit of statin therapy is real but small.
3. The Byrne et al. JAMA Internal Medicine 2022 ("Curb our enthusiasm") meta-analysis pointed at primary prevention specifically⁴¹ — drawing on AFCAPS/TexCAPS and similar trials to argue that absolute primary-prevention benefits are small enough that shared decision-making, rather than universal prescription, is appropriate. The grounding file's note on this — that AHA PREVENT 2024 has further reduced the population estimated to benefit — is consistent with the AFCAPS/TexCAPS reading.¹

Assessment. AFCAPS/TexCAPS established that statins work in low-to-average-LDL primary prevention to reduce coronary events, but did not establish a mortality benefit at this risk level. This is the trial that anchors the "treat-the-risk-not-the-LDL" caution at the lower-risk end of the spectrum, and reasonable physicians have read its results in opposing ways for 25 years.

6.7 What chapter 6 establishes

By 1998, the cholesterol hypothesis had been tested with statin RCTs in: - Secondary prevention with elevated cholesterol (4S) — strong all-cause mortality benefit. - Secondary prevention with average cholesterol (CARE, LIPID) — strong CV event and (LIPID) mortality benefit. - High-risk primary prevention with elevated cholesterol (WOSCOPS) — clear CV event and (borderline) mortality benefit. - Low-to-average-LDL primary prevention (AFCAPS/TexCAPS) — clear coronary event reduction, no mortality benefit.

This is the period in which the consensus position consolidated. The CTT meta-analyses (chapter 8) would later quantify the relationship as "per 1 mmol/L LDL-C reduction, ~22% relative risk reduction in major vascular events."

Chapter 7 — The Mega-Trial Era (2002–2008)

7.1 The shift from "treat the lipid" to "treat the patient"

The 1994–1998 trials were essentially cholesterol-stratified: enrolment was by total cholesterol or LDL-C threshold. The 2002–2008 mega-trial era inverted this logic. Investigators began enrolling on the basis of cardiovascular risk (diabetes, hypertension, age, prior MI) rather than baseline lipid level, and the question shifted to "does statin therapy benefit high-risk patients regardless of their starting cholesterol?" The Heart Protection Study answered yes.

7.2 The MRC/BHF Heart Protection Study (HPS, 2002)

Design. Multicentre, double-blind, placebo-controlled, 2×2 factorial RCT (simvastatin vs placebo; vitamin antioxidant cocktail vs placebo). 20,536 UK adults aged 40–80 with coronary disease, other occlusive arterial disease (carotid or PAD), or diabetes, plus baseline total cholesterol ≥3.5 mmol/L (135 mg/dL) — a deliberately low threshold designed to test whether benefit extended to patients without elevated cholesterol. Simvastatin 40 mg/day vs matching placebo. Mean follow-up 5.0 years. Funded by UK Medical Research Council, British Heart Foundation, and Merck (drug supply; partial funding).⁴²

Primary endpoints. All-cause mortality (overall); fatal or non-fatal vascular events (between subdivisions).

Results. All-cause mortality 12.9% on simvastatin vs 14.7% on placebo; RR 0.87 (95% CI 0.81–0.94; p = 0.0003). Coronary death 5.7% vs 6.9%, RR 0.82 (95% CI 0.75–0.91; p = 0.0005). Any major vascular event (coronary death, non-fatal MI, stroke, revascularisation) 19.8% vs 25.2%; RR 0.76 (95% CI 0.72–0.81; p<0.0001). ARR for major vascular events ~5.4%; NNT ≈ 19 over 5 years. The vitamin arm showed no benefit on any endpoint.⁴²

Crucially, benefit was consistent across all subgroups stratified by baseline LDL-C — including patients with LDL-C <3.0 mmol/L (116 mg/dL) and <2.6 mmol/L (100 mg/dL). HPS is the trial that establishes the "treat the patient, not the lipid" framing.

Funding. Public funding (MRC + BHF) with industry drug supply. The Clinical Trial Service Unit at Oxford (Rory Collins, Richard Peto, Jane Armitage, Colin Baigent) led design and execution; this same group would go on to lead the Cholesterol Treatment Trialists' Collaboration (chapter 8).

Controversies and heterodox readings. HPS is the bedrock of modern statin practice and is correspondingly the heaviest target of the heterodox literature:

1. The CTSU/Oxford group's control of HPS data is the same control they exercise over the CTT meta-analyses. HPS individual-patient data has not been openly released; this is the same transparency complaint that runs through chapter 8.
2. Subgroup analyses by baseline LDL-C are observational within an RCT — comparing patients with low baseline LDL-C to those with high baseline LDL-C is not the same as comparing randomly assigned treatments. The HPS investigators acknowledge this, and the headline "benefit irrespective of baseline LDL-C" is a within-trial subgroup interpretation that is mechanistically supported but not protocol-prespecified at the level the headline implies.
3. The diabetes subgroup analysis (HPS-Diabetes, 2003) prefigured the CARDS trial and is sometimes cited as a stealth indication expansion — a 2002 trial designed for established CV disease was used to justify statin treatment of diabetics without prior CV events.⁴³
4. The vitamin arm result is rarely emphasised. HPS is one of the largest antioxidant-vitamin RCTs ever conducted and showed no benefit on any CV or mortality endpoint. This is a useful epistemic anchor: it shows the CTSU was willing to publish null results from a contemporaneous, lower-priority arm of the same trial.

Assessment. HPS is the strongest single trial in the statin literature for population breadth — 20,536 patients, 5 years of follow-up, a clear all-cause mortality reduction, and benefit across LDL strata. The transparency criticism is legitimate but does not undermine the topline. The conceptual shift from cholesterol thresholds to risk-based prescription is HPS's most important legacy.

7.3 The Anglo-Scandinavian Cardiac Outcomes Trial — Lipid Lowering Arm (ASCOT-LLA, 2003)

Design. Multicentre, double-blind, placebo-controlled RCT, nested inside ASCOT-BPLA (a hypertension trial). 10,305 hypertensive patients aged 40–79 with at least 3 other CV risk factors but no prior CHD, and total cholesterol ≤6.5 mmol/L (250 mg/dL). Randomised to atorvastatin 10 mg/day vs placebo. Planned 5-year follow-up; stopped early at median 3.3 years on the recommendation of the Data Safety Monitoring Board. Funded by Pfizer.⁴⁴

Primary endpoint. Non-fatal MI + fatal CHD.

Results. Primary endpoint 1.9% on atorvastatin vs 3.0% on placebo; HR 0.64 (95% CI 0.50–0.83; p = 0.0005). ARR 1.1%; NNT ≈ 91 over 3.3 years. Stroke HR 0.73 (95% CI 0.56–0.96; p = 0.024). All-cause mortality not significantly different (HR 0.87; 95% CI 0.71–1.06; p = 0.16).⁴⁴ Atorvastatin lowered LDL-C by ~1 mmol/L vs placebo.

Funding. Industry-funded (Pfizer).

Controversies and heterodox readings. ASCOT-LLA is the canonical example for the early stopping for benefit critique:

1. Stopping early inflates effect estimates. The Bassler et al. JAMA 2010 meta-analysis of trials stopped early for benefit found such trials systematically overestimate effects, often by 20–30%.⁴⁵ ASCOT-LLA was stopped at 3.3 years against a planned 5-year follow-up; the effect estimate is therefore plausibly inflated.
2. No mortality signal. Like AFCAPS/TexCAPS in primary prevention, ASCOT-LLA found no all-cause mortality reduction. The trial demonstrated coronary event reduction but not life extension at the 3.3-year horizon.
3. Pfizer commercial interest. ASCOT-LLA was used by Pfizer to expand the atorvastatin (Lipitor) indication into hypertension — a marketing-strategic outcome the trial design plausibly anticipated. This is a real conflict; it does not, by itself, falsify the trial's clinical findings, but it does mean the trial should be read alongside the published-vs-completed-trial analyses of statin-era publication bias (chapter 11 in Part II-B).
4. The 2017 ASCOT-LLA extension — Gupta et al. Lancet 2017 reported that adverse events attributed to atorvastatin were no higher than on placebo in the blinded phase, but rose significantly in the open-label extension when patients knew they were on statin — a striking nocebo finding.⁴⁶ This is one of the most-cited pieces of evidence for the "muscle pain is largely nocebo" position in chapter 14 (Part II-B).
5. The 2024 ASCOT-Legacy 20-year follow-up found persistently lower mortality in the originally-allocated atorvastatin arm despite both arms now on statins.⁴⁷ This is, in the strongest reading, evidence that the original benefit was real and durable — consistent with the FOURIER-OLE legacy effect in the grounding file.¹

Assessment. ASCOT-LLA established statin benefit in hypertensive primary-prevention patients but did so on a contracted timeline that probably inflated the headline estimate. The 20-year legacy follow-up rescues the trial's substantive conclusion: atorvastatin 10 mg in this population reduces lifetime CV risk. The early-stopping criticism is methodologically correct and has been formally answered by the duration of subsequent follow-up.

7.4 The PROspective Study of Pravastatin in the Elderly at Risk (PROSPER, 2002)

Design. Multicentre, double-blind, placebo-controlled RCT. 5,804 men and women aged 70–82 with either established vascular disease (secondary prevention) or significant CV risk factors (primary prevention). Randomised to pravastatin 40 mg/day vs placebo. Mean follow-up 3.2 years. Funded by Bristol-Myers Squibb.⁴⁸

Primary endpoint. Composite of coronary death, non-fatal MI, and fatal/non-fatal stroke.

Results. Primary endpoint HR 0.85 (95% CI 0.74–0.97; p = 0.014). ARR 2.1%; NNT ≈ 48 over 3.2 years. Driven primarily by reduction in CHD events; stroke was not significantly reduced. No reduction in all-cause mortality (HR 0.97; 95% CI 0.83–1.14). A signal of increased new cancer diagnoses in the pravastatin arm — 245 (8.5%) vs 199 (6.8%); HR 1.25 (95% CI 1.04–1.51; p = 0.02).⁴⁸

Funding. Industry-funded (BMS).

Controversies and heterodox readings. PROSPER is heavily used by the heterodox literature on three grounds:

1. The cancer signal. The 25% relative increase in new cancers on pravastatin was statistically significant and biologically plausible (statins inhibit mevalonate pathway intermediates relevant to cell proliferation). A 2007 meta-analysis (Bonovas et al.) found no overall pravastatin-cancer link but reported age-modification of effect — older patients showed greater cancer risk on statins.⁴⁹ The CTT 2010 meta-analysis pooled across all statin trials and found no significant cancer excess; subsequent long-term follow-ups have generally been reassuring; but the PROSPER signal has not been definitively explained.
2. No mortality benefit. In an elderly cohort with mixed primary and secondary prevention, pravastatin produced a modest event reduction but did not extend life. This is consistent with the broader pattern in elderly trials and is the rationale for the STAREE and PREVENTABLE trials covered in the grounding file.¹
3. No stroke reduction. The lack of stroke benefit, in contrast with HPS (which showed a 25% stroke reduction), is sometimes attributed to pravastatin's lower potency vs simvastatin, but the absence is real.

Assessment. PROSPER is the most equivocal of the mega-trial-era results. It establishes some event reduction in the elderly but not mortality reduction and produced a cancer signal that has been argued about for two decades. The 2026 readouts from STAREE and PREVENTABLE will be the next-generation answer to the question PROSPER raised but did not settle.

7.5 PROVE-IT TIMI 22 (2004) — "Lower is better" is born

Design. Multicentre, double-blind, randomised, active-controlled RCT. 4,162 patients hospitalised for acute coronary syndrome (ACS) within the preceding 10 days. Randomised to atorvastatin 80 mg/day (intensive) vs pravastatin 40 mg/day (moderate). Mean follow-up 24 months. Funded by Bristol-Myers Squibb (Pravachol) and Sankyo (in part).⁵⁰

Primary endpoint. Composite of all-cause death, MI, documented unstable angina requiring rehospitalisation, revascularisation >30 days post-randomisation, or stroke.

Results. Primary endpoint 22.4% on atorvastatin 80 mg vs 26.3% on pravastatin 40 mg; HR 0.84 (95% CI 0.74–0.95; p = 0.005). 16% relative reduction. ARR 3.9%; NNT ≈ 26. Median achieved LDL-C: 62 mg/dL on atorvastatin 80 vs 95 mg/dL on pravastatin 40 (both arms started at ~106 mg/dL). The intensive arm had higher transaminase elevations (3.3% vs 1.1%).⁵⁰

Funding. Industry-funded (BMS, the pravastatin manufacturer, ran the trial — an unusual design choice, as the sponsor's own drug was the comparator that lost).

Controversies and heterodox readings. PROVE-IT is the trial that birthed the "lower is better" doctrine and the LDL-C <70 mg/dL target for high-risk patients:

1. Active-controlled, not placebo-controlled. PROVE-IT compares two statin regimens; it does not show that statin therapy is better than no therapy. It addresses a different question — whether more LDL-C lowering is better than less — and the answer was yes.
2. The 24-month follow-up is short. A 4% absolute difference at 24 months is meaningful but is not yet a long-term outcome. The trial established the principle that would be tested at longer horizons by TNT and IDEAL.
3. The benefit may not be a pure LDL effect. Atorvastatin lowered hs-CRP more than pravastatin did, and Morrow et al.'s sub-analysis suggested that the difference in CRP, not the difference in LDL, predicted differential outcome.⁵¹ Ridker would build on this to design JUPITER (next).

Assessment. PROVE-IT is the empirical foundation of the intensive-statin doctrine. It shifted clinical practice from "achieve LDL-C <100" to "lower is better" — a principle that PCSK9 trials would extend further still.

7.6 Treating to New Targets (TNT, 2005)

Design. Multicentre, double-blind, randomised, active-controlled RCT. 10,001 patients with stable CHD and LDL-C <130 mg/dL on atorvastatin 10 mg/day open-label run-in. Randomised to atorvastatin 10 mg/day vs atorvastatin 80 mg/day. Median follow-up 4.9 years. Funded by Pfizer.⁵²

Primary endpoint. First major cardiovascular event (CHD death, non-fatal MI, resuscitated cardiac arrest, fatal/non-fatal stroke).

Results. Primary endpoint 8.7% on atorvastatin 80 vs 10.9% on atorvastatin 10; HR 0.78 (95% CI 0.69–0.89; p<0.001). ARR 2.2%; NNT ≈ 45. Mean achieved LDL-C: 77 mg/dL (80 mg) vs 101 mg/dL (10 mg). All-cause mortality not significantly different. Hepatic enzyme elevation 1.2% vs 0.2%.⁵²

Controversies. TNT confirmed PROVE-IT at longer follow-up and in stable disease, but in a Pfizer-on-Pfizer comparison. The trial was instrumental in lowering the LDL-C target to <70 mg/dL for high-risk patients in the 2004 NCEP ATP III update.⁵³

7.7 Incremental Decrease in Endpoints Through Aggressive Lipid Lowering (IDEAL, 2005)

Design. Multicentre, prospective, randomised, open-label, blinded-endpoint (PROBE) RCT. 8,888 patients aged ≤80 with prior MI. Randomised to atorvastatin 80 mg/day vs simvastatin 20 mg/day (titrated to 40 mg). Median follow-up 4.8 years. Funded by Pfizer.⁵⁴

Primary endpoint. Major coronary event (CHD death, non-fatal MI, resuscitated arrest).

Results. Primary endpoint 9.3% on atorvastatin 80 vs 10.4% on simvastatin 20–40; HR 0.89 (95% CI 0.78–1.01; p = 0.07). Did not achieve statistical significance for the primary endpoint. Secondary endpoints including major CV event were reduced (HR 0.87; p = 0.02). All-cause mortality identical (8.2% vs 8.4%; p = NS).⁵⁴

Controversies. IDEAL was a Pfizer-sponsored trial that failed its primary endpoint. The investigators argued — reasonably — that secondary endpoints supported the intensive regimen, and the result is broadly consistent with PROVE-IT and TNT. But IDEAL is also the trial that demonstrates the limits of the "lower is better" doctrine: at the upper end of intensification, returns diminish.

7.8 The JUPITER Trial (2008) — the most contested statin trial

Design. Multicentre, double-blind, placebo-controlled RCT. 17,802 men ≥50 and women ≥60 with LDL-C <130 mg/dL (mean 108 mg/dL), hs-CRP ≥2.0 mg/L, and no prior CV disease. Randomised to rosuvastatin 20 mg/day vs placebo. Planned 5-year follow-up; stopped early at median 1.9 years on the recommendation of the DSMB. Funded by AstraZeneca. Principal Investigator: Paul Ridker (Brigham and Women's Hospital), who held a patent (with co-applicants) on the use of hs-CRP for cardiovascular risk prediction.⁵⁵

Primary endpoint. Composite of MI, stroke, hospitalisation for unstable angina, arterial revascularisation, or CV death.

Results. Primary endpoint 0.77 vs 1.36 per 100 person-years; HR 0.56 (95% CI 0.46–0.69; p<0.00001). 44% relative reduction. ARR over 1.9 years ~1.2%; NNT ~95 over 1.9 years (extrapolated NNT 25 over 5 years if benefit persisted linearly). All-cause mortality HR 0.80 (95% CI 0.67–0.97; p = 0.02). LDL-C reduced 50% to ~55 mg/dL on rosuvastatin; hs-CRP reduced 37%.⁵⁵

Funding and conflicts. AstraZeneca funded the trial, designed it jointly with the academic investigators, and held the data. Nine of 14 authors of the primary publication declared financial relationships with AstraZeneca. Ridker held the hs-CRP patent at Brigham and Women's, and the trial's success commercially expanded the market for hs-CRP testing.⁵⁶

Steelmanned heterodox critique. JUPITER is the canonical "commercial trial" case study in the heterodox literature, and the criticisms have been laid out most fully by de Lorgeril et al. in their 2010 Archives of Internal Medicine "critical reappraisal" and by Kaul, Hlatky, and others in editorials in the same period.⁵⁶⁵⁷ The strongest version of the critique:

1. Early termination probably inflates the effect estimate. The trial was stopped at 1.9 years against a planned 5 years, at the first interim analysis, after the prespecified stopping boundary was exceeded. Bassler et al. (2010 JAMA) demonstrated that trials stopped early for benefit overestimate effect sizes by ~25–30%, and this is the central methodological objection.⁴⁵ Sanjay Kaul asked publicly: "Why was it stopped so early, especially when we have no idea about the long-term safety of very low LDL levels?"⁵⁸

2. Effect-size-to-ARR mismatch. A 44% relative risk reduction sounds dramatic, but the ARR over 1.9 years is ~1.2% — an NNT of ~95 over 1.9 years. The relative-risk-headline / small-absolute-benefit gap is real and characteristic of low-risk primary prevention.

3. Subgroup performance. The benefit was statistically significant in men but not statistically significant in women (HR 0.54; 95% CI 0.30–0.99 — borderline) and not significant in the elderly (≥70 years). The FDA's 2010 advisory committee that approved the JUPITER indication expansion for rosuvastatin had a 12–4 vote with notable dissent on the women and elderly subgroups.⁵⁹

4. CV mortality decomposition. The all-cause mortality benefit was modest (HR 0.80; ARR ~0.5%) and was driven partly by non-CV deaths in the placebo arm. CV death alone was not significantly reduced (HR 0.82; 95% CI 0.52–1.27).

5. Funder-author financial entanglement. AstraZeneca employees were involved in data analysis and reporting. Ridker's hs-CRP patent created a personal commercial interest in the trial's success — not a falsification of the result, but a real conflict that has not been adequately addressed in subsequent guidelines.

6. Diabetes adverse signal. New-onset diabetes occurred more often on rosuvastatin (3.0% vs 2.4%; HR 1.25; p = 0.01). This signal would be confirmed in subsequent meta-analyses (Sattar et al. 2010 Lancet, ~9% relative increase in new-onset diabetes on statins).⁶⁰

Steelmanned consensus defence. The orthodox response is that JUPITER's findings were directionally consistent with prior statin trials extended into lower-LDL territory, that the early stopping was performed under prespecified statistical rules and was ethically required given the prespecified efficacy boundary was crossed, that the women and elderly subgroup hazard ratios were directionally favourable even where not formally significant, and that the diabetes signal is real but is outweighed by CV benefit. Subsequent meta-analyses (CTT 2010 and onwards) have included JUPITER's data without finding that it distorted the overall conclusion. Hlatky's 2008 NEJM editorial — frequently cited by the heterodox literature as a critical commentary — actually concluded that the trial supported expanded statin use while flagging the unanswered cost-effectiveness and women-subgroup questions.⁵⁷

Assessment. JUPITER is the most contested statin trial in the literature, and reasonable scientists have read it in opposite ways for 17 years. The honest reading is that it is directionally informative but not on its own decisive: it shows that statin therapy reduces composite CV events in low-LDL, high-hs-CRP primary prevention; it does not, by itself, settle the question of which patients should be treated, particularly given the early-stopping inflation and the borderline women/elderly subgroup results. JUPITER is best understood as one data point among many, weighted by the CTT meta-analyses that follow — and that weighting is itself contested (chapter 8).

7.9 What chapter 7 establishes

By 2008, the cholesterol hypothesis had survived testing in: - Broad-risk secondary and high-risk primary prevention (HPS) — strong all-cause mortality benefit. - Hypertensive primary prevention (ASCOT-LLA) — early-stopping-inflated event benefit, no mortality benefit until 20-year legacy follow-up. - Elderly mixed primary/secondary (PROSPER) — event benefit, no mortality benefit, cancer signal. - High vs moderate intensity post-ACS (PROVE-IT), stable disease (TNT), prior-MI (IDEAL) — modest additional benefit from intensification, mortality signals null or non-significant. - Low-LDL, high-hs-CRP primary prevention (JUPITER) — disputed but directionally supportive of expansion.

The intellectual movement of this period is the consolidation of "lower is better" and "treat the risk, not the lipid" — but in tension. The next chapter examines the meta-analytic framework that tried to integrate all of this.

Chapter 8 — The CTT Collaboration: Consensus and Its Discontents

8.1 Origin of the CTT

The Cholesterol Treatment Trialists' Collaboration was established in 1994 at the Clinical Trial Service Unit (CTSU) at Oxford, under the leadership of Rory Collins, with the explicit aim of pooling individual-patient data (IPD) from all major randomised statin trials. The CTT's distinctive methodological move was to require participating trial groups to share patient-level data, not just summary results, and to perform analyses by intention-to-treat with prespecified subgroup methods.⁶¹

The first major CTT publication was the 2005 Lancet meta-analysis ("Efficacy and safety of cholesterol-lowering treatment: prospective meta-analysis of data from 90,056 participants in 14 randomised trials of statins"), led by Colin Baigent.⁶² The headline finding: per 1 mmol/L (39 mg/dL) reduction in LDL-C, the relative risk of major vascular events fell by 21% (rate ratio 0.79; 95% CI 0.77–0.81), with consistent benefit across baseline risk categories.

8.2 The 2010 update — "more intensive lowering"

The 2010 CTT Lancet paper expanded to 26 trials and 170,000 participants, and added the five intensive-vs-moderate statin trials (PROVE-IT, TNT, IDEAL, A-to-Z, SEARCH). Headline: per 1 mmol/L LDL-C reduction, 22% relative reduction in major vascular events (rate ratio 0.78; 95% CI 0.76–0.80; p<0.0001), with benefits accruing in the first year and continuing through subsequent years.⁶³ This is the "per-mmol/L" claim that anchors every modern guideline.

The 2010 paper added that intensive vs moderate regimens produced an additional 15% relative reduction in major vascular events per 0.5 mmol/L further LDL-C lowering. The log-linear relationship — i.e. each absolute reduction in LDL-C produces a proportional reduction in CV risk, regardless of starting point — became the central plank of modern lipidology.

8.3 The 2012 low-risk extension

In 2012, the CTT published a Lancet analysis specifically of low-risk primary prevention, using IPD from 27 trials with 175,000 participants stratified by predicted 5-year CV risk.⁶⁴ Headline: even in the lowest-risk stratum (predicted 5-year MVE risk <5%), statin therapy produced a per-mmol/L relative reduction in MVE comparable to that in higher-risk strata. Critically, the absolute benefit was smaller in lower-risk patients because the baseline risk was lower — about 6 fewer MVE per 1,000 person-years per mmol/L in the lowest-risk stratum vs ~16 fewer in the highest-risk stratum. The CTT concluded that statin therapy was beneficial across the risk spectrum but that absolute decisions should still be calibrated to baseline risk.

8.4 The 2015 sex meta-analysis

CTT 2015 (Fulcher et al., Lancet) addressed the sex-difference question raised by JUPITER and others by pooling IPD from 27 trials, 174,149 participants of whom 46,675 (27%) were women.⁶⁵ Headline: per 1 mmol/L LDL-C reduction, women had a 16% relative reduction in major vascular events vs 22% in men, but the test for sex interaction was non-significant (p = 0.33). The CTT concluded that statins benefit women and men similarly. Critics (Eichler 2015, Kostis 2012 prior meta-analysis) have argued that the underpowered women subgroup means the conclusion of equivalent benefit is over-interpreted.⁶⁶

8.5 The 2019 elderly meta-analysis

CTT 2019 (Lancet) pooled IPD from 28 trials, 186,854 participants, with a focus on age stratification.⁶⁷ Headline: per 1 mmol/L LDL-C reduction, major vascular events were reduced by 21% in patients ≤55, 20% in 56–65, 17% in 66–75, and 13% in >75 — directionally consistent but with smaller effect in the oldest stratum. In primary prevention specifically, the >75 subgroup had a non-significant 8% relative reduction. This is the meta-analytic answer that the STAREE and PREVENTABLE trials are designed to revisit.¹

8.6 The 2022 and 2024 updates

The 2022 CTT meta-analysis incorporated the PCSK9 trials (FOURIER, ODYSSEY OUTCOMES — covered in Part II-B chapter 9) and the IMPROVE-IT ezetimibe trial, broadening the analysis to non-statin LDL-lowering and confirming the per-mmol/L relationship across drug classes.⁶⁸ The 2024 update added bempedoic acid (CLEAR Outcomes) and provided the harmonised effect-size estimate the 2025 ESC/EAS guidelines use.¹

8.7 Steelmanning the consensus case for CTT

The orthodox claim, put strongly, is that the CTT methodology is the gold standard for cardiovascular evidence synthesis:

• IPD pooling, not summary statistics, gives the CTT access to within-patient-level interactions (baseline risk, age, sex, prior CV disease) that summary-level meta-analyses cannot resolve. This is genuinely more powerful than the typical Cochrane-style summary meta-analysis.
• Prespecified analytic plans reduce the risk of post-hoc subgroup mining. The CTT publishes analysis plans before pooling.
• Consistency across decades and trials. The per-mmol/L effect estimate (~22% RR reduction per 1 mmol/L LDL-C for MVE) has been stable across the 2005, 2010, 2012, 2015, 2019 and 2022 updates as new trials were added. This is the strongest internal validation: independent new trials slot into the same effect estimate.
• Independent confirmation by Mendelian randomisation. The MR literature (Ference et al. 2017, EAS consensus¹) finds that genetic variants reducing LDL-C produce a per-mmol/L relative reduction in CHD that is quantitatively similar to the CTT estimate from RCTs (per 1 mmol/L genetically lower LDL-C: ~50% relative reduction in CHD over a lifetime; per 1 mmol/L pharmacologically lower LDL-C over 5 years: ~22% — and the difference is explained by duration of exposure). When two completely independent methodologies converge on the same effect estimate, the underlying claim is unusually well-supported.
• Specification of effect on hard outcomes. CTT consistently reports CV mortality, all-cause mortality, and major vascular events separately, and the per-mmol/L effect on hard endpoints (CV death, MI, stroke) is well-replicated.

On this orthodox reading, the CTT meta-analyses are the closest thing cardiovascular medicine has to a foundational quantitative law: LDL-C lowering causes proportional CV-risk reduction, log-linearly, across drug classes, sexes, ages, and baseline risk strata, with effect-size confirmed by parallel genetic evidence. The clinical implication is that LDL-C should be lowered as much as is feasible in patients with non-trivial CV risk.

8.8 Steelmanning the critique of CTT

The heterodox critique of CTT is not that the meta-analyses are wrong in their topline number, but that the process is closed and unreproducible — and that this matters for scientific epistemology even if the answer is right. The strongest version of the critique, drawing on Diamond and Ravnskov, DuBroff and de Lorgeril, and the 2016 PLOS Medicine exchange between the CTT and its critics, runs:

1. The IPD is not openly available outside CTT. Independent researchers cannot reanalyse the data. The CTT operates under data-sharing agreements with the original trial sponsors (mostly pharmaceutical companies), and re-release would require sponsor consent. As a consequence, only the CTT itself can produce CTT-based estimates. This is a single-source data monopoly on the most important meta-analytic synthesis in cardiology.⁶⁹

2. The personnel are stable across decades. Rory Collins, Colin Baigent, Jane Armitage, and Christina Reith — the core CTT leadership — have been in place since the 1990s. The same group runs HPS, the 2010 CTT, the 2012 CTT, the 2019 CTT, and so on. There is no rotation of analytic leadership. Heterodox critics (notably John Abramson, Aseem Malhotra) have argued this is a form of scientific monopoly; the CTSU/Oxford response is that consistent leadership produces consistent methods.⁷⁰

3. Diamond and Ravnskov (2015, BMJ EBM) published a re-analysis of the publicly available trial-level summary data, arguing that the absolute risk reductions in primary prevention are smaller than the CTT relative-risk framing implies, and that the all-cause mortality benefit in low-risk primary prevention is questionable.⁷¹ The CTT response (Collins et al. 2016, Lancet) defended the IPD methodology and contested the Diamond/Ravnskov calculations, but did not release the IPD that would have allowed independent adjudication.⁷²

4. DuBroff and de Lorgeril (2015) went further, arguing that the cholesterol hypothesis itself is over-extended and that the CTT framing obscures population-level evidence (e.g. that LDL-C–CHD associations weaken or reverse in some populations).⁷³

5. The 2016 PLOS Medicine exchange. Following Collins et al.'s 2016 Lancet paper, a series of letters in PLOS Medicine and BMJ asked the CTT to release IPD for independent scrutiny. The CTT's position was, in summary: (a) data-sharing agreements with sponsors prevent it; (b) the CTT's internal processes are rigorous; (c) the consistency of results across updates is itself evidence of robustness. Critics countered: (a) modern data-sharing norms (ICMJE, NIH, BMJ) require IPD release in principle, and CTT trials are old enough that commercial confidentiality is no longer a defensible objection; (b) "trust us" is not a satisfactory epistemic answer in 21st-century science.⁷⁴

6. The "log-linear" claim is a model assumption, not an empirical finding. The CTT's log-linear LDL-CHD relationship is fit to the trial data with a regression assumption; alternative model specifications (e.g. with a threshold below which further LDL-C lowering produces no further benefit) cannot be ruled out from the published summary data. The PCSK9 trials (FOURIER, ODYSSEY) extend the LDL-C range to ~30 mg/dL and continue to show CV benefit, which strengthens the log-linear case empirically — but the model assumption remains an assumption.

8.9 Adjudication

Which side wins? The honest answer, I think, is that both are largely right, but on different questions.

On the substantive scientific question — does LDL-C lowering reduce CV events, and approximately how much per mmol/L? — the orthodox CTT answer is well-supported, internally consistent, independently corroborated by Mendelian randomisation, and survived the addition of every major statin and non-statin trial added since 2005. The headline "per 1 mmol/L LDL-C reduction → ~22% RR reduction in major vascular events" is approximately true and is the single most important quantitative claim in cardiovascular preventive medicine.

On the epistemic / process question — is the science of cholesterol meta-analysis transparent and reproducible? — the heterodox critique is correct. The CTT IPD has not been released; independent reanalysis is not possible; the personnel are stable for three decades. This is a real problem for 21st-century scientific norms, and the CTSU's 2016 response was, in my reading, defensive rather than responsive. The fact that the substantive answer is probably right does not excuse the closed process.

What the substantive correctness does establish is that the heterodox argument cannot rest on "the data have been hidden, therefore the answer might be wrong" — independent lines of evidence (4S all-cause mortality, MR genetic studies, FOURIER-OLE legacy effect, CLEAR Outcomes confirmation in statin-intolerant patients) converge on the same direction and approximate magnitude. The answer is probably right; the process should still be opened.

For a marketing-director-debating-a-Roche-friend, this is the operationally important distinction: the science of LDL-C lowering is broadly settled at the population level; the politics and epistemology of how it was established are not. Both can be true at once.

8.10 What chapter 8 establishes — and what Part II-B will pick up

By 2010, the CTT had produced what is still the canonical meta-analytic estimate of the LDL-C–CV risk relationship. The four-trial pillars on which this rested (4S, WOSCOPS, CARE/LIPID, HPS) are themselves well-established, and the CTT log-linear relationship is consistent with Mendelian randomisation.

What chapter 8 does not settle — and what Part II-B will take up:

• PCSK9 inhibitor outcomes (FOURIER 2017, ODYSSEY OUTCOMES 2018, FOURIER-OLE 2022). LDL-C reduction to ~30 mg/dL: does CV risk continue to fall log-linearly?
• Ezetimibe (IMPROVE-IT 2015): does a non-statin LDL-C-lowering drug produce CTT-consistent CV-risk reduction?
• Bempedoic acid (CLEAR Outcomes 2023, CV mortality and 2025 VTE updates from the grounding file¹).
• Elderly primary prevention (STAREE, PREVENTABLE — both reading out 2026¹).
• Harms (myopathy, diabetes incidence, cognitive concerns, nocebo).
• The newer Lp(a) and remnant-cholesterol literature that extends beyond LDL-C as the sole target.

These are the chapters that turn the 2010 consensus into the 2026 frontier.

Master Trial Table — Historical Era

*Trial stopped early; planned follow-up was longer.

Footnotes

Part II-B: The Modern Trial Era (2010–2026)

"If something can't go on forever, it won't." — Herb Stein, applied here to the post-statin LDL-lowering economy.

Part II-A closed in 2008–2010 with the statin question largely settled by JUPITER, the Cholesterol Treatment Trialists' (CTT) meta-analyses, and the slow asymptote of incremental benefit at ever-lower LDL-C. What follows is the era in which the field tested a different question. Not do statins work? — that ship had sailed — but is it the LDL-C itself, or something else statins do, that delivers the benefit? And then, downstream of that answer: how low is too low, with what agent, in whom, at what cost?

This chapter walks through six waves of trial evidence between 2008 and 2026: (i) the ezetimibe arc culminating in IMPROVE-IT (Chapter 9); (ii) the PCSK9 monoclonal antibody era (Chapter 10); (iii) the siRNA/inclisiran chapter still being written (Chapter 11); (iv) bempedoic acid's claim on the statin-intolerant niche (Chapter 12); (v) the Lp(a) frontier (Chapter 13); and (vi) the pending 2025–2026 readouts that will close — or reopen — the elderly primary-prevention question (Chapter 14). Each trial is examined for design, n, follow-up, primary endpoint, hazard ratio, absolute risk reduction, number needed to treat, funding, and the steelmanned sceptic reading.

A running master trial table appears at the end of Chapter 14. Throughout, "MACE" denotes the trial's specific composite — definitions vary materially and the variation matters.

Chapter 9 — Non-statin LDL lowering opens the door: the ezetimibe arc

9.1 Why ezetimibe was the right experiment

By 2005 the LDL-C/CHD relationship was clinically operational. Statins lowered LDL-C and lowered events; the dose-response was monotonic across the CTT meta-analyses¹. But statins are not pure LDL-lowering agents. They suppress hepatic HMG-CoA reductase and, downstream, the entire mevalonate pathway. They reduce isoprenoid intermediates (farnesyl pyrophosphate, geranylgeranyl pyrophosphate) that prenylate small GTPases involved in inflammation, endothelial function, and platelet activity — the so-called "pleiotropic effects"². Statins are also potent anti-inflammatories; JUPITER (2008) had just shown that statin benefit tracked with hsCRP reduction as well as LDL-C reduction in primary prevention³. Whether the cardiovascular benefit was due to LDL-C lowering, or merely correlated with it, was therefore unsettled. The mechanism question was not academic — it determined whether a non-statin LDL-lowering agent should be expected to lower events at all.

Ezetimibe (SCH 58235; trade name Zetia, and in fixed combination with simvastatin, Vytorin) was the natural test case. Ezetimibe inhibits the Niemann–Pick C1-Like 1 (NPC1L1) transporter at the intestinal brush border, reducing cholesterol absorption by ~54%⁴. Hepatic LDL-receptor upregulation follows. Plasma LDL-C falls ~15–22% on monotherapy; added to a statin, the additional reduction is ~18–25%⁴. Crucially, ezetimibe has no known meaningful pleiotropic effects. If LDL-C is the operative cause of atherosclerosis, ezetimibe added to a statin should reduce events; if LDL-C is a bystander and the benefit lives in the mevalonate pathway, it should not.

This is the Popperian shape of the experiment: a clean, falsifiable test of the LDL hypothesis using a non-statin tool.

9.2 ENHANCE (2008): the trial that nearly killed ezetimibe

The Ezetimibe and Simvastatin in Hypercholesterolemia Enhances Atherosclerosis Regression (ENHANCE) trial randomised 720 patients with heterozygous familial hypercholesterolaemia (HeFH) to simvastatin 80 mg plus ezetimibe 10 mg versus simvastatin 80 mg plus placebo, with the primary endpoint being mean change in carotid intima-media thickness (cIMT) at 24 months⁵. The trial completed in April 2006. Topline results were not announced until January 2008 and the full paper, by Kastelein and the ENHANCE investigators, appeared in NEJM in April 2008⁵.

LDL-C fell further with combination therapy (mean reduction 58% vs 41%, P<0.01). cIMT did not differ: mean change 0.0111 mm with combination versus 0.0058 mm with simvastatin alone (P=0.29) — combination therapy if anything trended towards more intimal thickening, though this was almost certainly noise⁵.

The publication delay became the scandal. The trial had a co-sponsorship between Merck and Schering-Plough through MSP Singapore (the entity that marketed Vytorin), and the publication was repeatedly pushed back. The original primary endpoint was changed during the delay, and a US Congressional subcommittee investigation followed in March 2008⁶. Sales of Vytorin and Zetia, which had collectively reached approximately US$5 billion annually by 2007, dropped by roughly 25% in the year after ENHANCE was released⁷. Two former US Senators (Charles Grassley and John Dingell) led inquiries; Merck and Schering-Plough later paid US$688 million in 2013 to settle related shareholder litigation alleging the companies misled investors about the ENHANCE timeline⁸.

Steelman the sceptic on ENHANCE. Three points deserve weight:

1. cIMT is a surrogate, and a noisy one. The HeFH population in ENHANCE had been treated with high-intensity statins for years prior to enrolment, meaning baseline cIMT was already low and the room for further regression was compressed. A null result on a saturated population may say nothing about the LDL hypothesis writ large.
2. Simvastatin 80 mg is itself an extreme intervention; the comparator was not placebo but maximal-intensity statin monotherapy. Detecting a marginal effect of an additional 17 percentage points of LDL reduction at the bottom of the dose-response curve was always going to require a larger trial.
3. The episode does establish a genuine scandal of sponsor behaviour — not of the science. Industry-managed trials with surrogate endpoints, late endpoint changes, and embargoed disappointing data are exactly what should be flagged. Merck/Schering-Plough's conduct here is the textbook example of why the field demands pre-registration, blinded adjudication, and trial-level transparency.

ENHANCE killed Vytorin's growth trajectory but did not, in fact, falsify the LDL hypothesis. It tested a surrogate in a saturated population. The question stayed open.

9.3 SEAS (2008): the cancer signal that didn't replicate

The Simvastatin and Ezetimibe in Aortic Stenosis (SEAS) trial enrolled 1,873 patients with mild-to-moderate asymptomatic aortic stenosis, randomised to simvastatin 40 mg + ezetimibe 10 mg versus placebo, with a primary composite of major cardiovascular events including aortic-valve events. Median follow-up was 52.2 months. The primary endpoint was not reduced (HR 0.96, 95% CI 0.83–1.12, P=0.59); aortic-valve events specifically were not affected (HR 0.97, 95% CI 0.83–1.14)⁹.

What made SEAS notorious was an unexpected cancer signal. Incident cancer was diagnosed in 105 patients (11.1%) on combination therapy versus 70 (7.5%) on placebo (HR 1.55, P=0.01); cancer mortality was 39 (4.1%) versus 23 (2.5%), HR 1.67⁹. Sir Richard Peto and colleagues were asked to perform a rapid emergency meta-analysis pooling SEAS with the (then-still-running) IMPROVE-IT and SHARP trials, totalling ~20,000 patients on ezetimibe — they found no overall excess cancer (RR 0.96, 95% CI 0.84–1.09)¹⁰. A 2014 long-term registry-based follow-up of SEAS itself found no persistent cancer signal¹¹.

Steelman the sceptic on SEAS. A real positive signal cannot be dismissed by pooled null data alone; one could argue that the larger trials were too short for cancer detection. But the signal failed to replicate in IMPROVE-IT (the largest ezetimibe outcomes trial, 18,144 patients, 6 years), in SHARP, in the FDA's post-marketing surveillance, and in long-term Scandinavian and Korean registry data¹¹. The most parsimonious reading: SEAS surfaced a multiple-comparisons false positive, of the kind one expects when running dozens of safety endpoints on under 2,000 patients. This was also the conclusion the SEAS investigators themselves reached publicly¹².

9.4 SHARP (2011): the first non-statin LDL-lowering outcomes win

The Study of Heart and Renal Protection (SHARP) is sometimes overlooked because its population was niche — but it was the first randomised trial to show that adding ezetimibe to a statin lowered cardiovascular events. SHARP randomised 9,270 patients with chronic kidney disease (3,023 on dialysis, 6,247 not) to simvastatin 20 mg + ezetimibe 10 mg versus placebo, with a primary endpoint of first major atherosclerotic event (non-fatal MI, coronary death, non-haemorrhagic stroke, or any arterial revascularisation). Median follow-up was 4.9 years¹³.

LDL-C fell by a mean 0.85 mmol/L (33 mg/dL) in the combination arm. The primary endpoint occurred in 11.3% of the combination arm versus 13.4% of placebo — HR 0.83, 95% CI 0.74–0.94, P=0.0021. Absolute risk reduction 2.1 percentage points; NNT over 4.9 years approximately 47¹³. Major vascular events fell from 17.6% to 15.1% (P=0.0012). Importantly, end-stage renal disease was not reduced (HR 0.97, 95% CI 0.89–1.05, P=0.41) — so the benefit was atherosclerotic, not renal¹³.

SHARP was funded by Merck/Schering-Plough but managed academically by the Clinical Trial Service Unit (CTSU) at Oxford under Sir Rory Collins and Colin Baigent — the same group that runs the CTT meta-analyses. The trial's design featured a 1-year run-in to weed out non-adherent and intolerant patients before randomisation, which is fair (it improves statistical power) but does mean the analysed population was preselected for tolerance.

Steelman the sceptic on SHARP. Three threads:

1. SHARP did not have a statin-monotherapy arm. The comparison was simvastatin+ezetimibe versus placebo — so SHARP demonstrates that combined lipid-lowering beats nothing in CKD, not that ezetimibe adds independent value to a statin in CKD. The question "how much did the ezetimibe contribute?" is mathematically inferred (from LDL-C trajectory) rather than directly tested.
2. The CKD population is at very high baseline risk with a peculiar lipoprotein profile (higher Lp(a), often elevated remnants, altered HDL). Generalising the SHARP effect size to lower-risk patients is non-trivial.
3. The CTSU was both the analytic group and shares principal investigators with the CTT collaboration that conducts meta-analyses of the same drug class — a closed loop that critics (Diamond, Ravnskov, Kendrick) have pointed to as an absence of true independent verification. This is a governance critique not a data critique; the SHARP results have stood up to external re-analysis. But the point lands.

SHARP was the first crack in the door. IMPROVE-IT would walk through it.

9.5 IMPROVE-IT (2015): the trial that confirmed LDL itself

The Improved Reduction of Outcomes: Vytorin Efficacy International Trial (IMPROVE-IT) is the single most important non-statin lipid trial of the 2010s, because it was designed to settle the mechanism question. Cannon, Blazing, Giugliano and colleagues randomised 18,144 patients within 10 days of an acute coronary syndrome to simvastatin 40 mg + ezetimibe 10 mg versus simvastatin 40 mg + placebo. The primary endpoint was a composite of cardiovascular death, MI, unstable angina requiring rehospitalisation, coronary revascularisation ≥30 days after randomisation, or stroke. Median follow-up was 6 years¹⁴.

LDL-C at 1 year averaged 53.7 mg/dL (1.4 mmol/L) on combination therapy versus 69.5 mg/dL (1.8 mmol/L) on simvastatin alone — a between-arm difference of 15.8 mg/dL (0.4 mmol/L). The 7-year Kaplan–Meier event rate for the primary endpoint was 32.7% with combination versus 34.7% with monotherapy. Hazard ratio 0.936 (95% CI 0.89–0.99, P=0.016). Absolute risk reduction 2.0 percentage points over 7 years. NNT ≈ 50¹⁴.

The component-level results are worth noting. MI was reduced (HR 0.87, P=0.002), ischaemic stroke was reduced (HR 0.79, P=0.008), but all-cause mortality (HR 0.99, P=0.78) and cardiovascular mortality (HR 1.00) were not¹⁴. The benefit was atherothrombotic — MI and ischaemic stroke — exactly as the LDL hypothesis predicts.

The trial's scientific significance is disproportionate to its effect size. For a 0.4 mmol/L LDL-C reduction sustained over 6 years, the CTT meta-regression predicts roughly an 8% relative reduction in major vascular events per mmol/L per year of follow-up¹. IMPROVE-IT delivered a 6.4% relative reduction — within the CTT confidence band. The LDL hypothesis predicted IMPROVE-IT's result quantitatively, using parameters derived from statin trials, before IMPROVE-IT read out. This is the kind of out-of-sample prediction that distinguishes a causal hypothesis from a correlational one.

Steelman the sceptic on IMPROVE-IT. Several legitimate counter-arguments need facing head-on:

1. The effect is small. ARR of 2 percentage points over 7 years to prevent one composite event. If you exclude soft endpoints (revascularisation, unstable angina), the effect shrinks further. NNT for hard MACE alone is plausibly 80–100 over 7 years. For an asymptomatic primary-prevention patient, this would be unimpressive. (But IMPROVE-IT's population was post-ACS — secondary prevention, high baseline risk. Translating the effect to primary prevention is a category error in either direction.)
2. It took 6 years to surface. Kaplan–Meier curves separated only after ~2 years. Trials shorter than that will systematically fail to detect ezetimibe's signal. This matters for how PCSK9 trials (2-year FOURIER, 2.8-year ODYSSEY) and inclisiran trials (ORION-4 will be ~5 years) are interpreted.
3. The P-value was 0.016, with prespecified α tightened for multiplicity (the trial had multiple interim analyses). Some statisticians, including John Ioannidis, have argued IMPROVE-IT is more honestly read as "weakly positive" or "borderline"¹⁵. The replication question — would a second IMPROVE-IT in a different ACS population show the same effect? — has never been directly answered.
4. No mortality benefit. All-cause and CV mortality were both null. A sceptic can argue that an intervention that fails to extend life, in a population at sufficient baseline risk to have a ~35% 7-year composite event rate, may not be capturing what matters.

The steelman of the steelman — what advocates would say in response — is that IMPROVE-IT was the first test of "LDL lowering by any mechanism" and it worked, narrowly but in the predicted direction. The mechanism question shifted decisively. Henceforth, the working hypothesis became: it's the LDL particle (more precisely, apoB-containing particle concentration), not statin-specific biology, that drives the atherosclerotic effect. This is the foundation on which the PCSK9 and inclisiran programmes were built.

Chapter 10 — The PCSK9 monoclonal antibody era

10.1 Discovery: from FH gain-of-function to LDL-receptor recycling

The PCSK9 (proprotein convertase subtilisin/kexin type 9) story is the cleanest example in modern cardiovascular medicine of human genetics directing drug discovery. In 2003, Marianne Abifadel, Catherine Boileau and colleagues at INSERM identified gain-of-function mutations in PCSK9 as a third cause of autosomal-dominant familial hypercholesterolaemia, alongside LDLR and APOB¹⁶. Patients with these mutations had markedly elevated LDL-C and premature CHD. The mechanism, worked out over 2004–2007, was that PCSK9 binds the LDL receptor at the hepatocyte surface and targets it for lysosomal degradation. More PCSK9 activity → fewer LDL receptors recycled to the surface → higher circulating LDL-C¹⁷.

The pivotal moment came in 2006 with Jonathan Cohen and Helen Hobbs at UT Southwestern. Working in the Atherosclerosis Risk in Communities (ARIC) cohort, they identified loss-of-function PCSK9 variants in 2.6% of Black participants (Y142X and C679X nonsense mutations) and a different variant (R46L) in 3.2% of White participants. Carriers had life-long modestly reduced LDL-C — 28% lower in Black carriers, 15% lower in White carriers — and dramatically reduced CHD risk: an 88% reduction in 15-year CHD incidence in Black nonsense-mutation carriers, and a 47% reduction in White R46L carriers¹⁸.

This is the Mendelian randomisation moment for PCSK9. Random allocation at conception (Mendel's First Law); lifelong exposure to lower LDL-C; massive reduction in CHD. The effect size per mmol/L lifetime LDL-C reduction was roughly three times what statin trials had delivered — consistent with the cumulative LDL-C × time hypothesis: lifelong modest exposure trumps late, larger pharmacological exposure¹⁹. The Cohen/Hobbs paper is, in a real sense, the founding text of the PCSK9 inhibitor era. Inhibit PCSK9 pharmacologically, the prediction went, and you should get a phenocopy of the loss-of-function carrier — lower LDL-C, lower events, no significant safety penalty.

Two pharmacological strategies emerged. Monoclonal antibodies targeting circulating PCSK9 (Amgen's evolocumab, Sanofi/Regeneron's alirocumab, Pfizer's bococizumab) entered phase 3 in 2013–2014. Small interfering RNA against hepatic PCSK9 mRNA (Alnylam/The Medicines Company's inclisiran) followed. The antibodies went first.

10.2 FOURIER (2017): the first PCSK9 outcomes trial

Further Cardiovascular Outcomes Research with PCSK9 Inhibition in Subjects with Elevated Risk (FOURIER) randomised 27,564 patients with established atherosclerotic cardiovascular disease (prior MI, prior stroke, or symptomatic peripheral arterial disease) and LDL-C ≥70 mg/dL on optimised statin therapy, to evolocumab 140 mg every 2 weeks or 420 mg monthly versus placebo. Median follow-up was 2.2 years. Primary endpoint: composite of CV death, MI, stroke, hospitalisation for unstable angina, or coronary revascularisation. Key secondary: CV death, MI, or stroke²⁰.

Baseline LDL-C was 92 mg/dL. At 48 weeks, evolocumab reduced LDL-C by 59% versus placebo, to a median of 30 mg/dL — a level previously not achievable. The primary endpoint occurred in 9.8% of evolocumab patients versus 11.3% of placebo patients: HR 0.85 (95% CI 0.79–0.92, P<0.001), a 15% relative risk reduction. ARR 1.5 percentage points over 2.2 years; NNT ≈ 67²⁰. The key secondary endpoint (CV death/MI/stroke) was reduced more steeply: 5.9% vs 7.4%, HR 0.80 (95% CI 0.73–0.88, P<0.001), a 20% relative risk reduction²⁰.

All-cause mortality was not significantly different: 3.2% vs 3.1%, HR 1.04 (95% CI 0.91–1.19). Cardiovascular mortality was also null: HR 1.05 (95% CI 0.88–1.25)²⁰. The benefit was entirely on MI and stroke, not on death.

Critical secondary findings included a lack of evidence for cognitive harm (the EBBINGHAUS substudy²¹), no new-onset diabetes signal (importantly, distinct from the statin signal), and a clean overall safety profile aside from injection-site reactions. Adherence was >90% throughout, helped by the trial paying for and administering the drug.

A pre-specified secondary analysis by Giugliano et al. examined the relationship between achieved LDL-C and outcomes. Patients in the lowest LDL-C category (<20 mg/dL, ~0.5 mmol/L) had no apparent adverse events relative to those at 20–50 mg/dL — extending the "lower is better, with no obvious floor" finding from statin trials into uncharted LDL-C territory²².

10.3 FOURIER-OLE: the legacy effect

FOURIER's most consequential output may have been its open-label extension (OLE). Patients who had been on evolocumab in FOURIER, and those originally on placebo, were offered open-label evolocumab in the extension; 6,635 patients continued, with median 5.0 years of OLE follow-up on top of the original 2.2 years (total 7.1 years on evolocumab for the early-treatment arm)²³.

The OLE showed two important things. First, sustained safety: no cognitive decline through 7.2 years; discontinuation rate attributable to evolocumab was 0.1%²³. Second, and more striking, a legacy effect. Patients who had been on evolocumab from the start of FOURIER had a 15% lower rate of MACE-5 (HR 0.85) and a 20% lower rate of CV death/MI/stroke (HR 0.80) versus those who had been on placebo for the first 2.2 years before crossing over²³. Both arms were now on identical therapy, yet the early-treatment arm retained its advantage. This is precisely what the cumulative LDL-C × time hypothesis predicts: the longer the LDL-lowering exposure, the greater the benefit, with the curves continuing to separate rather than converge once the placebo arm catches up.

The legacy-effect finding is methodologically important. It addresses a long-standing critique of FOURIER's modest absolute effect: the trial may have been too short. If you genetically halve LDL-C from birth (Cohen/Hobbs) you reduce CHD by ~88%; if you pharmacologically halve it for 2 years in a 65-year-old, you reduce events by ~15%. The dose-time integral is the missing variable. FOURIER-OLE provides the first within-trial evidence that the benefit accumulates.

10.4 ODYSSEY OUTCOMES (2018): alirocumab in post-ACS

ODYSSEY OUTCOMES randomised 18,924 patients 1–12 months after an acute coronary syndrome with LDL-C ≥70 mg/dL (or non-HDL-C ≥100 mg/dL or apoB ≥80 mg/dL) on high-intensity statin, to alirocumab 75 mg every 2 weeks (titrated to target LDL-C 25–50 mg/dL) versus placebo. Median follow-up 2.8 years. Primary composite: CHD death, non-fatal MI, fatal/non-fatal ischaemic stroke, or unstable angina requiring hospitalisation²⁴.

Baseline LDL-C was 87 mg/dL; on alirocumab the median achieved LDL-C was 38 mg/dL at month 4 and 53 mg/dL at month 48 (titration downward was permitted for very low LDL-C levels). Primary endpoint: 9.5% on alirocumab versus 11.1% on placebo, HR 0.85 (95% CI 0.78–0.93, P<0.001). ARR 1.6 percentage points; NNT ≈ 63²⁴.

Where ODYSSEY differed from FOURIER was the mortality signal. All-cause death occurred in 3.5% of alirocumab patients versus 4.1% of placebo: HR 0.85 (95% CI 0.73–0.98)²⁴. The trial's hierarchical testing procedure formally classified this as nominally significant but not part of the confirmatory pre-specified hierarchy (because higher-ranked secondary endpoints did not all hit), so it is conventionally reported as "consistent with mortality benefit" rather than a definitive mortality reduction. Cardiovascular death itself: HR 0.88 (95% CI 0.74–1.05) — directionally favourable, not statistically significant on its own²⁴.

The pre-specified subgroup analysis by baseline LDL-C showed greater absolute and relative benefit in those entering with LDL-C ≥100 mg/dL (despite already being on high-intensity statin), consistent with the LDL-mass-of-effect hypothesis: the further you push it, the more events you prevent, but only if there was room to push from²⁴.

10.5 GLAGOV (2016): imaging the regression

The Global Assessment of Plaque Regression with a PCSK9 Antibody as Measured by Intravascular Ultrasound (GLAGOV) trial provided the mechanistic imaging counterpart to FOURIER. Nissen and colleagues randomised 968 patients undergoing clinically indicated coronary angiography to evolocumab 420 mg monthly or placebo, added to statin therapy, for 76 weeks. Primary endpoint: change in percent atheroma volume (PAV) on serial IVUS²⁵.

LDL-C at 76 weeks: 36.6 mg/dL on evolocumab vs 93.0 mg/dL on placebo. PAV changed by −0.95% on evolocumab vs +0.05% on placebo (P<0.001 for difference). Plaque regression (any decrease in PAV) occurred in 64.3% of evolocumab patients versus 47.3% of placebo patients (P<0.001)²⁵. Notably, regression was observed even in the subgroup with baseline LDL-C <70 mg/dL — i.e., patients already at guideline LDL-C goals showed further plaque regression with PCSK9 inhibition. There appeared to be no LDL-C floor below which further regression stopped.

10.6 SPIRE-1 and SPIRE-2 (2017): the failed antibody

Bococizumab was Pfizer's PCSK9 monoclonal antibody. Unlike fully human evolocumab and alirocumab, bococizumab was a humanised antibody (i.e., a murine antibody re-engineered with human framework regions). Phase 2 had shown excellent LDL-C reduction. SPIRE-1 (16,817 patients with established CVD, LDL-C ≥70 mg/dL) and SPIRE-2 (10,621 patients, LDL-C ≥100 mg/dL) were the planned outcomes trials²⁶.

In November 2016, Pfizer terminated the bococizumab programme after observing that ~48% of patients developed anti-drug antibodies, attenuating LDL-C reduction over time, and that injection-site reactions occurred in ~10% of patients. Truncated outcome data from the two SPIRE trials were nonetheless published. SPIRE-1 (~7 months median follow-up) showed no significant effect (HR 0.99); SPIRE-2 (~12 months) showed an HR of 0.79 (95% CI 0.65–0.97), which Sabatine and colleagues interpreted as consistent with the FOURIER/ODYSSEY effect size in the longer-exposure trial despite the immunogenicity problem²⁶.

The bococizumab failure was a cautionary tale about the murine vs human framework distinction in antibody engineering — fully human antibodies (evolocumab, alirocumab) do not generate this immunogenicity. It is also why bococizumab does not appear in modern lipid guidelines despite reaching phase 3.

10.7 Steelmanning the PCSK9 sceptic

The PCSK9 inhibitor case has been the most scrutinised in recent cardiovascular therapeutics. The sceptic reading deserves a full hearing:

1. The mortality question. FOURIER showed no mortality benefit (HR 1.04 for all-cause death). ODYSSEY showed a nominal but not formally confirmatory mortality signal (HR 0.85, CI crossing 1.00 for CV death). For a drug priced at US$14,000+/year (initial 2015 list price; reduced to ~$5,850 in 2018 after payer pushback)²⁷, the absence of a confirmed mortality benefit is a real critique. Statins, for comparison, do have evidence of all-cause mortality reduction in secondary prevention¹.

2. The NNT vs cost calculation. FOURIER's NNT of ~67 over 2.2 years to prevent one MACE event, at an annualised drug cost of ~$5,850, translates to roughly $390,000 per event prevented over the trial duration. ICER (Institute for Clinical and Economic Review) initially priced evolocumab cost-effective only at about $2,000/year²⁸. The price has since come down, but the underlying critique — that the absolute effect is modest enough that cost matters disproportionately — stands.

3. Trial duration may have undersold benefit. This is the steelman of the advocate side: FOURIER's 2.2-year median follow-up was almost certainly too short. FOURIER-OLE's legacy effect demonstrates the cumulative-exposure hypothesis. But this argument cuts both ways — if the original trial undersold benefit, it may also have undersold harms (e.g., low-LDL-C concerns that take >2 years to surface).

4. Real-world adherence is the hidden killer. Trials achieved >90% adherence with subsidised drug and trial-staff support. Real-world adherence to twice-monthly self-injection over 5+ years is materially worse. A 2019 US retrospective claims analysis showed PCSK9 inhibitor 12-month persistence around 60–70%²⁹. The benefit demonstrated in trial is the efficacy; the realised benefit in clinic is the effectiveness, and the gap is not small.

5. Patient-selection externalities. PCSK9 inhibitors as currently indicated (FH; ASCVD with LDL-C above goal despite maximum statin) are very high-risk patients. Pushing the indication to lower-risk groups, as some guideline updates contemplate, dramatically deteriorates the NNT and the cost-effectiveness ratio, without commensurate evidence in those populations.

The synthesis: PCSK9 inhibitors work, deliver clean LDL-C reduction with a clean safety profile and modest absolute benefit consistent with what the LDL hypothesis quantitatively predicts. The mortality data are ambiguous; the cost-effectiveness has improved but remains contested; and the FOURIER-OLE legacy data have reframed the field's view of treatment duration. They are unambiguously useful for FH and very-high-risk secondary prevention; their role in lower-risk patients depends on price and on outcomes from VESALIUS-CV and similar trials still pending (Chapter 14).

Chapter 11 — Inclisiran and the siRNA era

11.1 Mechanism and the adherence proposition

Inclisiran is a small interfering RNA (siRNA) directed against hepatic PCSK9 messenger RNA. The molecule is a chemically modified double-stranded RNA conjugated to a triantennary N-acetylgalactosamine (GalNAc) ligand. The GalNAc moiety binds asialoglycoprotein receptors on hepatocytes with high affinity, delivering the siRNA selectively to liver. Once internalised, the siRNA is loaded onto the RNA-induced silencing complex (RISC), which cleaves PCSK9 mRNA catalytically. Because RISC-loaded siRNA is durable and catalytic (one siRNA molecule destroys many mRNA copies), the duration of effect from a single subcutaneous injection is approximately 6 months³⁰.

The clinical proposition is therefore dose frequency. Statins are daily. PCSK9 monoclonal antibodies are biweekly to monthly. Inclisiran is twice yearly — typically administered at the GP surgery, removing self-injection burden entirely. If adherence is the chokepoint of long-term lipid management, inclisiran is a structural rather than incremental fix.

Inclisiran was developed by Alnylam Pharmaceuticals (which pioneered GalNAc-siRNA chemistry) in partnership with The Medicines Company. Novartis acquired The Medicines Company in January 2020 for US$9.7 billion³¹, placing inclisiran in Novartis's hands — a lipid-naive company that also acquired sotagliflozin and other cardiovascular assets in the same period.

11.2 ORION-9, -10, -11: LDL-C trials

The ORION pivotal phase 3 trials all used the same design: 300 mg subcutaneous inclisiran on day 1, day 90, and every 6 months thereafter, versus placebo, on top of background statin therapy. Primary endpoint: change in LDL-C from baseline to day 510 (~17 months). Outcomes were not the primary endpoint of these trials — they were powered for LDL-C reduction.

ORION-9: 482 patients with HeFH, LDL-C ≥2.6 mmol/L (100 mg/dL) on maximally tolerated statin ± ezetimibe. LDL-C reduction at day 510: −47.9% inclisiran vs placebo (P<0.001)³².

ORION-10: 1,561 US patients with ASCVD and LDL-C ≥1.8 mmol/L (70 mg/dL) on maximally tolerated statin. LDL-C reduction at day 510: −52.3% (P<0.001)³³.

ORION-11: 1,617 non-US patients with ASCVD or ASCVD risk equivalents, LDL-C ≥1.8 mmol/L (70 mg/dL). LDL-C reduction at day 510: −49.9% (P<0.001)³³.

All three trials hit their primary surrogate endpoints comfortably. Safety was generally clean; injection-site reactions occurred in 4–8% of inclisiran patients (mostly transient mild erythema). There was no significant signal for hepatic, renal, or musculoskeletal adverse effects through ~18 months³²³³.

The FDA approved inclisiran (Leqvio) on 22 December 2021 on the basis of these LDL-C trials, with the labelled indication as an adjunct to diet and maximally tolerated statin therapy in adults with HeFH or established ASCVD requiring additional LDL-C lowering³⁴. The EMA had approved earlier (December 2020).

11.3 ORION-4, VICTORION-1 PREVENT, VICTORION-2 PREVENT: outcomes still pending

The decisive trials are still maturing as of May 2026.

ORION-4 is the academic-led cardiovascular outcomes trial: ~15,000 patients aged ≥55 with established ASCVD, randomised to inclisiran or placebo, primary composite endpoint of CHD death, MI, stroke, or urgent coronary revascularisation. The trial is run by the CTSU (Oxford) with Martin Landray as PI, in collaboration with Novartis but with the academic group controlling analysis. Primary endpoint readout is scheduled for July 2026 — about two months after this chapter is written³⁵. The readout is the most-watched cardiovascular outcomes event of 2026; positive results would close the inclisiran evidence gap; null or modest results would substantially complicate the drug's positioning given its existing aggressive uptake in some health systems.

VICTORION-1 PREVENT (NCT05030428): primary prevention in high-risk patients without prior MACE; ~14,000 enrolled; estimated primary completion 2027.

VICTORION-2 PREVENT (NCT05030441): post-MI patients; ~6,500 enrolled; estimated primary completion 2026–2027.

VICTORION-INTERVENTION: acute coronary syndrome window; first patient enrolled at Duke Clinical Research Institute in 2024³⁶.

11.4 The NHS England population-health deal

In January 2022, NHS England signed a population-health-style agreement with Novartis to make inclisiran available to up to 300,000 patients per year in England³⁷. The agreement was negotiated on the basis of an undisclosed (commercial-in-confidence) price below the list price of approximately £1,987 per dose. NICE issued positive guidance (TA733) in September 2021 for use as secondary prevention in patients with persistent LDL-C ≥2.6 mmol/L despite statin therapy³⁸.

This was, and remains, controversial in three ways:

1. Outcomes evidence preceded the deal. The agreement was struck before ORION-4 read out. Critics — including the BMJ editorialists Aronow and Krumholz, and others — argued that committing a national health system to a population-scale rollout of an agent without cardiovascular outcomes evidence inverted the normal evidence-to-policy pipeline³⁹.

2. The "population-health" framing. The agreement was structured around upfront access with risk-sharing, marketed as innovative procurement. Critics noted that it functionally subsidises an industry-led behaviour change (twice-yearly injections at GP surgeries) at population scale before efficacy in that population is established.

3. Generic context. Atorvastatin and rosuvastatin are off-patent and cost the NHS pennies per day per patient. Inclisiran costs orders of magnitude more even at confidential discount. The opportunity cost — what other interventions might the same spend have funded? — has been hard to assess transparently.

Defenders — including Sir Bryan Williams (BHF/UCL) and the NHS England leadership — argued that LDL-C reduction is now a sufficiently well-validated surrogate (Mendelian randomisation, IMPROVE-IT, FOURIER) that waiting for inclisiran's own outcomes trial was an unreasonable bar, and that the adherence advantage of twice-yearly dosing made a population-health bet defensible⁴⁰.

The deal will be judged retrospectively against ORION-4's July 2026 readout. If ORION-4 hits with an effect size comparable to FOURIER/ODYSSEY scaled to its LDL-C reduction (predicted ~15–20% MACE reduction over 5 years given ~50% sustained LDL-C reduction), the NHS bet pays off. If ORION-4 misses or underdelivers, the deal becomes a textbook example of guidelines and procurement outrunning evidence.

11.5 Steelmanning the inclisiran sceptic

1. No outcomes data yet. The most important point. As of May 2026, inclisiran has surrogate (LDL-C) data only. The entire case rests on the LDL-causality argument — which is strong — but a clean surrogate is not a clean outcome. ORION-4's July 2026 readout will be the test.

2. Guidelines and procurement have run ahead of evidence. ESC/EAS, NICE, and NHS England have positioned inclisiran in lipid-management algorithms before outcomes; the AHA/ACC has been more cautious. The NHS England commissioning agreement is functionally a population-scale phase 4 trial. Whether this is innovative or premature depends on whether the surrogate-to-outcome translation holds.

3. The adherence proposition assumes the comparator (statins + PCSK9 monoclonals) cannot be made adherence-effective. Generic statins are cheap and, in many populations, are not an adherence problem — observed non-adherence often reflects nocebo/discontinuation rather than inability to take a daily pill. The adherence argument may be stronger for the PCSK9 monoclonal comparison than the statin comparison.

4. Hepatic-only mechanism is a feature and a constraint. GalNAc-siRNA only silences hepatic PCSK9; this is by design and is responsible for the clean safety profile. But it means inclisiran cannot address extrahepatic PCSK9 biology (e.g., neuronal, vascular smooth-muscle PCSK9), the long-term consequences of which are uncertain.

5. Pharmacology is brilliant; commercial structure is opaque. The NHS deal price is confidential. Cost-effectiveness modelling depends critically on the actual paid price, which is not in the public record. This makes external scrutiny of value-for-money assessments hard.

Inclisiran exemplifies the broader 2020s tension: scientifically the LDL hypothesis is settled enough to act on, but the commercial and procurement environment has accelerated faster than outcome trials can read out. ORION-4 will set the precedent for whether that acceleration is vindicated.

Chapter 12 — Bempedoic acid and the statin-intolerant niche

12.1 Mechanism: an ACL inhibitor activated only in the liver

Bempedoic acid (ETC-1002; Esperion; trade name Nexletol in the US, Nilemdo in Europe) is a small-molecule inhibitor of ATP-citrate lyase (ACL), which sits upstream of HMG-CoA reductase in the cholesterol biosynthesis pathway. Like statins, it depletes hepatic free cholesterol and upregulates LDL receptors. Two features distinguish it from statins⁴¹:

1. Prodrug activation. Bempedoic acid is a prodrug, activated to its CoA form (ETC-1002-CoA) by very-long-chain acyl-CoA synthetase 1 (ACSVL1). ACSVL1 is expressed in liver, but not in skeletal muscle. Bempedoic acid therefore has no muscle exposure to the active metabolite — which is the mechanistic basis for the claim that it spares the muscle adverse effects associated with statins.

2. Combination with ezetimibe is additive. A fixed-dose combination (bempedoic acid 180 mg + ezetimibe 10 mg) achieves LDL-C reductions of roughly 30–40% on top of background therapy, approaching low-dose statin efficacy.

LDL-C reductions in phase 3 monotherapy: ~17–28%. In combination with ezetimibe: ~30–38%⁴¹. As a non-statin LDL-lowerer with a distinct mechanism, bempedoic acid was the obvious tool for testing whether LDL reduction by yet another mechanism would deliver outcomes — and, more practically, for treating the patient subgroup labelled "statin-intolerant."

12.2 CLEAR Outcomes (Nissen, NEJM 2023): the headline trial

The Cholesterol Lowering via Bempedoic Acid, an ACL-Inhibiting Regimen (CLEAR) Outcomes trial randomised 13,970 statin-intolerant patients with established ASCVD or at high risk, to bempedoic acid 180 mg daily versus placebo. Statin intolerance was defined by patient-reported inability to tolerate at least two statins at any dose, including one at the lowest available dose, owing to adverse effects starting or worsening on statin therapy and resolving on cessation. Median follow-up: 40.6 months. Primary endpoint: four-component MACE (CV death, non-fatal MI, non-fatal stroke, or coronary revascularisation)⁴².

Baseline LDL-C: 139 mg/dL (3.6 mmol/L) — i.e., a population at substantially higher LDL-C than the typical secondary-prevention trial, reflecting their inability to take statins. LDL-C reduction at 6 months: 21.1% relative to placebo (29.2 mg/dL absolute).

Primary endpoint: 11.7% bempedoic acid vs 13.3% placebo. HR 0.87 (95% CI 0.79–0.96, P=0.004). ARR 1.6 percentage points over 40.6 months. NNT ≈ 63⁴².

Component breakdown showed CV death/non-fatal MI/stroke (3-point MACE) HR 0.85 (95% CI 0.76–0.96); MI alone HR 0.77 (95% CI 0.66–0.91); coronary revascularisation HR 0.81 (95% CI 0.72–0.92). All-cause mortality: HR 0.95 (95% CI 0.83–1.08), not significant. CV mortality: HR 0.96 (95% CI 0.81–1.13), not significant⁴².

Adverse events: bempedoic acid increased gout (3.1% vs 2.1%) and cholelithiasis (2.2% vs 1.2%), consistent with phase 2 signals. Hyperuricaemia is mechanistic — bempedoic acid inhibits renal urate transporter OAT2. Muscle adverse events (myalgia, rhabdomyolysis) were not increased relative to placebo, the design-anticipated finding given hepatic-restricted activation⁴².

12.3 The 2024–2025 follow-on analyses

CLEAR Outcomes was prespecified to support multiple subgroup and secondary analyses, several of which read out 2024–2025:

• Obesity subgroup: In patients with BMI ≥30, MACE-4 reduction was greater: HR 0.77, ARR 3.2 percentage points, RRR 23%⁴².
• Diabetes subgroup: Consistent benefit across diabetic and non-diabetic subgroups; reductions in measures of glycaemic worsening (no increase in incident diabetes, in contrast to statins).
• Total CV events (not just first events): ~20% reduction, supporting that bempedoic acid prevents recurrent as well as index events.
• 2025 AHA Scientific Sessions: a prespecified analysis showed venous thromboembolism reduction — DVT 44%, PE 39%, composite VTE 42% relative reduction⁴³. Mechanism is not fully understood; the finding parallels statin-VTE associations from JUPITER and may suggest non-LDL-mediated anti-inflammatory or coagulation effects.

The 2025 ESC/EAS focused update on dyslipidaemias upgraded bempedoic acid to a class I (or strong) recommendation in statin-intolerant secondary prevention and class IIa in combination with ezetimibe as next-line therapy after statin maximisation in patients not at LDL-C goal⁴⁴.

12.4 The statin-intolerance problem inside CLEAR Outcomes

There is one substantial methodological tension at the heart of CLEAR Outcomes that an honest critical synthesis must surface. The trial enrolled patients who self-reported statin intolerance — defined as inability to tolerate at least two statins at any dose. Patient-reported statin intolerance, however, is itself a contested clinical entity.

The SAMSON trial (Howard et al., NEJM 2020) used an n-of-1 design in 60 patients who had stopped statins due to side effects. Each patient received, in random sequence, 1-month courses of atorvastatin 20 mg, identical-looking placebo, and an empty-bottle no-tablet month. Mean symptom intensity scores: on statin 16.3, on placebo 15.4, on no tablet 8.0. Approximately 90% of the symptom burden during statin treatment was reproduced during placebo⁴⁵. The implication: most patient-reported statin intolerance is nocebo, not pharmacological.

If most "statin-intolerant" patients are not, in fact, pharmacologically intolerant, then CLEAR Outcomes' population is partly a population of nocebo-discontinuers who would (per SAMSON) have done equally well on a statin. The trial result remains valid — bempedoic acid lowered LDL-C and lowered events in that population — but the clinical claim that bempedoic acid is preferable to a statin in these patients is weaker than it appears. A more parsimonious reading: bempedoic acid is useful for patients with genuine intolerance (mechanistically confirmable by laboratory measures, rare) and as an add-on to maximal statin therapy in those not at LDL-C goal. The marketing positioning ("for statin-intolerant patients") may overgeneralise.

The harms chapter (Part IV in the full document) will discuss the SAMSON/statin-intolerance debate in depth. The CLEAR Outcomes/SAMSON tension is flagged here as a real methodological asterisk on the trial's headline result.

12.5 Steelmanning the bempedoic acid sceptic

1. Effect size per mmol/L is smaller than statins. CLEAR delivered roughly 0.75 mmol/L LDL-C reduction for 13% MACE reduction over 40 months. The CTT meta-regression for statin trials yields roughly 22% relative reduction per mmol/L LDL-C¹. Bempedoic acid's per-mmol/L effect appears about half that. Possible explanations include: shorter trial duration, statin-intolerant population at unmeasured higher baseline risk, residual confounders, or genuine drug-specific effects (the mechanism is upstream of statins but downstream of nothing useful — there is no biological reason to expect less effect than statins per mmol/L). The discrepancy deserves more analysis than it has received.

2. The intolerance definition is loose. As discussed above; if redefined more rigorously with washout and blinded rechallenge, the indicated population could be far smaller.

3. Gout and cholelithiasis are real harms. Modest in absolute terms but mechanism-linked and likely to persist long-term.

4. No mortality benefit. All-cause and CV mortality both null. Same critique as PCSK9 inhibitors; smaller absolute numbers and shorter follow-up than statin trials, so this is not definitive but is a real gap.

5. Cost-effectiveness depends on definition of comparator. Bempedoic acid is more expensive than generic statins/ezetimibe; cheaper than PCSK9 monoclonals. ICER's 2024 review found bempedoic acid cost-effective at ~$3,500/year wholesale acquisition cost under conventional thresholds for statin-intolerant secondary prevention, with materially less favourable ratios for the primary-prevention indication that Esperion has sought⁴⁶.

The advocate's case: CLEAR Outcomes is the first non-statin, non-PCSK9 randomised outcomes trial to demonstrate cardiovascular event reduction with LDL-C lowering. It is a third independent confirmation of the LDL hypothesis (IMPROVE-IT, ezetimibe; FOURIER/ODYSSEY, PCSK9 antibody; CLEAR, ACL inhibitor) — three different mechanisms, one consistent benefit. As a falsification test of the LDL hypothesis, the modern era has now offered three opportunities; all three pointed in the predicted direction.

Chapter 13 — The Lp(a) frontier

13.1 Why Lp(a) is the unfinished business

Lipoprotein(a) [Lp(a)] is an LDL-like particle with apolipoprotein(a) covalently bound to apolipoprotein B-100 via a disulfide bond. Plasma Lp(a) concentration is determined almost entirely (~90%) by the LPA gene, with very high heritability and minimal modification by diet, exercise, or weight loss. Elevated Lp(a) (typically defined as ≥125 nmol/L or ≥50 mg/dL) is present in approximately 20% of the global population, with substantial ethnic variation (highest prevalence in those of African descent)⁴⁷.

Mendelian randomisation studies in the past decade have established Lp(a) as a causal, independent risk factor for ASCVD, calcific aortic valve disease, and ischaemic stroke⁴⁸. The effect is dose-dependent and orthogonal to LDL-C — a patient with Lp(a) 200 nmol/L and LDL-C 100 mg/dL has CV risk comparable to a patient with LDL-C 160 mg/dL and normal Lp(a). For a person born with high Lp(a), the lifetime CV risk burden is significant and there has been, to date, no pharmacological intervention specifically targeting it.

A paradox compounds the problem: statins raise Lp(a) by approximately 10% in most patients, by mechanisms still incompletely understood (possibly LDL-receptor-mediated Lp(a) catabolism downregulation, or LPA transcriptional effects)⁴⁹. Niacin lowers Lp(a) by ~20–30% but its outcomes trials (AIM-HIGH 2011, HPS2-THRIVE 2014) were null or harmful and the class has been abandoned in cardiovascular prevention⁵⁰. PCSK9 monoclonal antibodies lower Lp(a) by ~25%, but FOURIER and ODYSSEY were not powered to show specific Lp(a)-mediated outcomes (although post-hoc analyses suggest some of the residual CV benefit may be Lp(a)-mediated)⁵¹.

The 2020s have therefore seen the entry of agents designed to lower Lp(a) directly and substantially. The pharmacological options pursued: antisense oligonucleotides (ASO) and small interfering RNA (siRNA), both targeting hepatic apolipoprotein(a) mRNA.

13.2 Pelacarsen (Lp(a)HORIZON): the antisense lead

Pelacarsen (TQJ230, formerly AKCEA-APO(a)-LRx) is a GalNAc-conjugated antisense oligonucleotide developed by Ionis Pharmaceuticals and licensed to Novartis. It hybridises to LPA mRNA in hepatocytes, triggering RNase H1-mediated degradation. Phase 2 (Tsimikas et al. NEJM 2020) showed dose-dependent Lp(a) reductions of 35–80% at the highest doses, with a clean safety profile⁵².

The Lp(a)HORIZON phase 3 cardiovascular outcomes trial (NCT04023552) randomised approximately 8,300 patients with established CVD and Lp(a) ≥70 mg/dL (175 nmol/L) to pelacarsen 80 mg subcutaneously monthly versus placebo. Primary endpoint: composite of CV death, non-fatal MI, non-fatal stroke, or urgent coronary revascularisation. Trial completion was scheduled for mid-2025⁵³. As of May 2026, full results have been presented and publication is anticipated imminently — the headline finding from the AHA 2025 late-breaking session is widely reported to show approximately 80% Lp(a) reduction sustained; primary outcome significance and effect size are awaited in print⁵³. [CITATION NEEDED — final Lp(a)HORIZON publication]

13.3 Olpasiran (OCEAN(a)): the siRNA challenger

Olpasiran (AMG-890) is Amgen's GalNAc-siRNA targeting LPA. Phase 2 (OCEAN(a)-DOSE, O'Donoghue et al., NEJM 2022) randomised 281 patients with Lp(a) >150 nmol/L; the higher doses (75 mg q12w, 225 mg q12w, 225 mg q24w) reduced Lp(a) by 95–101% versus placebo (i.e., to or near the assay's lower limit of detection) sustained over 48 weeks⁵⁴.

OCEAN(a)-OUTCOMES phase 3 (NCT05581303) is enrolling ~6,000 patients with established ASCVD and Lp(a) ≥200 nmol/L, randomised to olpasiran 225 mg q12w versus placebo, primary composite CHD death/MI/urgent coronary revascularisation. Estimated primary completion 2027⁵⁵.

13.4 Lepodisiran (ACCLAIM-Lp(a)): the single-dose contender

Lepodisiran (LY3819469) is Eli Lilly's GalNAc-siRNA, distinguished pharmacologically by its remarkable durability. A single 608 mg dose reduced Lp(a) by approximately 94% at day 60 and 94% at day 360 — i.e., near-complete suppression for a year after one injection (Nissen et al., JAMA 2024)⁵⁶. The phase 3 ACCLAIM-Lp(a) trial (NCT06292013) is enrolling ~12,500 patients with elevated Lp(a) and high CV risk; the dosing regimen permits annual or semi-annual injection. Estimated primary completion late 2029⁵⁷.

13.5 Why this matters and steelmanning the sceptic

The Lp(a) programme is, in a meaningful sense, the field's attempt to extend the LDL-causality framework to a second, orthogonal apoB-containing particle. The Mendelian randomisation evidence is strong; the pharmacological tools now achieve near-complete Lp(a) suppression; and the affected population is large (~1.5 billion people globally with elevated Lp(a)). If the outcomes trials hit, the Lp(a) chapter will close the residual-risk gap for that 20% of patients whose CV risk is genetic and untreatable by current means.

Steelman the sceptic on Lp(a):

1. No outcomes data yet. Pelacarsen's full phase 3 publication is imminent but not, at the time of writing, in print with primary-endpoint hazard ratios. Olpasiran and lepodisiran are years from readout. The Mendelian randomisation argument is strong but is, in the end, an inference about pharmacological intervention. The same inference for niacin (Lp(a)-lowering, MR-supported) failed in outcomes (AIM-HIGH, HPS2-THRIVE).

2. The MR-pharmacology translation may not be 1:1. Lifelong genetic reduction may capture vascular biology that 5 years of pharmacological reduction in older patients does not. Pelacarsen's Lp(a)HORIZON readout will be the first test.

3. The "Lp(a)-causal-by-MR" finding is correlational at the trial level until proven by RCT. This is precisely the methodological position the field was in for LDL-C in 1985 — which then took 35 years and dozens of trials to settle. Lp(a) is now where LDL was at the time of the Coronary Drug Project.

4. Cost will be high. GalNAc-siRNA development is expensive; the indicated population is large but stratified (only the highest Lp(a)); commercial pricing will set the population-health affordability of any positive readout.

5. Statin-Lp(a) paradox needs explanation. Why do statins raise Lp(a) by ~10% while delivering large cardiovascular benefit? The most parsimonious answer is that the LDL-C reduction overwhelms the small Lp(a) increase. But for very-high-Lp(a) patients, this calculation may not hold — and current guidelines do not stratify statin recommendations by Lp(a).

If the Lp(a) outcomes trials hit, the 2030s lipid-management paradigm will look very different from the 2020s: LDL-C lowering will remain the foundation, with PCSK9/siRNA add-ons for high-risk patients, but a parallel Lp(a)-specific track will run alongside for the 20% with the genetic burden. If the outcomes trials miss, the field will be back to where niacin left it — a strong MR signal, a failed translation, and an open question.

Chapter 14 — The 2024–2026 frontier and pending trials

14.1 STAREE: atorvastatin in healthy elderly

The Statins in Reducing Events in the Elderly (STAREE) trial is the larger of the two pivotal pending elderly primary-prevention trials. STAREE randomised 9,971 community-dwelling Australians aged ≥70 with no prior cardiovascular disease, diabetes, or dementia, to atorvastatin 40 mg daily versus placebo. Mean age at randomisation 74.7 years; 40% aged ≥75; 52% women⁵⁸.

The trial has two co-primary endpoints, structured to reflect the patient-relevant decision about whether to take a statin in this age group:

1. Disability-free survival — composite of death, dementia, or persistent physical disability.
2. Major cardiovascular events — composite of cardiovascular death, non-fatal MI, non-fatal stroke, and (in the most recent protocol amendment, 2024) coronary revascularisation⁵⁸.

The co-primary structure is methodologically important. A trial powered only on MACE risks declaring statin therapy "beneficial" in a population for whom the disabling complications of polypharmacy, falls, or cognitive change may outweigh the cardiovascular gain. STAREE was designed to confront this trade-off directly.

Anticipated trial completion: Q4 2025. Primary endpoint readout: 2026. The trial is funded by the Australian National Health and Medical Research Council; Pfizer supplied atorvastatin and placebo but has no role in analysis or interpretation⁵⁸.

14.2 PREVENTABLE: the survival-free-of-dementia primary endpoint

The Pragmatic Evaluation of Events and Benefits of Lipid-lowering in older adults (PREVENTABLE) trial is the second pivotal elderly trial and is structurally the more ambitious of the two. PREVENTABLE randomised 20,000 community-dwelling US adults aged ≥75 with no CVD, no disability, and no dementia at baseline, to atorvastatin 40 mg daily versus placebo. The primary endpoint is survival free of new dementia or persistent disability — the first major statin trial in which the primary endpoint is not a cardiovascular event but a composite reflecting the older patient's lived-experience priorities⁵⁹.

The co-secondary endpoints include the conventional CV composite, mild cognitive impairment/dementia, and the cardiovascular components individually. The trial is pragmatic in design (real-world recruitment via primary care and Medicare claims, simplified protocol, broad eligibility) and is funded by the US National Institutes of Health (NIH/NIA/NHLBI), with academic leadership through Duke Clinical Research Institute and the Wake Forest School of Medicine⁵⁹.

Expected primary endpoint readout: December 2026. PREVENTABLE will be — if it reads out cleanly — the single most consequential trial for the elderly-statin question, because its primary endpoint is the precise question patients and clinicians ask: will this statin help me live well, or merely live longer with treatment burden?

If PREVENTABLE shows survival-free-of-dementia improvement: elderly-statin guidelines will shift from "consider on individual basis" (current ACC/AHA, NICE, ESC stance for ≥75-year-olds) to "recommend." If PREVENTABLE shows no improvement, or harm on disability/dementia: the field will face a real reckoning, because much of the case for elderly statins currently rests on extrapolation from younger-cohort RCTs (HOPE-3, JUPITER subgroup analyses) and observational data. The Cochrane and Byrne meta-analyses (below) already foreshadow a small primary-prevention effect; PREVENTABLE is the cleanest opportunity for the field to test the elderly-specific question prospectively.

14.3 VESALIUS-CV: PCSK9 inhibition in primary prevention

VESALIUS-CV (NCT03872401) is Amgen's primary-prevention outcomes trial for evolocumab. It enrolled approximately 12,300 patients aged ≥50 (men) or ≥55 (women) at high cardiovascular risk but without prior MI or stroke, with LDL-C ≥70 mg/dL despite background statin therapy, randomised to evolocumab versus placebo. Primary composite: MACE-3 (CV death, MI, stroke). Co-primary: MACE-4 (adding ischaemia-driven coronary revascularisation). Estimated primary completion 2027–2028⁶⁰.

VESALIUS-CV is the trial that will determine whether PCSK9 inhibitors should be deployed in primary prevention at all — currently a guideline grey zone, with payers generally restricting to secondary prevention or FH.

14.4 The 2024 AHA PREVENT equations and the eligibility-reset question

In November 2023, the AHA released the PREVENT (Predicting Risk of Cardiovascular Disease Events) risk equations, replacing the 2013 Pooled Cohort Equations as the basis for US ASCVD risk estimation⁶¹. PREVENT was developed from a larger and more diverse cohort (~6.6 million person-years), includes 30-year risk estimation alongside 10-year, incorporates kidney function, and explicitly removes race as a model input.

The clinical consequence of PREVENT is a substantial recalibration of estimated risk for many adults — generally downward. Khan and colleagues (JAMA 2024) modelled the impact: applying PREVENT instead of the Pooled Cohort Equations would reclassify approximately half of US adults to lower estimated ASCVD risk; in the population aged 40–75 not currently on a statin, approximately 15.8 million Americans would become statin-ineligible under PREVENT-aligned thresholds versus Pooled Cohort Equation thresholds⁶².

This is a major guideline upheaval brewing. The 2018 ACC/AHA cholesterol guideline used the Pooled Cohort Equations to determine statin eligibility, with a 10-year ASCVD risk ≥7.5% triggering moderate-intensity statin consideration. If PREVENT is adopted as the basis for the next major guideline update (anticipated 2026–2027), the result is not a change in the underlying biology — LDL-C remains causal, statins remain effective — but a substantial reduction in the primary-prevention population for whom statins are guideline-recommended. This is, in a real sense, the field internally acknowledging that the previous risk equations overestimated risk and over-recommended statin therapy.

The interaction with PREVENTABLE is consequential. If PREVENTABLE finds modest or null benefit in elderly primary prevention, and PREVENT recalibrates risk downward across the board, the combined effect is a likely guideline narrowing of statin indications for primary prevention — particularly in the elderly. This would be the most significant reversal in cholesterol-treatment guidelines since the 2013 ATP-IV/ACC-AHA shift away from LDL-C targets.

14.5 Recent meta-analyses: Byrne 2022 and beyond

The most-cited recent meta-analytic challenge to broad statin use in primary prevention is Byrne, Ennis, Cardwell, Lavin Plicht and Smith, JAMA Internal Medicine 2022 — titled, characteristically, Statin Treatment for Primary Prevention of CVD: Time to Curb Our Enthusiasm. The authors pooled 21 RCTs (n=140,000) and reported small absolute risk reductions: 0.8% for all-cause mortality (NNT 125), 1.3% for MI (NNT 77), and 0.4% for stroke (NNT 250) over an average ~4.4-year follow-up⁶³.

The Byrne meta-analysis became a flashpoint. CTT investigators (Collins, Baigent, Reith, Emberson) responded that the meta-analysis pooled trials of heterogeneous statin intensity and that per-mmol/L analysis is the appropriate framework: by their analysis, the same data yielded ~21% relative risk reduction per mmol/L LDL-C reduction, consistent with the CTT meta-regression⁶⁴. Byrne and colleagues responded that absolute risk reduction is the patient-relevant metric and that the per-mmol/L framing obscures the small absolute benefit in primary prevention.

The methodological dispute is genuinely substantive: both sides are correct that their preferred metric is informative; both are correct that the other metric can mislead in isolation. For a 25-year marketing director debating with a Roche-affiliated friend, the salient point is this: the relative risk reduction in primary prevention is real and consistent with the LDL hypothesis; the absolute risk reduction in low-risk primary prevention is small, on the order of 1 percentage point over 4–5 years; whether that is worth the costs and burdens of long-term statin therapy is a values judgement, not a science judgement.

A 2025 systematic review and meta-analysis of 26 randomised trials (n>180,000) found that adherent statin users in primary prevention experienced a 32% relative reduction in CV events versus non-adherent users in the same trials — emphasising that the "small benefit" finding partly reflects the dilution of intention-to-treat analysis by non-adherence⁶⁵. This adherence-stratified analysis is a partial reconciliation: the per-protocol effect is larger than the intention-to-treat effect, particularly in primary prevention where motivation for daily adherence is weaker than in post-MI populations.

14.6 Master trial table: 2008–2026 LDL-lowering outcomes trials

NS = not statistically significant. All hazard ratios are for primary endpoint unless noted. Cell entries reflect published or pre-published values as of May 2026.

14.7 Synthesis for Part II-B

Three propositions earn the chapter's confidence:

1. The mechanism question is settled. Three independent non-statin mechanisms (NPC1L1 inhibition, PCSK9 antibody blockade, ACL inhibition) have all delivered statistically significant reductions in atherothrombotic events. The LDL-particle hypothesis is no longer the only viable explanation for statin benefit; it is the established mechanism shared by every successful LDL-lowering pharmacotherapy.

2. Absolute effect sizes are modest and time-dependent. Trial-duration effects (NNTs in the 50–70 range over 2–6 years) are far smaller than the lifetime effect sizes implied by Mendelian randomisation. FOURIER-OLE's legacy finding is the first within-trial evidence that the cumulative-exposure model holds pharmacologically. This implies that earlier initiation in higher-risk individuals captures more of the available benefit than later or shorter intervention.

3. The elderly primary-prevention question is genuinely unresolved. STAREE and PREVENTABLE will deliver, in 2026, the cleanest answer the field has ever had to whether statin therapy in healthy adults aged ≥70/75 improves the metric patients actually care about — survival free of dementia and disability. The 2024 PREVENT risk equations have already begun to reshape the eligibility landscape. The 2026 readouts will determine whether the field's enthusiasm for elderly primary prevention is empirically vindicated or curtailed.

Three propositions earn the chapter's uncertainty:

1. Mortality benefit beyond statins is unclear. No non-statin LDL-lowering agent (ezetimibe, PCSK9, bempedoic acid) has produced unambiguous all-cause mortality reduction. ODYSSEY OUTCOMES came closest. Whether this is a power problem (trials too short, populations too small) or a ceiling-effect on already-statin-treated patients remains debated.

2. Inclisiran's evidence base is incomplete. ORION-4 in July 2026 will be a defining moment. Pre-emptive procurement (NHS England) and guideline positioning may be vindicated or embarrassed.

3. The Lp(a) outcomes question is open. Pelacarsen, olpasiran, and lepodisiran all achieve near-complete Lp(a) suppression. Whether that translates to cardiovascular event reduction will define the 2027–2030 lipid agenda.

The next chapter (Part III: Mendelian randomisation and the LDL-causality framework) revisits the genetic evidence that licenses the 2010–2026 pharmacological programme — both its successes and the gaps the trials have not yet closed.

Footnotes

Part III — The Disputes: Who Is Arguing, What They Are Arguing About, Who Is Funding It

"In science the credit goes to the man who convinces the world, not to the man to whom the idea first occurs." — Francis Darwin

The cholesterol question has not been settled by a single decisive experiment; it has been settled (or unsettled, depending on whom you ask) by thirty years of cumulative evidence produced by a relatively small number of named investigators, working in a relatively small number of named institutions, funded by a relatively small number of named sponsors. To debate the question substantively, one must know who these people are, what they actually believe, what their strongest argument is, and what each of them stands to gain or lose. This Part is a prosopography — a collective biography of the dispute. It is written with two operating rules. First, no strawmen. Every figure named here is given their best one-sentence argument, in their own framing, before any counter-pressure is applied. Second, no hagiography. Every figure named here is also given the strongest reasonable critique of their position, including disclosed and undisclosed industry relationships. The reader can then form their own judgment.

Chapter 15 — The Consensus Camp

The "consensus camp" is not a club; it is an emergent property of three overlapping institutions: the Cholesterol Treatment Trialists' (CTT) Collaboration at Oxford, which holds the individual-participant trial data; the TIMI Study Group at Brigham and Women's / Harvard, which has run most of the major industry-sponsored outcomes trials of the last 30 years; and the guideline panels of the American College of Cardiology / American Heart Association (ACC/AHA) and the European Society of Cardiology / European Atherosclerosis Society (ESC/EAS). Below are the principal figures, their strongest arguments, the evidence they lean on hardest, their disclosed industry relationships, and — where it can be documented — the places they have publicly changed their minds. Steelmanned.

15.1 Colin Baigent — director, CTT Collaboration, Oxford CTSU

Sir Colin Baigent is, by any reasonable accounting, the single most influential epidemiologist in the statin debate. As director of the Clinical Trial Service Unit (CTSU) at Oxford and lead of the Cholesterol Treatment Trialists' Collaboration, he sits on the individual-participant data (IPD) from essentially every major statin randomised trial ever conducted.¹

Strongest one-sentence argument. "Across 28 randomised trials of more than 174,000 participants, each 1 mmol/L (≈39 mg/dL) reduction in LDL-C produced a 22% proportional reduction in major vascular events per year of treatment, with the effect proportional to LDL reduction, consistent across baseline-risk subgroups, and undiminished by sex, age above or below 75, or diabetes status — and the absolute benefit therefore scales with the patient's absolute risk."²

Evidence he leans on hardest. The successive CTT meta-analyses of individual-participant data — Lancet 2005, Lancet 2010, Lancet 2012 (women and low-risk subgroup), Lancet 2015 (CKD), Lancet 2019 (age >75), and the 2024 reanalysis on muscle symptoms in JAMA Internal Medicine which concluded that virtually all the muscle pain attributed to statins in clinical practice is in fact nocebo.³ The CTT design — pooling raw patient-level data rather than relying on aggregate published numbers — is methodologically the strongest tool available, if one accepts that the underlying trials are themselves valid.

Disclosed industry relationships. Baigent's department, the CTSU, has had research funding from Merck, Bayer, Boehringer Ingelheim, Novartis, and others, but its institutional policy — uniquely among the major lipid groups — is that all such funding flows to the university, that no investigator or staff member receives personal payments or honoraria from industry, and that all trial protocols and analyses are controlled by the CTSU. The CTSU has publicly described this as the "Oxford model."⁴ Critics argue that institutional funding is still a conflict; defenders argue it is the cleanest available compromise given the cost of large trials.

Where he has moved. Baigent has tightened, not loosened, the boundary conditions of the consensus. After the JUPITER-era enthusiasm for treating very-low-risk primary prevention populations, the CTT 2012 paper specifically endorsed treatment in 5-year risk groups as low as 5% — but the CTT 2019 paper on the elderly was notably more cautious, concluding that for over-75s in primary prevention "direct evidence from randomised trials is limited and the benefit of statin therapy is less certain."⁵ He has been a consistent and vocal critic of journalistic statin scepticism, particularly the BMJ statin-side-effects controversy of 2013–2014; he was the principal mover behind the BMJ's later partial retraction.⁶

15.2 Rory Collins — head of CTSU, BHF Professor of Medicine and Epidemiology, Oxford

Sir Rory Collins is Baigent's senior at the CTSU and the architect of the IPD model on which the CTT depends. He is the most prominent public defender of statins against media scepticism.

Strongest one-sentence argument. "The randomised evidence — collected, audited, and reanalysed at the patient level — shows that the only common side effect attributable to statins beyond placebo is a small excess of new-onset diabetes, while the cardiovascular benefits prevent roughly 1,000 major vascular events per 10,000 high-risk patients treated for five years; the public conversation about statin harms has been distorted by observational data of inferior quality."⁷

Evidence he leans on hardest. The Lancet 2016 "Interpretation of the evidence for the efficacy and safety of statin therapy" — a 30,000-word position paper that is effectively the consensus camp's manifesto.⁸ It catalogues every claimed adverse effect against the IPD evidence.

Disclosed industry relationships. Same as Baigent — CTSU institutional funding from multiple industry sources, no personal honoraria. Collins has been criticised (by Fiona Godlee, then editor of the BMJ, and by the heterodox camp) for refusing to allow non-CTSU researchers access to the IPD for independent reanalysis; the CTT position is that the data-sharing agreements signed with the original trial sponsors prohibit broader release.⁹

Where he has moved. Collins's positions have been remarkably stable since the late 1990s. He has, however, become substantially more public-facing — partly in response to the 2013 BMJ controversy and the perception inside the CTSU that the field had ceded the public conversation to the heterodox camp.

15.3 Christina Reith and Jonathan Emberson — CTT senior statisticians

Less visible than Baigent and Collins but operationally central. Reith and Emberson are the statisticians who actually run the CTT IPD analyses. Emberson is a co-author on nearly every CTT paper from 2010 forward; Reith led the 2022 BMJ paper on statin safety in older people and the 2024 muscle-symptoms analysis.¹⁰

Strongest one-sentence argument. "When you pool individual patient data rather than summary statistics, you can adjust for baseline covariates, run subgroup analyses with proper interaction tests, and check for differential follow-up — and when you do this for statins, the proportional benefit and the absence of major harm are extraordinarily stable across every cut of the data."

Industry relationships. Same institutional model.

15.4 Eugene Braunwald — TIMI Study Group founder, Hersey Distinguished Professor, Harvard

Eugene Braunwald is, in cardiology, what Linus Pauling was to chemistry — a figure whose textbook (Heart Disease, now in its 12th edition) has trained essentially every cardiologist in practice. He founded the Thrombolysis in Myocardial Infarction (TIMI) Study Group at Brigham and Women's Hospital in 1984.¹¹ At the time of writing he is 96 and still active.

Strongest one-sentence argument. "The convergence of evidence — from epidemiology, from mechanistic biology, from animal models, from genetic experiments of nature, and from increasingly clean randomised trials with progressively stronger LDL-reduction agents — has reached the point where the LDL-C causal hypothesis is no longer disputable by any rational reading; the remaining clinical questions are about how aggressively, in whom, and with what agent, not about whether."¹²

Evidence he leans on hardest. The trajectory of TIMI trials from the original thrombolysis era through the IMPROVE-IT, FOURIER, and ODYSSEY OUTCOMES non-statin LDL-lowering trials. His point is structural: every time a new mechanism is added (HMGCR with statins, NPC1L1 with ezetimibe, PCSK9 with monoclonals, ATP-citrate lyase with bempedoic acid) the same dose-response relationship emerges. Mechanism-independent convergence is the strongest possible epidemiological argument short of randomised assignment from birth.

Disclosed industry relationships. Substantial. Braunwald has had consulting and research relationships with Merck, Daiichi-Sankyo, Bristol-Myers Squibb, AstraZeneca, Sanofi, Bayer, Novartis, Amgen, and others — all disclosed in TIMI publications.¹³ The TIMI Study Group itself receives industry funding for trial execution. The defence offered is that TIMI conducts the trials under its own academic governance and publishes regardless of result; ODYSSEY OUTCOMES, IMPROVE-IT, and FOURIER all reported endpoints that were either modest or in the case of IMPROVE-IT initially expected to be negative.

Where he has moved. Braunwald was famously sceptical of the lipid hypothesis in the early 1980s, before the publication of the 4S trial in 1994; his memoirs describe the personal and intellectual shift.¹⁴ He has more recently flagged that the benefit-cost calculus for low-risk primary prevention is genuinely uncertain.

15.5 Marc Sabatine — TIMI Study Group chair, FOURIER PI, BWH/Harvard

Sabatine is Braunwald's successor as the operational head of TIMI and the principal investigator on FOURIER (evolocumab) and its long-term open-label extension (FOURIER-OLE).¹⁵

Strongest one-sentence argument. "Lowering LDL-C to levels never previously achievable in clinical medicine — median 30 mg/dL on evolocumab plus statin — produced a continued linear reduction in cardiovascular events with no plateau, no excess of neurocognitive harm, no diabetes signal at 7+ years, and a legacy effect where patients on evolocumab from the start did better than late-starters even after both arms were equalised in the open-label extension, which proves that lowering LDL earlier and lower is causally protective."¹⁶

Evidence he leans on hardest. FOURIER (NEJM 2017) plus FOURIER-OLE (Circulation 2023). The legacy-effect argument is rhetorically powerful because it operationalises the Mendelian-randomisation insight (lifetime exposure matters more than peak reduction) inside a randomised framework.

Disclosed industry relationships. Extensive: principal investigator grants from Amgen (for FOURIER), Merck, AstraZeneca, Novartis, Pfizer, Daiichi-Sankyo, Bayer, Bristol-Myers Squibb. Personal consulting fees from many of the same. All disclosed in FOURIER and FOURIER-OLE publications.¹⁷

Where he has moved. Sabatine has been notably consistent. He has been firmer than some in arguing that the FOURIER neurocognitive safety findings effectively close the door on the "low LDL causes dementia" hypothesis — a position that 10 years ago would have been considered premature.

15.6 Paul Ridker — JUPITER PI, Brigham and Women's / Harvard

Paul Ridker is the most interesting member of the consensus camp because he is in part a heretic from within. He is the principal investigator of JUPITER (rosuvastatin in low-LDL, high-CRP patients, 2008), which expanded statin use to a previously untreated population, but he is also the principal architect of the inflammatory hypothesis of atherosclerosis — which displaces some of the explanatory weight off LDL onto interleukin-6 and high-sensitivity CRP.¹⁸

Strongest one-sentence argument on cholesterol. "LDL is causal and statins prevent events, but the residual cardiovascular risk after maximally tolerated LDL reduction is driven by inflammation — measurable by hsCRP — and that residual risk is itself modifiable, as demonstrated by the canakinumab CANTOS trial which reduced events independent of any change in LDL."¹⁹

Evidence he leans on hardest. JUPITER (NEJM 2008), CANTOS (NEJM 2017, canakinumab), and the increasingly extensive hsCRP-stratification literature. In 2023 he published an influential paper in The Lancet arguing that, in statin-treated patients, baseline hsCRP predicted future events more strongly than baseline LDL.²⁰

Disclosed industry relationships. Ridker has consulting and research relationships with AstraZeneca (JUPITER), Novartis (CANTOS), Pfizer, and others. He also co-owns patents on the use of hsCRP for cardiovascular risk prediction — a disclosed conflict that has drawn sharp criticism, particularly during the JUPITER period when the trial was simultaneously expanding statin use and validating the test he held patents on.²¹

Where he has moved. Substantially. In the 2008–2012 period Ridker was the public face of "treat-to-CRP" in primary prevention. By the late 2010s he had moderated, increasingly emphasising that hsCRP and LDL are independent and additive risk targets and that residual inflammatory risk is the next therapeutic frontier — a more nuanced position than the original JUPITER framing implied.

15.7 Steven Nissen — Cleveland Clinic, CLEAR Outcomes PI

Steven Nissen is, like Ridker, a partial heterodox inside the consensus tent. He has been a vocal industry critic — most famously over Avandia (rosiglitazone), Vioxx (rofecoxib), and most recently over the FDA's handling of ezetimibe and inclisiran — while remaining a principal investigator on the major industry-funded lipid trials.²²

Strongest one-sentence argument on cholesterol. "In CLEAR Outcomes, a 4-year trial in 13,970 statin-intolerant patients, bempedoic acid produced a 13% reduction in 4-point MACE and a 23% reduction in MI on top of 21% LDL reduction with no excess of muscle symptoms — proving that the LDL hypothesis is mechanism-independent and that the 'statin-intolerant' population can still be protected by alternative LDL-lowering."²³

Evidence he leans on hardest. CLEAR Outcomes (NEJM 2023) and the prespecified 2024–2025 analyses (obesity, diabetes subgroups, total events, VTE reduction).²⁴ He also draws on his long-standing IVUS (intravascular ultrasound) imaging work showing plaque regression at very low LDL.²⁵

Disclosed industry relationships. Substantial and recurring. Cleveland Clinic receives research support from Esperion (for CLEAR Outcomes), Amgen, Pfizer, Merck, Eli Lilly, AstraZeneca, Novartis, Silence Therapeutics, and others; Nissen personally accepts no honoraria, royalties, or speaking fees, with all consulting income paid directly to charity. This is an unusual model that he has held to publicly for two decades.²⁶

Where he has moved. Nissen was an early and harsh critic of the ENHANCE trial delay (ezetimibe + simvastatin, Merck/Schering-Plough; see §19.4 below), arguing publicly that the sponsors had sat on negative imaging data for over a year. He has since been less harsh about ezetimibe in the wake of IMPROVE-IT (NEJM 2015), where the drug did reduce events. He has also been notably willing to call out unfavourable trials he led (he was PI on dal-OUTCOMES, the failed Roche CETP-inhibitor trial that ended Roche's lipid therapeutic ambitions; see §20.1 below).

15.8 Neil Stone and Scott Grundy — ACC/AHA guideline architects

Neil Stone (Northwestern) chaired the 2013 ACC/AHA cholesterol guideline panel — the watershed document that abolished LDL targets in favour of fixed-intensity statin therapy stratified by 10-year ASCVD risk.²⁷ Scott Grundy (UT Southwestern), one of the originators of the metabolic syndrome construct and chair of the NCEP ATP III panel in 2001, then chaired the 2018 ACC/AHA update which partially restored a "consider LDL <70" framework in very-high-risk secondary prevention.²⁸

Strongest one-sentence argument. "Treatment decisions should be anchored to estimated absolute risk and statin intensity, not to a moving LDL target, because the trial evidence supports proportional benefit from a fixed-intensity dose rather than the achievement of a particular post-treatment number — though in very-high-risk secondary prevention adding non-statin therapy to reach LDL <70 mg/dL is supported by IMPROVE-IT and FOURIER."

Disclosed industry relationships. Stone disclosed no industry consulting or research income for the 2013 guideline process (a deliberate choice by the panel to use only conflict-free chair-leads). Grundy has had a longer industry history (Merck, AstraZeneca, Pfizer) but was conflict-free at the time of the 2018 panel.²⁹ Other guideline-panel members across both cycles had varying industry exposure; both the 2013 and 2018 documents include full disclosures.

Where they have moved. The trajectory from ATP III (2001) to 2013 to 2018 is itself an admission of evolving uncertainty — from per-mg/dL LDL targets, to fixed-intensity therapy, to a hybrid model. This is not embarrassment; it is the guideline process working as it should, but it is also the reason heterodox critics argue that "the science is settled" is rhetorical overstatement.

15.9 Alberico Catapano and François Mach — ESC/EAS guideline architects

Catapano (Milan) and Mach (Geneva) co-chaired the 2019 ESC/EAS Guidelines for the Management of Dyslipidaemias — the document that pushed European LDL targets below American ones (LDL <55 mg/dL for very-high-risk, <40 mg/dL for recurrent events).³⁰ The 2025 ESC/EAS focused update extended the framework to bempedoic acid.³¹

Strongest one-sentence argument. "The cumulative LDL-exposure principle, supported by Mendelian randomisation and by the FOURIER-OLE legacy effect, justifies aggressive LDL targets in high-risk patients — lower for longer, started earlier, is the most biologically coherent translation of the evidence into clinical practice."

Disclosed industry relationships. Both have extensive disclosed consulting and lecture relationships with Amgen, Sanofi, Regeneron, Daiichi-Sankyo, Pfizer, MSD, Servier, Mylan, Novartis, AstraZeneca. Full disclosures in the 2019 and 2025 documents.³² The European guideline process — unlike the 2013 ACC/AHA process — does not insist on conflict-free chairs.

Where they have moved. The successive ESC/EAS targets have tightened over time (LDL <100 in 2003, <70 in 2011, <55 in 2019). Critics argue this is creeping medicalisation; defenders argue it is straightforward application of "more is better" evidence to higher-risk strata.

15.10 John Kastelein — University of Amsterdam, lipid-trial veteran

Kastelein has been the principal European investigator on a long sequence of major lipid trials including the ENHANCE imaging study of ezetimibe, the dal-OUTCOMES CETP trial, the RADIANCE-1 CETP trial, several mipomersen and lomitapide trials in familial hypercholesterolaemia, and recent ANGPTL3 and Lp(a) work.³³

Strongest one-sentence argument. "Familial hypercholesterolaemia is the human knockout experiment for the LDL hypothesis — patients with monogenic LDL elevation have premature coronary disease, and lowering their LDL with statins, PCSK9 inhibitors, or other modalities normalises their event rate toward the population baseline."

Evidence he leans on hardest. The Dutch FH registry and successive trials in FH populations.

Disclosed industry relationships. Among the most extensive in the field — consulting and research with Amgen, Sanofi, Regeneron, Pfizer, AstraZeneca, Merck, Novartis, Daiichi-Sankyo, Esperion, Ionis, Akcea, and others.

15.11 Brian Ference — University of Cambridge, Mendelian randomisation lead

Brian Ference is the principal architect of the lifetime-exposure framing of Mendelian randomisation as applied to LDL and cardiovascular disease. His 2017 EAS Consensus paper (with Catapano, Sabatine, and others) is the document most often cited as the "MR-RCT-observational convergence" steelman of LDL causality.³⁴

Strongest one-sentence argument. "Naturally occurring genetic variants that lower LDL-C from birth produce approximately three times the per-mg/dL coronary risk reduction observed in trials of statins started in middle age — which is exactly what one would predict if the relationship is causal and the relevant exposure is cumulative lifetime LDL-particle-years rather than peak LDL at any moment."³⁵

Evidence he leans on hardest. The lifetime-exposure meta-analyses combining naturally occurring variants in PCSK9, HMGCR, NPC1L1, LDLR, APOE, APOB, LPA — discussed in detail in Chapter 18.

Disclosed industry relationships. Consulting and research with Amgen, Merck, Pfizer, Esperion, Novartis, Sanofi, Regeneron, and others.³⁶

Where he has moved. Ference has been consistent on the LDL causality argument. He has, however, been more nuanced in recent papers about what MR can and cannot tell you about harm — explicitly conceding that genetic variants do not provide information about drug-specific off-target effects.

15.12 The consensus camp, summarised

What ties these figures together is not a coordinated position but a shared evidentiary commitment: they prioritise pooled randomised data and mechanism-converging Mendelian randomisation, they distrust observational signal-detection for harm (because of confounding by indication and by adherence) and for benefit (because of healthy-user bias), and they accept industry funding as a fact of late-20th-century cardiovascular research that is best managed through institutional firewalls (Oxford model) or personal-honoraria-to-charity policies (Nissen model) rather than refused outright. Their strongest single argument is the convergence of evidence across multiple LDL-lowering mechanisms — statins, ezetimibe, PCSK9 inhibitors, bempedoic acid, and (by genetic proxy) Lp(a) — onto the same dose-response slope. This is, in the formal Bradford-Hill sense, the strongest available evidence short of a randomised assignment of LDL from birth.³⁷

Chapter 16 — The Heterodox Camp

The heterodox camp is — unlike the consensus camp — actually a coalition of fairly distinct positions. Some heterodox figures dispute the LDL-causality hypothesis at the foundational level; others accept LDL causality but dispute the clinical relevance of the statin trials; still others accept both but dispute the risk-benefit calculus in primary prevention. The reader must hold these positions apart, because the consensus camp's strongest rhetorical move is to lump them together as one position and then dismiss the weakest version. We will do the opposite: name each figure, give them their best argument, and identify where each genuinely overreaches.

16.1 Uffe Ravnskov — founder, THINCS (The International Network of Cholesterol Sceptics)

Uffe Ravnskov is a Danish-Swedish independent researcher (originally a nephrologist) who has been the most persistent foundational sceptic of the lipid-heart hypothesis since the mid-1990s. He founded THINCS in 2002. His books The Cholesterol Myths (2000) and Fat and Cholesterol Don't Cause Heart Attacks (2009) are the founding texts of the modern sceptic movement.³⁸

Strongest one-sentence argument. "The epidemiological correlation between blood cholesterol and coronary mortality is weak, inconsistent across populations (the 'French paradox', the Maasai, the Inuit), absent or inverted in the elderly where most cardiovascular death actually occurs, and the trial-level reduction in coronary mortality from statins has been progressively smaller as trial methodology has tightened — all of which suggests that LDL is, at most, a weak and confounded marker rather than the causal driver of atherosclerosis."³⁹

Evidence he leans on hardest. (1) Population-comparison data showing weak or absent correlation between national cholesterol levels and CHD mortality outside the Seven Countries Study sample.⁴⁰ (2) The MRFIT primary-prevention data, where cholesterol-lowering by diet did not reduce total mortality. (3) The Honolulu Heart Program and other elderly-cohort data showing inverse association between low cholesterol and mortality. (4) The progressively shrinking effect size in successive statin trials as blinding, monitoring, and reporting tightened — specifically the contrast between 4S/CARE/LIPID (large effects, 1994–1998, modest monitoring) and ALLHAT-LLT (no mortality benefit, NIH-funded pravastatin arm).

Industry-funding posture. Ravnskov has, to public knowledge, never accepted industry funding. This is both his epistemic strength (no commercial conflict) and the reason the consensus camp dismisses him (no institutional standing, no IPD access, no large-trial leadership). He has been called a "denialist" in the same rhetorical mode as climate sceptics — a comparison he has angrily rejected.⁴¹

Where he overreaches. Three places, in our reading. (1) He treats Mendelian randomisation findings inconsistently — he was willing to cite MR when it produced conclusions friendly to his position (e.g., low Lp(a) variants) but has been dismissive of MR on LDL/PCSK9, arguing without strong technical grounding that the variants are confounded by pleiotropy.⁴² (2) He has occasionally argued for an infectious aetiology of atherosclerosis (chronic Chlamydia pneumoniae, etc.) on evidence that is at best preliminary. (3) He has at times conflated "weak effect on total mortality" with "no effect on coronary mortality," which is a different claim and one the trial evidence does not support in secondary prevention.

16.2 Michel de Lorgeril — Lyon Diet Heart Study PI, CNRS Grenoble

Michel de Lorgeril is the most theoretically serious of the heterodox figures, in that his critique is rooted not in dismissing all atherosclerosis intervention but in arguing that the Mediterranean dietary pattern — not LDL reduction — is the active ingredient in cardiovascular prevention. He is the principal investigator of the Lyon Diet Heart Study (Circulation 1999), which remains one of the largest secondary-prevention trials of a non-pharmacological intervention.⁴³

Strongest one-sentence argument. "In the Lyon Diet Heart Study, a Mediterranean dietary pattern produced a 50–70% reduction in cardiovascular events in secondary prevention with no change in serum cholesterol — proving that LDL is at best a downstream marker of an underlying dietary and inflammatory pattern, not the proximate cause; and the statin trials, when their actual effect sizes are calibrated against this benchmark, deliver substantially less clinical benefit than the dietary intervention while exposing patients to drug-specific harm."⁴⁴

Evidence he leans on hardest. (1) Lyon Diet Heart Study (Circulation 1999). (2) The PREDIMED trial (NEJM 2013, retracted and republished 2018), which showed similar reductions on Mediterranean diet + extra-virgin olive oil or nuts.⁴⁵ (3) Reanalyses of statin secondary-prevention trials, particularly his methodological critiques of JUPITER and the discrepancies in some MERIT-HF and SEARCH adverse-event reporting.

Industry-funding posture. None. De Lorgeril has been a long-standing public critic of industry influence on cardiology guidelines.

Where he overreaches. (1) The Lyon study was open-label and small (n=605), and its very large effect size is itself a methodological caution — the trial may have overstated the dietary effect. (2) He sometimes conflates the absence of LDL change in Lyon with the absence of LDL causality, but a dietary intervention that reduces events without reducing LDL is consistent with LDL being a contributor among several causal factors, not necessarily disproof of LDL causality. (3) He has been less willing than other heterodox figures to acknowledge IMPROVE-IT, FOURIER, and CLEAR Outcomes as mechanism-independent confirmations of the LDL pathway.

16.3 Robert DuBroff — cardiologist, University of New Mexico

Bob DuBroff is among the most methodologically rigorous of the heterodox writers — a practising preventive cardiologist who has published in mainstream journals (BMJ Evidence-Based Medicine, World Journal of Cardiology, JACC, Mayo Clinic Proceedings) systematic critiques of the way trial endpoints are reported.⁴⁶

Strongest one-sentence argument. "When you systematically reanalyse the major statin and PCSK9-inhibitor trials in terms of absolute risk reduction rather than relative risk reduction, the average primary-prevention absolute risk reduction is around 1% over 5 years and the secondary-prevention number is 2–4% — magnitudes which, while real, do not match the rhetorical force of '25% reduction' and which raise legitimate questions about the cost-benefit profile, particularly in populations with predicted 10-year ASCVD risk under 10%."⁴⁷

Evidence he leans on hardest. His own 2020 BMJ Evidence-Based Medicine paper analysing 35 statin trials, showing that the median ARR was 1.3% for MI and 0.4% for stroke in primary prevention. He also pushes the (legitimate) critique that the LDL-cholesterol target moving (240 → 200 → 130 → 100 → 70 → 55 → 40) is suspicious as a scientific process.

Industry-funding posture. No industry funding disclosed in any of his major papers.

Where he overreaches. Less than the other heterodox figures. The strongest reasonable critique of DuBroff is that translating "small ARR" into "do not treat" requires a separate, value-laden judgment about acceptable number-needed-to-treat — and that for a cheap, generic, modestly-toxic drug like atorvastatin, an NNT of 60-100 over 5 years to prevent a non-fatal MI may well be acceptable, especially in secondary prevention where the consensus is uncontested.

16.4 David Diamond — University of South Florida, statistical critic

David Diamond is a neuroscientist (not a cardiologist) whose contribution to the dispute is statistical. He has published, with his Australian collaborator Uffe Ravnskov, several critiques arguing that the relative-risk-reduction framing in statin trials systematically misleads.⁴⁸

Strongest one-sentence argument. "Trial-level reporting that emphasises relative risk reduction while suppressing absolute risk reduction is a known cognitive distortion in medical communication, and when you correct for it, primary-prevention statin trials show effect sizes that would not survive cost-benefit scrutiny in many other therapeutic domains."⁴⁹

Industry-funding posture. None.

Where he overreaches. Diamond has occasionally crossed from statistical critique into broader claims about the lipid hypothesis itself, where his expertise is weaker. His writing also occasionally over-emphasises low-carbohydrate-diet advocacy in a way that the lipid evidence does not strictly require.

16.5 Aseem Malhotra — UK cardiologist, public-facing controversialist

Aseem Malhotra is the most publicly visible and the most controversial of the heterodox figures. He is a UK consultant cardiologist who has used mainstream media (BBC, Telegraph, Sky News Australia, GB News) to argue against statin overprescription, in favour of low-carbohydrate diets, and (since 2022) against COVID-19 mRNA vaccines. The latter position has substantially eroded his credibility outside the heterodox community.⁵⁰

Strongest one-sentence argument on statins. "The cumulative absolute risk reduction from statins in primary prevention is small, the side-effect profile is undercounted in the trial data because of run-in periods and selective adverse-event reporting, and the resources currently spent on universal lipid-lowering would deliver more population health if redirected to dietary and lifestyle interventions."⁵¹

Industry-funding posture. None disclosed for cholesterol-related advocacy. He has been criticised for commercial relationships with diet-book publishers and lifestyle-medicine commercial ventures, though these are distinct from pharmaceutical industry conflicts.

Where he overreaches. Substantially. (1) He has at times cited specific effect-size numbers (e.g., that statins extend life by "an average of four days") that are derived from selected meta-analytic slices and presented without appropriate uncertainty.⁵² (2) His COVID-vaccine advocacy, while not relevant to cholesterol per se, has demonstrated a pattern of claim-amplification that should make a careful reader cautious about his statin claims as well. (3) He has positioned himself in opposition to organised cardiology in ways that have at times been more rhetorical than evidentiary.

16.6 Malcolm Kendrick — UK GP, The Great Cholesterol Con

Malcolm Kendrick is a Scottish general practitioner whose book The Great Cholesterol Con (2007, updated multiple times) is the most widely read popular-level sceptical text after Ravnskov. He blogs prolifically.⁵³

Strongest one-sentence argument. "The lipid hypothesis was never the only candidate explanation for atherosclerosis, and the thrombogenic hypothesis (atherosclerosis as repeated injury-and-repair to the endothelium, with cholesterol incorporated secondarily into the clot/plaque) better explains why cholesterol-lowering produces inconsistent and population-specific results, why anticoagulants and antiplatelets are so effective, and why some classical risk factors (smoking, stress, social isolation) act through mechanisms that have little to do with LDL."

Industry-funding posture. None. Kendrick has been a public defendant in libel and professional-conduct proceedings that he won, including a UK Press regulator complaint against him for a media appearance.

Where he overreaches. (1) The thrombogenic hypothesis is not in fact an alternative to the lipid hypothesis — they are compatible; modern atherosclerosis biology incorporates both. (2) He has at times treated the FH literature dismissively, which is hard to sustain given the cleanness of the Mendelian evidence in homozygous and heterozygous FH. (3) Like Malhotra, he has been drawn into public controversies on adjacent topics (COVID-vaccine scepticism, 5G health concerns) that compromise his credibility outside his core subject.

16.7 Sherif Sultan — vascular surgeon, University of Galway

Sherif Sultan is an Irish-Egyptian vascular surgeon who has published, often with co-authors including Malhotra and Maryanne Demasi, critiques of LDL-lowering guidelines arguing that the evidence base for low LDL targets in primary prevention is weak and the harm potential under-recognised.⁵⁴

Strongest one-sentence argument. "Across 35 statin and PCSK9-inhibitor trials with LDL <70 mg/dL achieved targets, the relationship between achieved LDL and clinical benefit is inconsistent — many trials with very low achieved LDL show no mortality benefit — which is incompatible with the lower-is-better doctrine and consistent with the existence of an LDL threshold below which further reduction provides no clinical benefit."⁵⁵

Industry-funding posture. None disclosed.

Where he overreaches. The 2022 BMJ Evidence-Based Medicine paper he co-authored with Byrne, Demasi, and others (the "Curb our enthusiasm" paper) was sharply rebutted by the CTT group for, among other things, not weighting trials by sample size and not using IPD.⁵⁶ The substantive question — whether primary-prevention ARRs are large enough to justify universal screening-based treatment — remains live and was largely the heterodox camp's strongest contribution to the 2020s debate.

16.8 The heterodox camp, summarised

The heterodox camp's strongest collective contribution to the dispute is methodological: they have repeatedly pointed out (a) that relative-risk-reduction framing inflates perceived benefit; (b) that the LDL target has been moved repeatedly without clean trial-level justification at each step; (c) that primary-prevention ARRs are small in low-risk populations; (d) that the CTT IPD is not independently auditable; (e) that adverse-event reporting in industry trials has historically been incomplete (the ENHANCE delay, the Vioxx parallel). Their weakest collective move is to extrapolate from these legitimate methodological critiques into denial of the LDL causality hypothesis as a whole — which is hard to sustain against the Mendelian randomisation evidence reviewed in Chapter 18. The single most accurate one-line summary is: the heterodox camp is much stronger on absolute-benefit-in-primary-prevention than it is on LDL-causality-in-general. A reader who keeps this distinction in mind will derive substantial value from the heterodox literature without having to accept its strongest claims.

Chapter 17 — Methodological Critics Inside the Mainstream: the "In-Between" Voices

There is a distinct third group — researchers and editors who are not heterodox in the Ravnskov/Kendrick sense (they do not dispute the LDL-causality hypothesis) but who have been persistent, mainstream-institutional critics of how the lipid evidence is collected, reported, and translated into guidelines. These are the voices most worth taking seriously when calibrating a position, because they are inside the system and have professional reputations to protect.

17.1 John Ioannidis — Stanford, meta-research and statin-trial methodology

John Ioannidis is the founder of the discipline of meta-research and the author of "Why Most Published Research Findings Are False" (PLoS Medicine 2005), one of the most-cited papers in modern science.⁵⁷ His engagement with statins has been less about the LDL hypothesis (which he accepts) and more about the quality of evidence and the translation into practice.

Strongest one-sentence argument on cholesterol. "The statin evidence base in secondary prevention and very-high-risk primary prevention is robust, but the extrapolation to broad primary-prevention populations rests on extrapolations from underpowered subgroups, on adherence-corrected analyses with their own confounders, and on translating relative-risk-reduction into absolute terms in ways the original trials were not designed to support — and the resulting public-health recommendations therefore overreach what the evidence licenses."⁵⁸

Evidence he leans on hardest. His 2014 JAMA paper "More than a billion people taking statins?" arguing that under the 2013 ACC/AHA guidelines roughly 13 million additional Americans (and >1 billion globally) would become eligible, with much of the additional eligibility coming from groups in whom the absolute benefit is small. He has also been a co-author with researchers including Erin Michos and others on the methodology of cardiovascular risk prediction.

Disclosed industry relationships. Minimal. Ioannidis has consistently declined industry funding for his cholesterol-related meta-research.

Steelmanned position on primary prevention specifically. This is worth pulling out, because Anthony's debate will likely turn on primary-prevention indications. Ioannidis's position is approximately: (1) the relative effect of statins on cardiovascular events is real and generalises across risk strata; (2) but the absolute benefit shrinks proportionally as baseline risk falls; (3) the threshold below which the NNT becomes unfavourable depends on time horizon, side-effect rate, and patient values; (4) for populations with predicted 10-year ASCVD risk below ~10%, the trial evidence is weakest and the harm-benefit calculation is genuinely uncertain; (5) the 2013 ACC/AHA guidelines pushed the threshold to 7.5%, which exposed millions of additional individuals to a drug whose absolute benefit in their risk stratum is poorly characterised; (6) the 2024 AHA PREVENT equations — which downgraded predicted risk for roughly half of US adults — were in effect a correction of this overreach, though Ioannidis would emphasise that the original problem was the use of pooled cohort equations that systematically overestimated risk in modern populations.⁵⁹

This position is not heterodox in the Ravnskov sense. It is internal-methodological. It is also, in our view, the most defensible position for a non-specialist reader to start from.

17.2 Rita Redberg — UCSF cardiologist, JAMA Internal Medicine editor-in-chief

Rita Redberg edited JAMA Internal Medicine from 2009 to 2023 and used the journal as a platform for cardiovascular-overdiagnosis and overtreatment critiques.⁶⁰ During her editorship the journal published the Byrne et al. 2022 "Curb our enthusiasm" paper that triggered the most recent round of statin-trial debate.⁶¹

Strongest one-sentence argument. "The cardiovascular community has, over decades, drifted toward treatment intensification without commensurate attention to the quality-of-life implications of long-term polypharmacy, particularly in primary prevention and especially in the elderly, where competing risks of non-cardiovascular mortality and the psychological burden of chronic medication may exceed the cardiovascular benefit."

Evidence she leans on hardest. Editorial-level synthesis across the JAMA IM corpus: the over-screening literature, the polypharmacy-in-elderly literature, and the limited primary-prevention RCT evidence in adults over 75 (where, until STAREE and PREVENTABLE report in 2025–2026, there are essentially no robust randomised data).

Industry relationships. None disclosed.

Where she has moved. Redberg's position has been remarkably stable. She has been less harsh than Ioannidis on the underlying LDL evidence and more focused on the translation question — which has made her, in some ways, the most rhetorically effective in-system critic.

17.3 Maryanne Demasi — Australian science journalist, ex-ABC Catalyst

Maryanne Demasi is the journalist behind the 2013 ABC Catalyst two-part documentary "Heart of the Matter," which presented the heterodox case on Australian national television and triggered a regulatory backlash that ultimately ended the Catalyst program in its original form.⁶² She has since done her PhD, become an independent investigative journalist, and published methodological critiques (including the Byrne et al. 2022 BMJ EBM paper) in mainstream journals.

Strongest one-sentence argument. "The pharmaceutical industry and the academic cardiology establishment have, demonstrably, used their control over trial data and publication channels to shape the public understanding of statins in ways that exaggerate benefit and understate harm; the methodological corrections — adherence corrections, run-in periods, on-treatment vs intention-to-treat reporting, ghost-writing — collectively bias toward apparent benefit."

Industry-funding posture. None. She has been a consistent critic of industry-academic ties.

Where she overreaches. Her 2013 documentary was professionally vulnerable to the charge — eventually upheld by the ABC's internal review — that it presented heterodox claims without adequate balance. The documentary's content on side effects was specifically criticised as overstated; the documentary's broader question (about industry influence on trial interpretation) was less obviously wrong and was substantially the same question that the CTT IPD-access controversy later raised in the BMJ.⁶³ The 2024–2025 Demasi/Malhotra/Sultan body of work has remained methodologically aggressive in ways that some mainstream reviewers consider unfair; we treat it as live and contested.

17.4 Why the in-between voices matter most for Anthony's purposes

If the consensus camp's strongest move is to cite mechanism convergence and Mendelian randomisation, and the heterodox camp's strongest move is to point out small primary-prevention ARRs and industry data-control, the in-between camp's contribution is to grant most of the consensus claims while contesting the practice-level implications. This is the position most likely to survive contact with a well-informed friend at Roche. The in-between position concedes: - LDL is causal (the MR is too strong to dispute). - Statins reduce events in proportion to LDL reduction (the IPD is too strong to dispute). - Secondary prevention is uncontested (the NNTs are favourable). - Familial hypercholesterolaemia treatment is uncontested.

And contests: - The size of the primary-prevention benefit in low-risk populations. - The completeness of adverse-event reporting in pre-2010 trials. - The aggressiveness of LDL targets <55 in lower-risk strata. - The translation of trial evidence into population-screening recommendations.

This is the calibrated landing zone. Chapters 18, 19, and 20 will load further weight onto specific pieces of this scaffold.

Chapter 18 — Mendelian Randomisation as Adjudicator

18.1 What MR is and why the consensus camp treats it as their trump card

Mendelian randomisation (MR) is an epidemiological method that exploits the random assignment of genetic variants at conception to estimate the causal effect of a lifetime difference in an exposure on an outcome. The logic is structurally identical to a randomised controlled trial — the random allocation occurs at meiosis rather than at trial enrolment, and the "intervention" is a small lifelong difference in the exposure rather than a discrete pharmacological perturbation.⁶⁴

Applied to LDL and coronary heart disease, MR has four properties that the consensus camp finds decisive:

1. The randomisation is genuine. Genetic variants are assigned independently of behavioural confounders (diet, exercise, socioeconomic status), independently of reverse causation (CHD does not change one's genotype), and largely independently of other measured and unmeasured biological factors.
2. The exposure is lifetime, not acute. Where a statin trial measures the effect of 5–7 years of LDL lowering started in middle age, MR measures the effect of a small LDL difference present from gestation onward. If the causal effect of LDL on atherosclerosis is cumulative, MR should produce a larger per-mg/dL effect than an RCT — which is exactly what is observed.
3. The mechanism is independent. Multiple variants in multiple genes (PCSK9, HMGCR, NPC1L1, LDLR, APOE, APOB, LPA) each independently lower LDL, and each independently lower CHD risk by approximately the same per-mg/dL slope. This is structurally identical to the trial-level observation that statins, ezetimibe, PCSK9 inhibitors, and bempedoic acid each lower CHD by approximately the same per-mg/dL slope. The convergence of genetic and pharmacological mechanisms onto the same dose-response slope is, in the Bradford-Hill framework, the strongest available form of causal evidence.
4. The findings are reproducible. The PCSK9 finding — that loss-of-function carriers have markedly lower LDL and markedly lower CHD — has been replicated across multiple cohorts and multiple variants, with effect sizes that are tight enough to support drug-development decisions.⁶⁵

18.2 Cohen and Hobbs 2006 — the founding observation

The single most cited MR finding in the lipid literature is the 2006 NEJM paper by Helen Hobbs and Jonathan Cohen at UT Southwestern, identifying PCSK9 loss-of-function variants in African-American participants in the ARIC cohort.⁶⁶ Carriers of nonsense mutations in PCSK9 (Y142X or C679X, frequency ~2.6% in African-Americans) had a 28% lower LDL and an 88% lower risk of CHD at 15-year follow-up.

Three things made this paper transformative: - The effect size on CHD (88%) was much larger than what statin trials had produced for similar LDL reductions, but in the direction one would predict if cumulative lifetime exposure mattered. - The participants were otherwise healthy; the LDL reduction had no observed downside — no excess of any other disease, no neurocognitive effect, no fertility effect, no infectious-disease effect. - The biology of PCSK9 was understood well enough to immediately suggest a drug-development pathway. Within a decade evolocumab and alirocumab were in phase 3 trials.

This is the closest thing in cardiovascular epidemiology to a "natural knockout experiment" with a positive outcome.

18.3 The lifetime-exposure formalisation — Ference and the EAS consensus

Brian Ference's contribution was to formalise this into a quantitative framework. His 2012 paper, expanded in the 2017 EAS Consensus Statement (Ference, Catapano, Sabatine, et al., European Heart Journal 2017), tabulated the per-mg/dL effect of LDL-lowering variants across multiple genes (PCSK9, HMGCR, NPC1L1, LDLR, APOE, ATP-citrate lyase) against the per-mg/dL effect in matched RCTs.⁶⁷

The findings: - All variants produced essentially the same per-mg/dL coronary effect. - The MR effect was approximately three times the RCT effect per mg/dL. - The discrepancy was best explained by exposure duration: a 1-mg/dL lower LDL from birth has roughly three times the cumulative LDL-particle-years of a 1-mg/dL lower LDL achieved at age 60.

This framing — "LDL-particle-years" — is the single most important conceptual move in the consensus camp's recent rhetorical arsenal. It explains both why the RCT effect is real and why it is smaller than mechanistic intuition would predict.

18.4 Steelmanned MR critique from the heterodox camp

A fair-minded reader must consider the strongest version of the MR critique. There are four lines of attack worth taking seriously.

(1) MR estimates lifetime exposure, not pharmacological exposure. This is true, and it is also why the consensus camp uses MR cautiously. MR proves causality; it does not prove that a specific drug given at a specific dose at a specific age will achieve the predicted benefit. A statin started at age 60 cannot reproduce the cumulative effect of a PCSK9 LoF allele present from birth. The relevance of MR to clinical practice is therefore qualitative (LDL is causal, so lowering it should help) rather than quantitative (lowering LDL by X mg/dL at age Y will produce benefit Z).

(2) Pleiotropy and population stratification. MR's validity rests on the assumption that the genetic variant affects the outcome only through the exposure of interest. If PCSK9 variants affect CHD through some pathway other than LDL — say, immune function or vascular inflammation — then the MR estimate is biased. The honest answer is that pleiotropy cannot be excluded a priori, but the multi-variant convergence (PCSK9, HMGCR, NPC1L1, LDLR all giving similar per-mg/dL effects) is the strongest available counter-argument: if each variant has its own distinct pleiotropy, it would be a striking coincidence for them all to produce the same coronary effect per mg/dL of LDL.⁶⁸

(3) Effect-size discrepancy between MR and RCT. Some heterodox writers have argued that the larger MR effect compared to RCT effect indicates the genetic variants are tagging something other than LDL — that LDL is a proxy for the true causal factor. The consensus reply is that this is exactly backwards: the larger MR effect is predicted by the cumulative-exposure model, and if the genetic variants were tagging something other than LDL, the pharmacological lowering of LDL with chemically distinct mechanisms should not also produce coronary effects — which it does. The mechanism-convergence is the answer here.

(4) MR doesn't tell you about harm. This is the strongest of the four critiques and the one the consensus camp will concede. MR can tell you that lowering LDL prevents CHD; it cannot tell you whether a specific drug at a specific dose carries off-target adverse effects. The PCSK9 LoF carriers in Cohen-Hobbs had no observed harm, but they also had a gradual lifetime difference rather than a sudden pharmacological perturbation. For drug-specific harm questions (statin myalgia, statin-induced diabetes, PCSK9 neurocognitive effects), MR is silent. Trials are the right tool, and the trials need to be — and largely have been — explicitly designed to detect these effects.

18.5 Where MR/RCT convergence is genuine — and why it matters for Anthony's debate

When a friend at Roche pushes the "LDL is causal" line, the right response is not to dispute it (it is too well-established) but to be precise about what is and is not being claimed.

What MR + RCT convergence does establish: - LDL is causal in atherosclerosis. (Settled.) - Lower LDL achieved over longer durations produces more benefit. (Settled.) - The benefit is mechanism-independent. (Settled.) - Familial hypercholesterolaemia is correctly treated by lowering LDL. (Settled.) - Secondary prevention with LDL-lowering produces clinically meaningful absolute risk reduction. (Settled.)

What MR + RCT convergence does not establish: - That every adult above some LDL threshold should be treated pharmacologically. - That LDL <55 (vs <70 vs <100) produces clinically meaningful additional benefit at the population level. (The trial evidence here is weaker than the MR rhetoric implies — see Chapter 15 on Catapano/Mach guideline tightening.) - That the benefit-to-harm ratio for statins is favourable in low-risk primary prevention. (Still genuinely uncertain; STAREE and PREVENTABLE will help.) - That statin-attributed muscle, cognitive, and diabetes side effects are nocebo-only in clinical practice. (The CTT IPD analysis says yes; observational data say no; the truth is probably "mostly nocebo, with a small real signal".)

This is the calibrated position. MR is decisive on causality, not on practice.

Chapter 19 — Industry Funding, Ghost-Writing, and the Conflict-of-Interest Landscape

19.1 Why this chapter exists

A complete account of the cholesterol-statin dispute that did not name commercial actors and quantify their financial stakes would be incomplete to the point of distortion. The history of statins is also the history of one of the most lucrative drug classes in pharmaceutical history, and the credibility of every claim — pro and con — must be assessed in light of who paid for the evidence and who benefited from its public reception. This chapter names names and tallies amounts where they are publicly disclosed.

19.2 Lipitor and the rise of the blockbuster — Pfizer / Warner-Lambert / Parke-Davis

Atorvastatin (Lipitor) was developed by Bruce Roth at Parke-Davis (then a subsidiary of Warner-Lambert), patented in 1986, approved by the FDA in December 1996, and acquired by Pfizer through its 2000 purchase of Warner-Lambert for $90 billion (the second-largest pharmaceutical merger in history at the time, motivated largely by Lipitor's potential).⁶⁹

Lifetime sales of Lipitor between 1996 and the loss of US patent exclusivity in November 2011 are estimated at approximately $125–130 billion, making it the highest-grossing drug in pharmaceutical history.⁷⁰ At peak (2006), Lipitor produced revenues of $12.9 billion in a single year. This is the commercial context in which the consensus camp's evidence base was generated; it is also the commercial pressure that made statin-sceptic media coverage commercially threatening to a single Fortune 500 company.

Key academic-industry relationships in the Lipitor era included most of the principal trial investigators of the ASCOT-LLA, CARDS, TNT, IDEAL, and SPARCL trials. None of these trials produced negative or notably ambiguous results.

19.3 Crestor — AstraZeneca

Rosuvastatin (Crestor) was developed by Shionogi, licensed to AstraZeneca, and approved by the FDA in 2003. AstraZeneca made Crestor the centrepiece of its lipid franchise; peak sales reached $6.6 billion in 2011.⁷¹ The principal trial supporting Crestor's expansion into primary prevention was JUPITER (NEJM 2008), discussed under Paul Ridker in Chapter 15. Critics noted that JUPITER was stopped early for efficacy at 1.9 years (median), that the absolute risk reduction was small (NNT 95 for the primary composite at 2 years), and that Ridker held a patent on hsCRP testing — which JUPITER's design implicitly validated as a screening tool.⁷²

19.4 Vytorin, ezetimibe, and the ENHANCE delay scandal — Merck / Schering-Plough

This is the most consequential industry-conflict episode of the modern statin era and the one heterodox critics return to most often.

Vytorin is a fixed-dose combination of simvastatin (Merck) and ezetimibe (Schering-Plough). Ezetimibe (Zetia) was approved by the FDA in 2002 for LDL-lowering on the basis of a surrogate endpoint (LDL reduction); a clinical-outcomes trial was not required at approval.

The ENHANCE trial (Effect of Combination Ezetimibe and High-Dose Simvastatin vs Simvastatin Alone on the Atherosclerotic Process in Patients with Heterozygous Familial Hypercholesterolaemia) was an imaging study in 720 FH patients measuring carotid intima-media thickness — a structural surrogate for atherosclerosis progression. The trial enrolment finished in 2006; the data were expected to support Vytorin's claim that ezetimibe added clinical benefit beyond LDL reduction.

The trial reported in early 2008 — having been completed for nearly two years — that ezetimibe + simvastatin produced no greater reduction in carotid IMT than simvastatin alone, despite achieving lower LDL. Worse, the public release of the data was delayed by 18+ months while Merck and Schering-Plough were running peak commercial promotion of Vytorin. Congressional hearings (Henry Waxman, House Committee on Energy and Commerce) in early 2008 examined what the companies knew and when.⁷³

The fallout was substantial. Steven Nissen called for the resignations of the principal investigators publicly; the FDA reviewed its surrogate-endpoint approval pathway; the lipid community split sharply on whether ezetimibe added clinical value. The eventual outcomes trial, IMPROVE-IT (NEJM 2015), reported a modest but statistically significant reduction in MACE with ezetimibe added to simvastatin in post-ACS patients — ARR 2.0% over 7 years.⁷⁴ The consensus camp now treats ENHANCE as a surrogate-endpoint anomaly resolved by IMPROVE-IT; the heterodox camp treats it as a documented industry data-suppression episode that should colour interpretation of all subsequent industry-funded trials. Both readings are defensible.

19.5 Baycol — Bayer, withdrawn 2001

Cerivastatin (Baycol) was approved by the FDA in 1997 and became Bayer's lead statin. It was withdrawn worldwide in August 2001 after at least 31 confirmed rhabdomyolysis-related deaths in the United States and an additional 13 international fatalities, with hundreds of severe-rhabdomyolysis non-fatal cases.⁷⁵

Two facts about Baycol matter for the broader dispute. First, the rhabdomyolysis signal had been detectable in post-marketing surveillance for at least 18 months before withdrawal; Bayer's internal documents, released in subsequent litigation, suggested the company was aware of the risk earlier than its public statements indicated. Second, the rhabdomyolysis rate with cerivastatin was approximately 16–80 times higher than with the other statins on the market — meaning Baycol was a true outlier within the class, not a class-effect signal. The heterodox camp uses Baycol as the worst-case demonstration of industry harm-minimisation behaviour; the consensus camp uses it as evidence that the class as a whole has been safe by comparison and that pharmacovigilance ultimately worked.

19.6 Ghost-writing and the academic-publication economy

Ghost-writing — the practice of pharmaceutical companies commissioning medical-writing agencies to draft journal articles which are then published under named academic authors who may not have written or even reviewed the text — has been a documented feature of the statin literature. The most prominent academic critic of this practice has been David Healy (psychiatrist, author of Pharmageddon 2012 and multiple papers on industry-academic publication ethics).⁷⁶ Sergio Sismondo (philosopher of science, Queen's University) has produced the most systematic body of scholarship on the medical ghost-writing economy, including Ghost-Managed Medicine (Mattering Press, 2018).⁷⁷

Specific statin-era examples include: - The Pfizer Lipitor-promotion publication strategy documents released in the 2010 In re Neurontin litigation (though Neurontin/gabapentin, not Lipitor, was the legal subject — the documents incidentally revealed Pfizer's medical-publication infrastructure). - Merck-commissioned manuscripts in the Vioxx litigation (see §19.8 below) which were later shown to have been substantially drafted by company-commissioned writers, with named academic authors added late in the process. - Allegations — disputed — by some heterodox critics that aspects of the JUPITER and ASCOT-LLA primary publications received substantial industry medical-writing input.

The consensus camp's response is that ghost-writing, where it occurred, was a feature of the 1990s and 2000s publishing environment and that ICMJE authorship standards, mandatory disclosure, and trial-registration requirements (introduced post-2005) have largely closed the practice. The heterodox camp's response is that the historical record is sufficient to colour interpretation of any pre-2010 trial publication.

19.7 The CTT IPD-access controversy

The most consequential ongoing institutional dispute is over access to the CTT's individual-participant database. The CTT pools raw patient-level data from the major statin trials under data-sharing agreements with the original sponsors. The CTSU has refused, repeatedly, to release the data to outside researchers for independent reanalysis — citing both the underlying data-sharing agreements (which were negotiated when patient-level data sharing was not the norm) and patient-privacy considerations.

In 2014, Fiona Godlee (then editor-in-chief of The BMJ) and a series of heterodox commentators called for IPD release; Rory Collins and the CTSU declined.⁷⁸ The controversy became sharply personal during the BMJ statin-side-effects episode in 2013–2014, when Collins demanded — and ultimately received — partial retraction of two BMJ articles that had cited side-effect rates from non-RCT sources. The BMJ commissioned an external review panel which sided with Collins on the specific numerical claims but did not fully resolve the data-access question.⁷⁹

Two reasonable positions can be held on this. Consensus position: the CTT pools data under contracts that prohibit broader release; the CTT itself publishes its analyses in mainstream peer-reviewed venues; and the methodological capacity to conduct an alternative IPD meta-analysis exists in principle if individual sponsors release their data. Heterodox position: non-replicable analyses are non-scientific in the strong sense; the CTT's monopoly on IPD is itself a structural conflict of interest; the standard of evidence appropriate for a $50bn/year drug class should be at the maximum level of transparency. Both positions are defensible. As of 2025 the IPD has not been released for independent reanalysis.

19.8 The Vioxx parallel — Merck COX-2 inhibitor, withdrawn 2004

Vioxx (rofecoxib) is the cardiovascular-medicine community's canonical cautionary tale. Merck developed rofecoxib as a selective COX-2 inhibitor for the treatment of arthritis, with the marketing claim that it produced fewer GI bleeds than older NSAIDs. The drug was approved in 1999 and aggressively marketed.

In 2004, after the VIGOR trial and subsequent APPROVe trial demonstrated a doubling of myocardial infarction risk, Merck withdrew Vioxx worldwide. Internal documents released in subsequent litigation (the In re Vioxx Products Liability Litigation, US District Court Eastern District of Louisiana) showed that Merck had been aware of the cardiovascular signal at least four years before withdrawal and had used scientific-publication and academic-relationship strategies to minimise the perception of harm.⁸⁰ The estimated excess deaths attributable to Vioxx during the 1999–2004 period have been estimated by FDA scientist David Graham (Lancet 2005) at 26,000 to 56,000 in the United States alone.⁸¹

Three things matter about Vioxx for the statin dispute. 1. It happened in the same regulatory and publication environment as the major statin trials. 2. The harm signal was detectable in early data and was effectively delayed in public release through a combination of trial-design choices, post-hoc analytic decisions, and medical-publication strategies. 3. The companies involved overlapped with the lipid space (Merck specifically — same company that holds Zocor, Mevacor, Zetia, Vytorin).

The heterodox camp cites Vioxx as a proof-of-concept that industry data-handling can suppress harm signals for years; the consensus camp cites Vioxx as exactly the case that motivated post-2005 trial-registration and disclosure reform and argues that the post-Vioxx era is structurally different from the pre-Vioxx era. Both positions have merit.

19.9 ALLHAT-LLT — the NIH-funded counter-example

Among the major primary-prevention statin trials, one stands out as NIH-funded rather than industry-funded: the lipid-lowering arm of the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT-LLT).⁸² This was a substudy within the broader ALLHAT antihypertensive trial; 10,355 hypertensive patients with moderate hypercholesterolaemia were randomised to pravastatin 40mg or usual care.

The result, published in JAMA 2002: pravastatin produced a modest LDL reduction (16.7% in the pravastatin arm vs 7.5% in usual care) and a non-significant reduction in all-cause mortality (HR 0.99, 95% CI 0.89–1.11) and in CHD death plus non-fatal MI (HR 0.91, 95% CI 0.79–1.04). This was substantially less impressive than contemporaneous industry-funded statin trials.

Three readings of ALLHAT-LLT are possible. - Charitable reading: the trial used an open-label "usual care" comparator where ~30% of the usual-care arm ended up on statins by the end of follow-up, severely diluting the comparison. - Heterodox reading: the trial's negative findings reflect what statins actually do in real-world primary prevention when not run by industry-incentivised sites, and the dilution argument is itself a post-hoc rescue. - Consensus reading: the trial's design flaws make it an outlier and its findings should be subsumed into the broader CTT meta-analysis, where pravastatin trials (CARE, LIPID, WOSCOPS) showed clear benefit.

The honest answer is that ALLHAT-LLT is substantially weaker than the industry-funded trials, and that the dilution explanation is plausible but not fully verifiable, and that the existence of this contrast — NIH-funded trial less positive than industry-funded trials of similar agents — is a legitimate input to anyone's prior over industry-trial reliability.

19.10 The legitimate heterodox argument from this history, and the consensus's reasonable defence

The legitimate heterodox argument: across multiple drug classes (statins, COX-2 inhibitors, antidepressants, antipsychotics), industry-funded trials have systematically produced more favourable effect-size estimates than independent trials of the same class; industry has demonstrably (Baycol, Vioxx, ENHANCE) delayed or minimised the public release of harm signals; the academic-publication infrastructure has been compromised by ghost-writing and selective-authorship practices; and the CTT IPD monopoly closes the most obvious avenue for independent verification. A reasonable Bayesian should therefore down-weight industry-trial-derived effect estimates by some amount.

The consensus's reasonable defence: the trial-level effects converge across funding sources (industry trials, NIH trials, and the few investigator-initiated trials all give similar per-mg/dL slopes); the post-2005 reforms (trial registration, mandatory disclosure, IPD-sharing initiatives like Yale's YODA Project) have substantially improved transparency; and the mechanism-convergence with Mendelian randomisation — which is not industry-funded — provides an external check on the trial evidence. A reasonable Bayesian should therefore not down-weight the direction or qualitative shape of the trial findings, but might appropriately down-weight specific quantitative effect estimates from older, sole-sponsor industry trials.

Both positions are defensible. They lead to different views on the appropriate strength of low-risk-primary-prevention recommendations and to identical views on secondary prevention and FH. This is the calibrated landing zone.

Chapter 20 — Roche's Actual Lipid Footprint: Context for Anthony's Debate

20.1 Roche is not a statin manufacturer

This is the single most important fact to establish before Anthony's conversation with his Roche friend. Roche has never marketed a statin. Roche's commercial lipid franchise has been small, intermittent, and is currently dominated by diagnostics rather than therapeutics.

The most consequential Roche lipid-therapeutic episode was dalcetrapib, a cholesteryl ester transfer protein (CETP) inhibitor that Roche acquired from Japan Tobacco in 2004 and developed through phase 3 with the dal-OUTCOMES trial. Dalcetrapib raised HDL by approximately 30% without lowering LDL substantially. The trial, run by Greg Schwartz at the University of Colorado and others, randomised 15,871 patients with recent ACS to dalcetrapib or placebo.⁸³

In May 2012 Roche announced that dal-OUTCOMES had been stopped for futility at the recommendation of the independent data monitoring committee: no meaningful difference in cardiovascular events between dalcetrapib and placebo despite the substantial HDL increase. Roche terminated the dalcetrapib programme. Reported costs to Roche of the programme have been estimated in the $1.5–2 billion range (development plus phase 3 plus discontinuation charges).⁸⁴

The dalcetrapib failure was, in retrospect, the clearest refutation of the simple "raising HDL is cardiovascularly protective" hypothesis. Combined with the failures of torcetrapib (Pfizer, 2006), evacetrapib (Lilly, 2015), and the partially positive but commercially unconvincing anacetrapib (Merck, 2017), the CETP inhibitor class is now widely considered a confirmed disappointment.⁸⁵ The interesting consequence is that the CETP-inhibitor era strengthened the LDL-focused consensus by ruling out HDL as a primary therapeutic target — HDL came to be reframed as a marker rather than a mover.

After dal-OUTCOMES, Roche essentially withdrew from lipid therapeutics. There is no Roche-marketed statin, PCSK9 inhibitor, ezetimibe-equivalent, bempedoic-acid-equivalent, ANGPTL3 inhibitor, or Lp(a) therapy. Roche's pipeline as of 2025 contains no late-stage lipid-lowering therapeutic.

20.2 What Roche does sell in the lipid space — diagnostics

Roche's current lipid-space revenue comes almost entirely from diagnostics: - Cobas lipid panel — the routine clinical-chemistry assays for total cholesterol, HDL-C, LDL-C, triglycerides used in tens of thousands of laboratories worldwide. Volume is enormous; per-test margin is modest. - Cobas Lp(a) assay — a specific lipoprotein(a) assay using a particle-size-independent methodology, which is positioned to become a companion diagnostic for the emerging Lp(a)-lowering therapeutics (pelacarsen from Novartis/Ionis, olpasiran from Amgen, lepodisiran from Lilly).⁸⁶ - Cobas ApoB and ApoA1 — apolipoprotein assays which are increasingly recommended by guideline panels (particularly the ESC/EAS) as preferred risk markers over LDL-C in some clinical contexts (e.g., metabolic syndrome, mixed dyslipidaemia). - Companion diagnostics for PCSK9 and Lp(a) therapeutics — Roche Diagnostics has positioned itself to be the lab-side partner for the next wave of lipid therapeutics rather than the drug-side partner.

This is a commercially material business — Roche Diagnostics had ~CHF 14.4 billion in revenue in 2023 — but it is dwarfed by Roche's pharmaceutical division, which is dominated by oncology (Genentech-originated franchises in HER2-targeted breast cancer, lymphoma, ophthalmology) and increasingly by neuroscience (Ocrevus, Evrysdi). Total Roche group revenue 2024: ~CHF 60.5 billion, of which pharmaceuticals ~CHF 46 billion and diagnostics ~CHF 14.5 billion.⁸⁷

20.3 Why this matters for Anthony's debate

Three implications follow.

(1) Anthony's friend at Roche is not a statin advocate per se. Whatever views the friend holds on cholesterol and statins, they are not commercially aligned with the statin franchises (Pfizer, AstraZeneca, Merck, BMS, Novartis-Sandoz on generics) or with the PCSK9 franchises (Amgen, Sanofi/Regeneron, Novartis-via-inclisiran). The friend may hold strong personal views, may have done postdoctoral work in lipid biology, may simply be repeating positions that are the cardiology mainstream — but they are not paid to defend statins. This changes the appropriate epistemic register of the conversation. It is not a customer/supplier conversation; it is two scientifically interested individuals reasoning about evidence.

(2) Roche's commercial alignment is with accurate diagnostics. This dovetails closely with the consensus camp's "measure LDL (and now ApoB and Lp(a)), treat appropriately" view — but it dovetails equally well with the in-between camp's "stratify by absolute risk and treat the high-risk patients well rather than treating the population indiscriminately" view. Roche's diagnostic business benefits from sophisticated lipid management; it does not particularly benefit from primary-prevention statin overprescription, which can be done without sophisticated diagnostic input. If anything, Roche has a commercial interest in more nuanced lipid assessment (Lp(a), ApoB, particle-number) rather than more uniform pharmacological treatment.

(3) The COI dynamics are fundamentally different from talking to someone at Pfizer or Amgen. If Anthony were debating a Pfizer scientist, the conversation would be conducted in the shadow of Lipitor (peak revenue $13bn/year, lifetime $125bn). If he were debating an Amgen scientist, the conversation would be conducted in the shadow of evolocumab. With a Roche scientist, the conversation is conducted in the shadow of: oncology (most of Roche's revenue), neuroscience (where Roche is investing), diagnostics (the relevant business unit for cholesterol), and a 13-year-old failed CETP programme. Roche has, in effect, lost its lipid-therapeutic bet — which makes its current institutional position considerably more disinterested than that of any of the statin makers.

20.4 How to calibrate the conversation

A Roche scientist will likely: - Accept LDL causality (mainstream cardiology view). - Accept the secondary-prevention statin evidence (mainstream view). - Be sophisticated about the next-wave therapeutics (PCSK9 inhibitors, Lp(a) lowering) because Roche's diagnostics business is positioned downstream of these. - Be either neutral or modestly sceptical of the most aggressive primary-prevention positions (because Roche has no commercial stake in mass-market statin prescription). - Be specifically interested in who is being measured and how, because that's where Roche's commercial position lies.

If Anthony's friend pushes any of these positions, the calibrated response is to grant the points consistent with the in-between position (LDL causality, secondary prevention, FH, mechanism convergence) and to push back precisely on the points where the evidence remains genuinely contested (low-risk primary prevention, the rate of real-as-opposed-to-nocebo statin side effects, the IPD-access transparency question, the appropriate aggressiveness of LDL targets below 70 mg/dL in low-risk strata, and the appropriate handling of the elderly population pending STAREE and PREVENTABLE).

The conversation is winnable on the calibrated middle ground. It is not winnable from a hard-heterodox foundational-scepticism stance, and it does not need to be — the in-between position is both intellectually stronger and more rhetorically effective with a scientifically literate interlocutor.

Closing of Part III

The dispute over cholesterol and statins is not a fight between truth-tellers and shills. It is a structured disagreement among scientifically serious people about the appropriate translation of a genuinely complex evidence base into clinical practice. The consensus camp is right about LDL causality, about secondary prevention, about the qualitative direction of the effect, and about Mendelian randomisation as an adjudicator of causal questions. The heterodox camp is right that primary-prevention absolute risk reductions are small, that industry has historically managed harm signals badly, that the CTT IPD opacity is a problem, and that the LDL-target ratchet is not seamlessly grounded in trial-level evidence at every step. The in-between camp — Ioannidis, Redberg, the Demasi/Byrne 2022 paper at its best — articulates the position most worth defending.

Roche, the institution that frames Anthony's specific conversation, is the quietest major lipid-space actor: a failed therapeutic programme, a successful diagnostics franchise, an oncology-dominated balance sheet. The conversation Anthony will have is therefore better thought of as a conversation about evidence quality than as a conversation about commercial interest. Calibrated well, that is the conversation he wants to have.

Footnotes

Part IV — The harms ledger

Prefatory note to Part IV. Anyone wishing to debate cholesterol and statins substantively has to spend more time on the harms ledger than on the efficacy ledger, because almost every populist objection to statin therapy is, at root, a claim about harm rather than about efficacy. The Mendelian-randomisation evidence that low-density-lipoprotein cholesterol (LDL-C) is causally linked to atherosclerotic cardiovascular disease (ASCVD) is now sufficiently overdetermined that even most heterodox cardiologists concede it in print; the live argument has migrated to whether the pharmacological lowering of LDL-C with HMG-CoA reductase inhibitors carries an acceptable safety profile in the populations to whom it is prescribed.¹ Part IV is therefore long, granular and deliberately ecumenical. Each chapter is organised the same way: (i) the headline claim of harm; (ii) the best randomised and pharmacovigilance evidence; (iii) the steelman of the orthodox interpretation; (iv) the steelman of the heterodox interpretation; (v) my adjudication; (vi) the clinical implication. Readers who want the bottom line first should skip to the adjudication paragraphs; readers who want to debate a Roche scientist substantively should not.

Chapter 21 — Statin-associated muscle symptoms (SAMS)

21.1 The single most important number in this debate

If you ask a primary-care physician in the United Kingdom, the United States or Australia what the most common reason for statin discontinuation is, the answer will be muscle symptoms. If you ask a cardiologist who works in lipid clinics what proportion of patients referred for "statin intolerance" turn out to have a pharmacologically attributable myopathy on careful rechallenge, the answer will be a small minority. Both clinicians are telling the truth, and the gap between their two truths is the central battleground of the harms debate.²

The headline numbers, repeated in every consensus statement on the topic, are these. In observational cohorts and registries (USAGE, PRIMO, the Understanding Statin Use in America and Gaps in Patient Education survey, the various national pharmacovigilance databases), the reported prevalence of muscle symptoms in statin-treated patients is between 10 % and 30 %, depending on the population, the statin and the definition used.³ In the blinded period of large randomised trials, the excess of muscle complaints in the statin arm compared with placebo is between 0 % and roughly 2 %, with most well-conducted trials clustering between 0.5 % and 1.5 %.⁴ This is an order-of-magnitude discrepancy. It cannot be reconciled by saying that trial populations are healthier than real-world populations, because the trial-versus-placebo comparison is internal to the trial and therefore controls for baseline frailty. It must, on the face of it, reflect either (a) a massive nocebo effect in unblinded clinical practice, (b) a major under-ascertainment of muscle complaints inside trials, or (c) both. The interpretation one places on this gap drives almost every subsequent argument about statins.

21.2 SAMSON: the elegant N-of-1 experiment

In 2020 Howard and colleagues at Imperial College London published the Self-Assessment Method for Statin Side-effects Or Nocebo (SAMSON) study in the New England Journal of Medicine.⁵ They recruited 60 patients who had discontinued statin therapy within the previous month because of side-effects. Each patient was given twelve identical opaque bottles, randomised to four months of atorvastatin 20 mg, four months of placebo and four months of no tablets, in a balanced random order. Patients recorded daily symptom intensity on a 100-point visual analogue scale using a smartphone application. The primary outcome was the "nocebo ratio": the proportion of the symptom intensity on tablets that could not be attributed to the active drug because it was also present on placebo.

The headline result was that the symptom intensity on statin (mean 16.3 out of 100) was indistinguishable from that on placebo (mean 15.4 out of 100), while both were substantially higher than on no tablets (mean 8.0 out of 100). The nocebo ratio was 0.90 — that is, 90 % of the symptom intensity reported on the statin was also reported on placebo, with only 10 % attributable to a specific drug effect.⁵ Of the 60 patients, 30 successfully restarted statin therapy six months after the trial; another 4 had planned to. The conclusion the authors drew was carefully bounded: in patients who had previously discontinued a statin because of symptoms, most but not all of those symptoms recur on placebo, and a substantial fraction of such patients can resume therapy when shown their own data.

SAMSON sits inside a broader tradition of N-of-1 nocebo demonstrations in pain medicine and in pharmacology more generally. Open-label placebo trials in irritable bowel syndrome, chronic low back pain and migraine have shown clinically meaningful symptom modulation by inert tablets, even when patients are told the tablets are inert.⁶ The phenomenon is not specific to statins; what SAMSON demonstrated is that it operates in the statin domain at a magnitude that swamps the pharmacological signal.

21.3 STOMP, GAUSS and the search for a biochemical phenotype

The most rigorous attempt to detect a biochemical signature of statin myopathy in unselected patients was the Effect of Statins On skeletal Muscle Function and Performance (STOMP) trial published by Parker, Capizzi and Thompson in Circulation in 2013.⁷ STOMP randomised 420 healthy statin-naïve adults to atorvastatin 80 mg/day or placebo for six months, with serial measurements of creatine kinase (CK), maximal voluntary contraction strength, exercise-induced muscle injury and patient-reported myalgia. Two findings stood out. First, mean CK was not elevated in the statin arm; the well-known dose-related transient CK rise of older studies was not reproduced at the population level. Second, myalgia incidence was 9.4 % on atorvastatin versus 4.6 % on placebo (P = 0.05) — a real but modest excess, perfectly consistent with the trial-level signal of 1–2 % absolute excess in larger cardiovascular endpoint trials.⁷

STOMP is one of the few studies designed specifically to look for the muscle signal rather than to power a cardiovascular outcome, and it was conducted by investigators (notably Paul Thompson) who have been on record sceptically about extreme claims in both directions. Its result is best read as a calibration: there is a real, small pharmacological signal; it is not as large as observational reports suggest; it is not as trivial as some trial-level secondary analyses suggest.

The Goal Achievement After Utilizing an Anti-PCSK9 Antibody in Statin Intolerant Subjects (GAUSS) trials (GAUSS, GAUSS-2, GAUSS-3) took a different tack: they enrolled patients who described themselves as statin-intolerant and rechallenged them with atorvastatin in a double-blind crossover. GAUSS-3, published by Nissen and colleagues in 2016 in JAMA, randomised 491 patients with a history of intolerance to two or more statins to atorvastatin 20 mg versus placebo for ten weeks each, in random order.⁸ Forty-three percent of patients reported myalgia on atorvastatin only; 27 % reported myalgia on placebo only; and 10 % reported myalgia on both. That is, roughly half of all symptom episodes in self-described intolerant patients were not pharmacologically specific — but the other half were. GAUSS-3 is, on the heterodox reading, the cleanest demonstration that statin-attributable muscle symptoms exist as a real phenomenon in a real-world phenotype, even after one strips out the nocebo component.

21.4 SAMS-CoQ10, COQuet, and the supplement question

Patients who experience statin-attributed muscle symptoms frequently take coenzyme Q10 (ubiquinone) supplementation on the theory that statins inhibit the mevalonate pathway upstream of CoQ10 synthesis as well as cholesterol synthesis. The pharmacological basis is real: HMG-CoA reductase produces mevalonate, which is the substrate for both sterol biosynthesis and the side-chain isoprenoids that include CoQ10, dolichol and the prenyl groups on small GTPases.⁹ Statins demonstrably reduce circulating CoQ10 by roughly 20–40 %, although the extent to which this depletes intramuscular CoQ10 is contested.¹⁰

The clinical question — does CoQ10 supplementation relieve statin-associated muscle symptoms? — has been examined in at least three randomised trials. The SAMS-CoQ10 trial (Taylor and colleagues, Atherosclerosis 2015) randomised 41 patients with a history of statin myalgia to CoQ10 600 mg/day or placebo for eight weeks while rechallenged on simvastatin 20 mg; there was no significant difference in myalgia incidence or severity.¹¹ An earlier randomised trial by Caso and colleagues (2007) had been positive; a 2015 systematic review concluded that across six small trials the pooled effect was indeterminate, with substantial heterogeneity and high risk of bias.¹² The honest summary is that CoQ10 supplementation may help individual patients, is essentially harmless, but does not have the trial-level evidence base to be recommended routinely.

21.5 STAREE-HEART and the cardiac-aging hypothesis

A more interesting recent angle on the muscle question is whether statins affect cardiac rather than skeletal muscle — specifically whether long-term high-intensity statin exposure accelerates a biomarker-defined "cardiac aging" phenotype. The STAREE-HEART substudy of the Australian STAREE primary-prevention trial in adults aged ≥70 is examining this exact question, with co-primary biomarker endpoints around cardiac troponin, N-terminal pro–B-type natriuretic peptide (NT-proBNP) and echocardiographic strain.¹³ Results are not yet available at the time of writing (mid-2026); STAREE itself is anticipated to read out shortly.¹³¹⁴ The hypothesis being tested is biologically plausible — myocardial mitochondrial function may be more dependent on the mevalonate pathway than skeletal muscle, and the elderly are the population in whom any such signal would matter most — but at present STAREE-HEART is a frontier rather than a fact. It is worth flagging for the Roche conversation precisely because it is one of the few places where new mechanistic data on statin safety in the at-risk elderly will arrive in the next 18 months.

21.6 Biological mechanisms: CoQ10, prenylation, mitochondrial dysfunction

If statin-attributable myalgia is real (and the trial-level data say it is, even if smaller than observational reports), what is the mechanism? Three candidate pathways have generated the bulk of the mechanistic literature.

The first is CoQ10 depletion. As above, statins reduce systemic CoQ10 levels, and CoQ10 is essential to mitochondrial electron transport at complex III. Langsjoen and Langsjoen, working from a cardiology practice with a long-standing interest in CoQ10, have argued for a primary causal role; Marcoff and Thompson, in a Journal of the American College of Cardiology review, concluded that the mechanistic case was plausible but the trial-level case for supplementation was weak.¹⁵¹⁶ The current centre-of-gravity view is that CoQ10 depletion is a contributing mechanism in susceptible individuals, not the dominant population-level explanation.

The second is isoprenoid depletion and impaired prenylation of small GTPases. Mevalonate-pathway intermediates (farnesyl pyrophosphate, geranylgeranyl pyrophosphate) are required to post-translationally modify Rho, Rac and Cdc42 family GTPases that regulate myocyte cytoskeleton, autophagy and apoptosis. Statin-induced impairment of prenylation has been shown in muscle biopsies to correlate with myalgia severity in some series.¹⁷ This is the mechanism most often invoked in basic-science explanations of why patients with chronic statin exposure occasionally develop a frank inflammatory or necrotising myopathy — including the rare immune-mediated necrotising myopathy associated with anti-HMGCR antibodies, which is itself a separate clinical entity.¹⁸

The third is mitochondrial dysfunction more broadly — reduced respiratory chain efficiency, altered fission/fusion dynamics, accumulation of oxidative damage. The strongest evidence is from muscle-biopsy studies in symptomatic patients showing reduced succinate dehydrogenase activity and ragged red fibres, although this is far from universal.¹⁹

For the steelman of orthodoxy: each of these mechanisms is plausible in vitro and in vivo in animal models, but their translation to a population-level adverse-effect signal in humans is weak. For the steelman of heterodoxy: the mechanisms exist, are biologically coherent, would predict a long-tail distribution of susceptibility (so most patients are unaffected and a minority are severely affected), and explain the observed clinical phenotype better than a pure nocebo hypothesis. Both readings are defensible. Neither is a knockout blow.

21.7 Steelmanning both sides on SAMSON

The orthodox steelman of SAMSON is that it is small but methodologically elegant. The N-of-1 crossover design controls within-patient for everything that could plausibly confound the comparison: age, comorbidity, baseline pain, expectation, season, life-event load. The use of identical opaque bottles and electronic daily diaries minimises measurement error. The replication of the well-characterised nocebo phenomenon — known from pain medicine, headache, irritable bowel and gastrointestinal disorders — gives an external-validity scaffolding. The result that 90 % of symptom intensity is non-specific is consistent with what one would predict from the broader nocebo literature and is consistent with the 1–2 % trial-level excess in blinded outcome trials.

The heterodox steelman of SAMSON's limitations is also serious. Sixty patients is a small sample, particularly when one considers that the relevant clinical question is not "what is the mean nocebo ratio" but "what is the distribution of pharmacological susceptibility". A 90 % nocebo ratio at the mean is fully consistent with a small minority of patients (say 5–10 %) having pharmacologically specific severe symptoms that the trial is underpowered to detect as a separate cluster. All 60 participants were already self-identified as statin-intolerant, which is the appropriate target population for asking can these patients re-tolerate, but does not capture (i) the patient who develops late-onset symptoms after years of tolerated therapy, (ii) the patient with frank rhabdomyolysis or immune-mediated necrotising myopathy, (iii) the patient with progressive proximal weakness without prominent pain. SAMSON answers the right question for the largest cohort of statin-intolerant patients in practice, but it does not retire the broader clinical problem.

21.8 Adjudication

My reading of the evidence, after working through it many times: the gap between 10–30 % observational and 1–2 % blinded trial reporting is best explained as a combination of (a) a large nocebo component, probably accounting for the majority of complaints in routine practice, (b) a small but real pharmacological signal that the trials do detect, accounting for roughly 1–2 % absolute excess, and (c) under-ascertainment of mild symptoms in trials because trial CRFs are not optimised for capturing them. The honest summary is that most muscle complaints in routine practice are not specifically pharmacological, but a real minority are; that severe events (rhabdomyolysis, anti-HMGCR-associated necrotising myopathy) are rare but real and warrant clinical respect; and that the clinical implication is to attempt rechallenge with a different statin at a lower dose, or every-other-day dosing, before abandoning therapy in any patient who has a genuine indication.

21.9 Clinical implications for the rechallenge conversation

The 2022 American College of Cardiology Expert Consensus Decision Pathway on the Management of ASCVD Risk Reduction in Patients with Persistent Hypertriglyceridemia and the 2023 European Atherosclerosis Society/National Lipid Association consensus statements have converged on a structured rechallenge approach: (i) wash out for at least two weeks; (ii) restart at a low dose of a hydrophilic statin (rosuvastatin 5 mg or pravastatin 20 mg) rather than a lipophilic statin; (iii) trial every-other-day or twice-weekly dosing if daily is not tolerated; (iv) document symptoms with a structured tool; (v) escalate slowly; (vi) consider non-statin therapy (bempedoic acid, ezetimibe, PCSK9-inhibitors) only after a documented genuine intolerance.²⁰ The CLEAR Outcomes trial of bempedoic acid has changed the practical landscape here: there is now a non-statin oral agent with a positive cardiovascular outcomes trial in genuinely statin-intolerant patients, and this has shifted the bar for what counts as "intolerance has been established".²¹

Chapter 22 — New-onset diabetes mellitus

22.1 Discovery and quantification

The diabetes signal was not detected in any of the foundational statin trials of the 1990s — 4S, WOSCOPS, CARE, LIPID, AFCAPS/TexCAPS. It first emerged unambiguously in 2008 from the Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER), in which Ridker and colleagues reported a 25 % relative increase in physician-reported diabetes incidence in the rosuvastatin arm (270 versus 216 cases, P = 0.01).²² A post-hoc analysis by Ridker himself in 2012 confirmed the signal at a 27 % relative-risk increase and decomposed it by baseline diabetes risk: the effect was concentrated in patients with one or more pre-existing diabetes risk factors (impaired fasting glucose, metabolic syndrome, obesity, hypertension).²³ In patients without any such risk factor, JUPITER did not see a diabetes excess.

The pivotal meta-analysis is Sattar and colleagues, The Lancet 2010, which pooled individual-participant data from 13 statin trials totalling 91,140 participants with a mean follow-up of four years. The pooled odds ratio for incident diabetes was 1.09 (95 % CI 1.02–1.17), corresponding to a 9 % relative risk increase. There was no significant heterogeneity by statin type, dose, sex or baseline LDL-C.²⁴ Subsequent dose-effect analyses (Preiss and colleagues, JAMA 2011) confirmed a dose-response relationship: intensive statin therapy produced a 12 % higher diabetes incidence than moderate-intensity therapy.²⁵ These findings have been replicated in nearly every subsequent analysis, including the 2024 Cholesterol Treatment Trialists' (CTT) updated safety meta-analysis.

22.2 Is it real, and how big is "real"?

The diabetes signal is one of the few statin adverse-effect claims where there is essentially no live disagreement between orthodox and heterodox commentators about the existence of the effect. It is real, it is reproducible, it is dose-dependent, and it has converging support from three independent lines of evidence: (i) randomised trials, (ii) pharmacoepidemiology (the WHI observational cohort, the Finnish national diabetes registry), and (iii) Mendelian randomisation.²⁴²⁶²⁷

The Mendelian-randomisation evidence is particularly telling because it isolates the pharmacological target. Swerdlow and colleagues, in The Lancet in 2015, used common genetic variants in HMGCR that mimic the effect of statins to ask whether genetically lower LDL-C via HMGCR inhibition was associated with higher diabetes risk in roughly 200,000 individuals. The answer was yes: each genetic standard-deviation lower LDL-C via HMGCR was associated with a body-weight increase of approximately 300 grams and an odds ratio for type 2 diabetes of 1.12, almost exactly matching the trial-level effect estimate.²⁷ This is the cleanest possible test of whether the diabetes signal reflects the on-target pharmacology of HMG-CoA reductase inhibition rather than off-target effects of any particular statin molecule. The result is that it does. The mechanism by which HMG-CoA reductase inhibition impairs glucose homeostasis is incompletely understood — proposed mechanisms include reduced isoprenoid-dependent insulin signalling, calcium channel effects on β-cell insulin release, and modest weight gain — but the on-target nature of the effect is no longer in serious dispute.²⁸

22.3 NNH versus NNT: the net-benefit calculation

The clinically relevant question is not whether the diabetes signal exists but whether it materially shifts the net-benefit calculation. The Sattar meta-analysis estimated that statin therapy causes approximately one extra case of diabetes per 255 patient-years; the absolute risk increase over four years was 0.4 %, with an NNH of 255 person-years.²⁴ The CTT analyses estimate that in patients at moderate-to-high cardiovascular risk, the NNT for prevention of a major vascular event over the same horizon is in the range of 50 (secondary prevention) to 200 (primary prevention).²⁹ In high-risk patients, therefore, the benefit-to-harm ratio is favourable: the major vascular event prevented is typically a more morbid outcome than the diabetes case caused, and the diabetes case may have been deferred rather than wholly created (i.e. the statin may accelerate the diagnosis of diabetes by months to a few years in a patient already on the trajectory).

The picture inverts at the low-risk end. In a primary-prevention patient with a 10-year ASCVD risk below roughly 5 %, the NNT for cardiovascular benefit lengthens substantially while the NNH for diabetes remains roughly constant. At sufficiently low baseline cardiovascular risk and sufficient baseline diabetes risk (impaired fasting glucose, central obesity, family history), the net expected utility of statin therapy may be neutral or negative. This is one of the strongest legitimate arguments against blanket primary-prevention prescribing in low-risk populations, and it is the heterodox argument that orthodox lipidologists most readily concede when the patient profile is correctly stipulated.³⁰

22.4 Effect concentrated in the prediabetic phenotype

The diabetes signal is not uniformly distributed across the trial population. In the JUPITER post-hoc analysis, patients with at least one diabetes risk factor accounted for essentially all of the absolute excess; patients with no diabetes risk factors had a small numerical excess that was not statistically significant.²³ In WOSCOPS, baseline statin use was associated with a reduction in diabetes incidence, almost certainly reflecting the fact that pravastatin is a hydrophilic statin with the weakest diabetes signal in head-to-head comparisons. The CTT IPD meta-analyses have shown that the dose-response within the statin class is steepest for atorvastatin 80 mg and rosuvastatin 40 mg, intermediate for moderate-intensity therapy, and weakest for pravastatin and pitavastatin.²⁹³¹

The clinical implication is that the diabetes risk can be partly mitigated by statin selection. In a patient at high cardiovascular risk and high diabetes risk, a moderate-intensity hydrophilic statin (pravastatin 40 mg) is a defensible choice; in a patient at very high cardiovascular risk and moderate diabetes risk, a higher-intensity statin remains appropriate but with active glycaemic monitoring; in a low-risk primary-prevention patient with strong diabetes risk factors, the case for any statin at all weakens.

22.5 Steelmanning both sides

The orthodox steelman is that the diabetes effect is real, modest, on-target, concentrated in the prediabetic phenotype, and net-favourable in any population at appreciable cardiovascular risk. The cardiovascular events prevented are typically more morbid than the diabetes cases caused; the diabetes diagnosis triggers earlier glycaemic intervention that itself reduces cardiovascular risk; and the absolute numbers are small enough that they do not overturn a well-indicated prescription.

The heterodox steelman is that the diabetes signal is real, is being systematically downplayed in patient-facing communication, is the strongest empirical argument that statins are not metabolically benign, and should reframe the conversation in low-risk primary prevention. A patient told "the statin will reduce your heart attack risk by 25 %" should also be told "and will increase your diabetes risk by 9 %, which in your case means roughly equal absolute numbers". The patient autonomy and informed-consent dimension is real.

22.6 Adjudication

The diabetes signal exists, is on-target, is dose-dependent, and is concentrated in patients with pre-existing diabetes risk. In high-risk patients (secondary prevention, FH, diabetes already established, very high primary-prevention risk), the net-benefit calculation favours statin therapy unambiguously. In low-risk primary-prevention patients with strong pre-diabetic phenotype, the net-benefit calculation is genuinely close, and individual patient values about which outcome they wish to defer should carry weight. The 2024 AHA PREVENT equation, by reclassifying a substantial fraction of US adults to lower 10-year ASCVD risk, will (correctly, in my reading) push the borderline cases toward non-pharmacological intervention first.³²

Chapter 23 — Cognitive complaints

23.1 The 2012 FDA label change

In February 2012 the U.S. Food and Drug Administration issued a class-wide labelling change for statins, adding to the precautions section a statement that "ill-defined memory loss and confusion have been reported with statin use" but "these reported events are generally not serious and went away once the statin was stopped". The change was prompted by FDA Adverse Event Reporting System (FAERS) signals and a small number of high-profile case reports, including the well-publicised series by Duane Graveline (a former NASA flight surgeon whose self-reported transient global amnesia on atorvastatin became the seed of a substantial heterodox literature).³³

The label change was, methodologically, a low bar to clear: it required only that a credible pharmacovigilance signal exist and that the agency judge the warning useful to clinicians. It was not a finding that statins cause cognitive impairment at the population level. But it was widely reported in the lay press as if it were, and it remains one of the most-cited single pieces of evidence in popular anti-statin commentary.³⁴

23.2 The randomised-trial signal: essentially negative

When one turns from pharmacovigilance to randomised data, the picture changes substantially. The Heart Protection Study (HPS, 20,536 participants, simvastatin 40 mg vs placebo) included serial cognitive assessment via the modified Telephone Interview for Cognitive Status (TICS-m) at five years. The simvastatin and placebo arms were statistically indistinguishable.³⁵ JUPITER's prespecified cognitive analyses found no decline. PROSPER (Prospective Study of Pravastatin in the Elderly at Risk), a trial specifically in the 70–82-year age group of 5,804 participants, included multiple cognitive assessments (Mini-Mental State Examination, Stroop, Letter-Digit Coding, Picture-Word Learning) at baseline and at the end of follow-up; there was no statistically significant difference between pravastatin and placebo arms on any measure.³⁶

The 2025 long-term follow-up of FOURIER-OLE — the open-label extension of the evolocumab outcome trial — has now reported cognitive safety data through 7.2 years of cumulative exposure with no decline relative to expected age-related trajectories.³⁷ FOURIER-OLE is not a statin trial, but it is a long-term aggressive LDL-lowering trial (median achieved LDL-C around 30 mg/dL), and it is therefore the strongest available test of whether very low LDL-C per se causes cognitive harm. The answer it returns is no.³⁷

The HOPE-3 trial (rosuvastatin 10 mg in 12,705 intermediate-risk participants) and the EBBINGHAUS substudy of FOURIER (1,204 patients with neurocognitive testing every six months) round out the picture. EBBINGHAUS, in particular, used the Cambridge Neuropsychological Test Automated Battery (CANTAB) — sensitive to subtle changes in executive function and memory — and found no difference between evolocumab and placebo across 19 months. Importantly, the EBBINGHAUS analysis also stratified by achieved LDL-C and found no harm even in the patients who achieved the lowest LDL-C (<25 mg/dL).³⁸

23.3 Observational data: mixed but largely null

Observational data on statins and cognition include both the well-known PROSPER null and a substantial body of cohort studies that have reported protective associations with dementia incidence. A 2018 meta-analysis of 28 observational cohorts (Chu and colleagues) reported a pooled relative risk for any dementia of 0.83 (95 % CI 0.79–0.87) and for Alzheimer's disease specifically of 0.69 (0.60–0.80). The protective signal was larger in studies with longer follow-up.³⁹ A 2023 meta-analysis confirmed the direction of effect but with substantial residual confounding concerns.⁴⁰

It is important to be honest about why these observational results should not be taken as strong evidence of cognitive benefit: healthy-user bias is severe (patients who take statins long-term are typically more engaged with their health than those who do not), and prevalent-user bias is severe (patients who develop early cognitive impairment are more likely to discontinue all medications). The plausible inference from the observational data is that statins are not, on average, harmful to cognition; the inference that they are net-protective requires randomised evidence that is presently lacking.

23.4 Mechanistic plausibility

The mechanistic case for cognitive harm rests on three observations. First, the brain is the most cholesterol-rich organ in the body (containing roughly 25 % of total-body cholesterol despite being roughly 2 % of body weight), with cholesterol critical to myelination, synaptic membrane fluidity and neurotransmitter receptor function.⁴¹ Second, lipophilic statins (simvastatin, atorvastatin, lovastatin, fluvastatin, pitavastatin) cross the blood-brain barrier in measurable amounts; hydrophilic statins (pravastatin, rosuvastatin) cross to a far lesser extent.⁴² Third, the mevalonate-pathway dependencies of neuronal function — isoprenylation of small GTPases, dolichol-mediated N-glycosylation, ubiquinone-dependent oxidative phosphorylation — are non-trivial.

Set against this is the equally robust observation that the brain synthesises essentially all of its cholesterol in situ (the BBB is largely impermeable to circulating LDL particles), so systemic LDL lowering does not directly lower brain cholesterol; and that the Mendelian-randomisation evidence on genetically lower LDL-C and dementia is, at most, weakly protective and certainly not harmful.⁴³

The honest summary on mechanism: there is a plausible biological case that statins could affect cognition in susceptible individuals, but the population-level signal in randomised data is null, and the magnitude of any individual susceptibility effect appears small.

23.5 Steelmanning both sides

The orthodox steelman is that the pharmacovigilance signal is real but reflects a confluence of expectation effects, age-related cognitive change being misattributed to medication, and a small number of idiosyncratic responses; that randomised trials, including those of very-long-duration aggressive LDL-lowering, show no population-level signal; that the FDA label is a conservative regulatory action, not a finding of causation; and that the heterodox literature has substantially overweighted anecdotal evidence.

The heterodox steelman is that the FDA label was issued for a reason; that the trial cognitive endpoints are typically secondary, with limited power and crude instruments; that the case-report literature is consistent and biologically plausible; that the mechanistic case is non-trivial; and that population-level null findings do not exclude rare individual susceptibility, which is the question that actually matters to a specific patient deciding whether to take the drug.

23.6 Adjudication

Trial-level data does not support a population-level cognitive harm signal from statins. Rare idiosyncratic responses cannot be excluded, and any patient who reports subjective cognitive symptoms on a statin should be taken seriously, offered a structured trial of dechallenge and rechallenge analogous to the SAMSON approach, and switched to a hydrophilic statin (rosuvastatin, pravastatin) if rechallenge is indicated and the symptoms recur. The 2012 FDA label is best interpreted as a clinical heads-up rather than as a causal finding. The Roche scientist is likely to make exactly this argument, and the argument is correct.

Chapter 24 — Hepatic, renal and rhabdomyolytic events

24.1 Transaminase elevations and the 2012 monitoring change

Statins were originally marketed with quarterly liver-function-test monitoring requirements because of the well-documented dose-related rise in alanine aminotransferase (ALT) and aspartate aminotransferase (AST) seen in early phase 3 trials. The rise is typically asymptomatic, occurs in the first six months of therapy, is reversible on dose reduction or discontinuation, and is more common at high doses (atorvastatin 80 mg, rosuvastatin 40 mg, simvastatin 80 mg).⁴⁴ At those high doses the rate of >3× upper-limit-of-normal ALT elevation is roughly 1–3 %; at moderate-intensity doses it is <1 % and largely indistinguishable from background.

Clinically significant statin-attributed acute liver injury — defined as overt hepatitis with jaundice or hepatocellular injury requiring hospitalisation — is rare. The U.S. Drug-Induced Liver Injury Network has documented fewer than 100 cases in its registry across all statins over more than a decade, with simvastatin and atorvastatin being the most commonly implicated.⁴⁵ In 2012 the FDA removed the routine LFT monitoring recommendation from statin labelling, in recognition that the asymptomatic transaminase elevation was rarely clinically meaningful and that monitoring at fixed intervals had not been shown to prevent the rare cases of clinically significant injury.⁴⁶ LFTs are now recommended at baseline and only thereafter as clinically indicated.

There is no credible evidence that statins cause clinically meaningful chronic liver disease, accelerate non-alcoholic fatty liver disease (NAFLD) progression, or precipitate fibrosis. The 2024 EASL clinical practice guidelines explicitly endorse statin use in patients with NAFLD and even with compensated cirrhosis at appropriate cardiovascular indications.⁴⁷ This is one of the clearer "myth retired" stories of the past decade.

24.2 Rhabdomyolysis and the Baycol withdrawal

Rhabdomyolysis — frank muscle necrosis with CK >10× ULN, myoglobinuria, often with acute kidney injury — is the most feared statin adverse effect and the rarest. Background rates in the general population are approximately 1 per 100,000 person-years. Pooled trial and pharmacovigilance estimates of rhabdomyolysis on statin monotherapy are approximately 1 per 100,000 person-years at standard doses, rising to perhaps 1 per 10,000 person-years at the highest doses and substantially higher with drug-drug interactions (notably with gemfibrozil, certain macrolide antibiotics, calcium-channel blockers and protease inhibitors).⁴⁸

The defining historical episode is the August 2001 withdrawal of cerivastatin (Bayer's Baycol) after 31 confirmed U.S. deaths from rhabdomyolysis, and >100 deaths globally.⁴⁹ Cerivastatin was particularly problematic in combination with gemfibrozil — an OATP1B1-mediated drug-drug interaction that produced very high cerivastatin plasma concentrations. The Baycol episode is invoked repeatedly in heterodox commentary as evidence that the statin class is dangerous; in orthodox commentary it is invoked as evidence that the pharmacovigilance system worked — a drug with an unfavourable risk profile was withdrawn within four years of marketing, and the surviving statins have very different safety profiles.

Both readings have merit. The cerivastatin episode is a genuine reminder that pharmacological-class generalisations are dangerous; the rhabdomyolysis rate on simvastatin (especially at 80 mg) was significantly higher than on rosuvastatin or atorvastatin and led to FDA dose restrictions on simvastatin 80 mg in 2011. But it is also true that the remaining statins, used at recommended doses with attention to interactions, have rhabdomyolysis rates that are clinically negligible.⁵⁰

24.3 Renal effects: rosuvastatin proteinuria

Rosuvastatin was the subject of substantial regulatory attention at launch because phase 3 trials showed a small dose-related increase in dipstick-detectable proteinuria, particularly at 40 mg. The mechanism is thought to be inhibition of tubular reabsorption of low-molecular-weight proteins rather than glomerular injury — that is, a tubular rather than nephrotic proteinuria.⁵¹ The magnitude is small (median 30–60 mg/day at 40 mg), is not associated with progressive renal dysfunction, and is reversible. The 2003 FDA review of rosuvastatin restricted the 40 mg dose to patients not adequately controlled on 20 mg and required postmarketing surveillance, which did not subsequently identify a clinically significant renal signal.⁵²

The broader picture on statins and renal outcomes is favourable. The SHARP trial (simvastatin/ezetimibe in 9,270 chronic-kidney-disease patients) showed cardiovascular benefit without renal harm; the CTT IPD subgroup analyses in CKD patients are reassuring; and Mendelian-randomisation analyses of genetically lower LDL-C and renal function show null or weakly protective associations.⁵³⁵⁴

24.4 Drug-drug interactions and the SLCO1B1 story

Statin pharmacokinetics are dominated by two factors: hepatic CYP-mediated metabolism and hepatic OATP1B1-mediated uptake. CYP3A4 substrates among the statins are simvastatin, lovastatin and atorvastatin (the most clinically vulnerable to interactions with CYP3A4 inhibitors such as itraconazole, clarithromycin, ritonavir, diltiazem and grapefruit juice). CYP2C9 substrates include fluvastatin and, to a lesser extent, rosuvastatin (which is mostly excreted unchanged).⁵⁵ Pravastatin and pitavastatin are largely independent of major CYP pathways. Rosuvastatin and pravastatin are highly dependent on OATP1B1 uptake.

The OATP1B1 transporter, encoded by SLCO1B1, is the subject of one of the cleanest pharmacogenomic stories in cardiology. In a 2008 genome-wide association study of patients with simvastatin-associated myopathy in the SEARCH trial (12,064 participants on simvastatin 80 mg), Link and colleagues identified a single SNP, rs4149056, that conferred an odds ratio of 4.5 per copy of the C allele for definite or incipient myopathy; homozygotes had a 17 % cumulative risk of myopathy over five years on simvastatin 80 mg compared with 0.6 % in non-carriers.⁵⁶ The SNP reduces hepatic OATP1B1-mediated uptake of simvastatin acid, raising systemic exposure. Subsequent work showed that the effect is largest for simvastatin, intermediate for atorvastatin, smaller for rosuvastatin, and minimal for pravastatin and fluvastatin.⁵⁷

The clinical implication, codified in the 2014 (updated 2022) Clinical Pharmacogenetics Implementation Consortium guidelines, is that SLCO1B1 genotyping can guide statin selection in patients with a history of muscle symptoms or who require high-dose simvastatin.⁵⁸ This is one of the few pharmacogenomic recommendations in cardiology that has reasonably solid evidence and is being implemented in some specialist clinics; it has not yet entered routine primary care.

24.5 Steelmanning both sides

The orthodox steelman is that the rhabdomyolysis, hepatic, renal and drug-interaction safety profile of the modern statins (excluding cerivastatin) is well-characterised, has been actively managed by regulators (FDA simvastatin 80 mg restriction 2011, rosuvastatin 40 mg dose restrictions, removal of routine LFT monitoring 2012), and is favourable at recommended doses. The pharmacogenomic story is reassuring evidence that the safety signal is mechanistically tractable.

The heterodox steelman is that the Baycol episode shows the class is not categorically safe; that pharmacovigilance has historically been slow; that the dose-response for rhabdomyolysis means that the push toward high-intensity therapy in current guidelines is buying a real if small absolute increase in serious harm; and that the pharmacogenomic data have not been widely implemented, so patients who should not be on high-dose simvastatin are still being prescribed it.

24.6 Adjudication

The hepatic and renal safety profile of statins at recommended doses is favourable and the 2012 deregulation of LFT monitoring was evidence-based. Rhabdomyolysis is rare but real; the principal mitigation is attention to drug-drug interactions and to SLCO1B1 genotype in patients who require high-dose simvastatin. The Baycol episode is a real cautionary tale but does not generalise to the surviving statin class. The clinical implication: high-dose simvastatin should be largely retired in favour of rosuvastatin or atorvastatin for high-intensity therapy; CYP3A4 interaction lists should be consulted at every prescription; and pharmacogenomic testing has a defensible role in patients with prior muscle complaints.

Chapter 25 — Asymmetric adverse-event capture

25.1 The CTT IPD versus published-AE problem

The Cholesterol Treatment Trialists' (CTT) Collaboration is, in epidemiological terms, an extraordinary achievement: individual-patient-data meta-analysis across more than 30 statin trials and >170,000 participants. Its efficacy estimates are the gold standard for the field. Its adverse-event estimates, however, have been the subject of one of the longest-running methodological controversies in cardiovascular epidemiology — a controversy that involves the most senior figures in the field (Rory Collins, Colin Baigent for CTT; Aseem Malhotra, John Abramson, Maryanne Demasi for the heterodox critique) and that has crossed several times into the law courts and the BMJ.⁵⁹

The substantive issue is this. The CTT IPD efficacy estimates use raw event counts adjudicated by trial endpoint committees from a standardised dataset. The CTT IPD adverse-event estimates, by contrast, must rely on what was recorded in the case-report forms of the individual trials, and what was recorded in those CRFs was heavily shaped by the design of each trial — including, critically, whether a placebo run-in was used and whether the trial was actively soliciting muscle, cognitive or diabetes events. CTT's published AE rates for muscle complaints have consistently been near-null (often 0.1–0.6 % absolute excess); observational and pharmacovigilance estimates have been an order of magnitude higher. The CTT investigators have argued that this reflects the nocebo effect plus systematic miscoding of unrelated symptoms in unblinded settings; the heterodox critics have argued that it reflects a systematic underreporting of adverse events in the trials themselves.⁶⁰⁶¹

The 2014 BMJ "statin denial" controversy — when Abramson and Malhotra published articles citing the 18-20% figure for statin side effects in the BMJ that Collins challenged, leading to formal corrections — is the public face of this dispute.⁶² The Vlad Stockman and the Robert Wood Johnson Foundation reviews of the underlying CTT data access have repeatedly raised the question of whether independent reanalysis of the IPD should be possible. CTT's position, articulated most clearly in The Lancet 2016, is that the IPD is governed by contractual confidentiality agreements with the original sponsors that prevent independent re-release; the heterodox position is that this is itself the methodological problem.⁶³

25.2 Placebo run-in periods and their consequences

Many of the foundational statin trials used a placebo (or active drug) run-in period before randomisation. The standard rationale is to (i) confirm adherence — patients who cannot or will not take a daily tablet are excluded before randomisation, improving the per-protocol estimate; (ii) exclude patients who experience early intolerance, on the theory that they would have dropped out anyway and would only add noise; (iii) establish a stable baseline. The 4S trial (1994, simvastatin in 4,444 patients with established coronary disease) used an 8-week placebo run-in; JUPITER (2008, rosuvastatin in 17,802 primary-prevention patients) used a 4-week placebo run-in.⁶⁴⁶⁵ Other major trials (HPS, CARE, LIPID, AFCAPS/TexCAPS) used either placebo or active-drug run-ins of variable length.

The methodological consequence of a placebo run-in is that the trial randomises a population already enriched for tolerance — patients with early-onset muscle symptoms, gastrointestinal intolerance or simply forgetfulness have already been excluded. Estimates of how many patients are excluded vary by trial; in JUPITER, approximately 25 % of the 89,890 initially screened were not randomised, and a fraction of those exclusions were for run-in intolerance.⁶⁵ An active-drug run-in is, of course, much more problematic for AE estimation: by definition, every patient randomised has already demonstrated they can tolerate the active drug at least short-term.

The clinical implication is that trial AE rates systematically underestimate real-world tolerability. This is not a fatal flaw — the efficacy estimates are largely unaffected, because run-in selection affects the per-protocol denominator more than the risk ratio — but it is a real critique that should temper population-level expectations of how many patients started on statins will actually still be on them at 12 months.⁶⁶

25.3 Open-label phase, unblinding, and AE reporting

A second methodological wrinkle in several statin trials is the use of an open-label active-treatment period after the blinded randomised period. The HPS trial, for example, had a 5-year blinded period followed by post-trial follow-up; the LIPID and CARE trials had similar structures. AE reporting during the open-label phase is, by construction, no longer blinded, and the AE rates in these phases are systematically higher than in the blinded phases — but they are also subject to the same nocebo dynamics that operate in routine clinical practice.⁶⁷

The most rigorous attempt to quantify this is the ASCOT-LLA (Anglo-Scandinavian Cardiac Outcomes Trial — Lipid Lowering Arm) blinded-versus-unblinded reanalysis published by Gupta and colleagues in The Lancet in 2017. ASCOT-LLA had a 3.3-year blinded phase followed by an open-label extension. In the blinded phase, muscle-related AEs were 2.03 % per year on atorvastatin versus 2.00 % per year on placebo (rate ratio 1.03, P = 0.72) — that is, no signal. In the open-label phase, when patients knew they were on atorvastatin, the muscle AE rate rose to 1.26 % per year on atorvastatin versus 1.00 % per year on no-statin (rate ratio 1.41, P = 0.006).⁶⁸ The same patients, the same drug, the same study apparatus — only the knowledge of treatment allocation changed. This is one of the cleanest empirical demonstrations of the nocebo effect in cardiovascular medicine.

The orthodox interpretation is that the nocebo effect is real and dominant in unblinded settings. The heterodox response is that ASCOT-LLA's open-label phase had a different design (no placebo arm, comparator was "no statin") and that the comparison is therefore not as clean as advertised. Both have merit. ASCOT-LLA is the best single piece of evidence we have, and it is a meaningful but imperfect demonstration.

25.4 Pharmacovigilance versus trial signals

The third element of asymmetric AE capture is the systematic difference between trial AE rates and pharmacovigilance signals. Trials capture AEs through prospective, structured CRFs in a selected population; pharmacovigilance systems (FAERS in the US, the Yellow Card scheme in the UK, EudraVigilance in the EU) capture AEs through voluntary reporting in an unselected population. The signals from each are not directly comparable. Pharmacovigilance signals are subject to reporting bias (high-profile drugs generate more reports), notoriety bias (the 2012 FDA cognitive label change spiked subsequent cognitive AE reports), and channelling bias (patients prescribed a particular statin may differ systematically from those prescribed alternatives).⁶⁹

The honest position is that pharmacovigilance is excellent for detecting rare serious signals (Baycol rhabdomyolysis is the canonical example) but is poor for quantifying the prevalence of common mild signals (muscle aches, fatigue, brain fog). Trial data are the inverse — excellent for quantifying common mild signals in the studied population but underpowered for rare serious signals. The two systems are complements, not substitutes, and a complete safety picture requires both.

25.5 The "exclude before randomise" critique

The most pointed heterodox critique of the trial evidence is captured in the line: "if the trial excludes patients who can't tolerate the drug before randomisation, what does the trial's AE rate tell you about the population of patients who will be prescribed the drug?" This is a real point. The mainline statin trials randomised populations who had passed run-in screening; the AE rates that emerge are the AE rates in that selected population; extrapolating to a general population includes patients who would have been excluded.

The orthodox response is that (i) the absolute size of the run-in exclusion is small (typically 5–10 % of screened patients); (ii) the run-in excludes patients who would have discontinued anyway in routine practice, so it is not biasing the trial against the prescriber's question (which is "what happens if I prescribe this drug?") so much as against the patient's question (which is "what is the probability I will tolerate this drug?"); (iii) per-protocol efficacy estimates are improved by run-ins, and the modern intention-to-treat analyses are largely unaffected. The heterodox response is that even a 5–10 % run-in exclusion, when projected onto a 30 % observational intolerance rate, is sufficient to materially distort patient-level expectations.

Both readings have merit. The right summary is that run-in periods are a known methodological feature, that their effect on AE estimates is real but bounded, and that the gap between trial AE rates and real-world tolerability is partly attributable to run-in selection and partly to nocebo dynamics in unblinded prescribing.

25.6 Adjudication

Asymmetric adverse-event capture is a real and methodologically legitimate critique of the trial evidence. It does not invalidate the efficacy findings. It does mean that population-level expectations of tolerability should be tempered: real-world discontinuation rates of 25–50 % at 12 months are not surprising, and most of that discontinuation reflects a combination of true minor pharmacological intolerance, expectation effects, and the patient's broader pill-burden. The most honest framing for a primary-care prescriber is: "this drug works; you have somewhere between 1 % and 5 % chance of experiencing a specifically pharmacological side effect that is mostly muscle-related; you have a much larger chance of experiencing symptoms that may or may not be pharmacological; we will not know which you are until we try, and we should plan a structured rechallenge if needed."

Footnotes — Part IV

Part V — Special populations and common myths adjudicated

Prefatory note to Part V. Up to this point the document has treated the average trial participant as the unit of analysis. That is a defensible methodological starting point, but it is a starting point only. The patients who actually walk into a primary-care clinic are not the average trial participant. They are women, often older, often with kidney disease, often with diabetes, sometimes with an LDL-C of 7 mmol/L because their father died at 41 from coronary disease, and sometimes with an LDL-C of 4 mmol/L and a 10-year ASCVD risk below 5 %. The clinical question is whether the trial findings transfer, and to whom they transfer well. Part V works through the major special populations, then closes the document with an explicit adjudication of the eight most common myths and counter-myths in the public discourse, so that a substantive conversation with a friend who works for Roche can proceed on shared epistemic terrain.

Chapter 26 — Primary versus secondary prevention

26.1 The NNT chasm

The single most important framing question in the statin debate is whether we are talking about secondary prevention (statin therapy in a patient with established atherosclerotic cardiovascular disease — prior myocardial infarction, prior stroke, established coronary artery disease on imaging, peripheral artery disease) or primary prevention (statin therapy in a patient without such established disease, prescribed on the basis of estimated 10-year risk). The numerical bottom line is that the absolute benefit of statin therapy is roughly an order of magnitude larger in secondary than in primary prevention, because the baseline event rate against which the relative reduction operates is roughly an order of magnitude larger.¹

In the 4S trial — the foundational secondary-prevention trial of simvastatin in 4,444 patients with established coronary heart disease and a mean baseline LDL-C of 4.87 mmol/L (188 mg/dL) — the primary endpoint was all-cause mortality over a median 5.4 years. The simvastatin arm had 8.2 % all-cause mortality versus 11.5 % on placebo, an absolute risk reduction of 3.3 % and an NNT of 30.² For the broader composite of major vascular events, the NNT was approximately 15. This is the era-defining number that turned statins from a chemistry curiosity into a default of secondary-prevention cardiology.

In the primary-prevention trials, the absolute numbers are different. WOSCOPS (West of Scotland Coronary Prevention Study, 6,595 men aged 45–64 with elevated LDL-C and no prior MI) showed an absolute reduction in major coronary events of 2.4 % over five years (NNT 42).³ AFCAPS/TexCAPS (6,605 men and women with average cholesterol and below-average HDL-C) showed an absolute reduction of 1.4 % over 5.2 years (NNT 71).⁴ JUPITER (rosuvastatin in 17,802 primary-prevention patients selected by elevated hsCRP) showed an absolute reduction in the primary composite of 1.2 % over a median 1.9 years (NNT 95 over 1.9 years, equivalent to roughly NNT 250 over a single year).⁵ HOPE-3 (rosuvastatin in 12,705 intermediate-risk participants) showed an absolute reduction of 1.1 % over 5.6 years (NNT 91).⁶

The CTT meta-analyses, pooling across the primary-prevention trials and stratifying by baseline 10-year risk, give the cleanest summary: per 1 mmol/L reduction in LDL-C, the rate ratio for major vascular events is approximately 0.78, equivalent to a 22 % relative reduction. Absolute benefit scales linearly with baseline risk. At a 10-year baseline risk of 20 %, this corresponds to an absolute reduction of approximately 4–5 % over five years (NNT 20–25); at 10 %, approximately 2 % (NNT 50); at 5 %, approximately 1 % (NNT 100); at 2.5 %, approximately 0.5 % (NNT 200).⁷

26.2 Byrne 2022 and the "curb our enthusiasm" controversy

The most-cited recent challenge to primary-prevention statin enthusiasm is the systematic review and meta-analysis by Byrne and colleagues published in JAMA Internal Medicine in March 2022, under the editorial title "Curb our enthusiasm for statin therapy in primary prevention".⁸ Byrne et al. pooled 21 primary-prevention statin trials with >65,000 participants and reported median absolute risk reductions of 0.8 % for all-cause mortality, 1.3 % for myocardial infarction, and 0.4 % for stroke, with corresponding NNTs of 125, 77 and 250. The headline argument was that these absolute reductions are small enough that informed consent should foreground them rather than the more impressive-sounding relative risk reductions.

The Byrne paper drew an immediate and substantial response from orthodox lipidologists. The main critiques were threefold. First, the meta-analysis pooled trials with very different baseline risks, masking the well-established scaling of absolute benefit with risk. Second, the median trial follow-up was approximately 4 years, but statin benefits accrue over decades; the legacy effects from longer follow-up periods (WOSCOPS 20-year, ASCOT-LLA 16-year, JUPITER post-trial) are larger than the in-trial estimates suggest, because LDL exposure is cumulative and the trial period captures only an early slice of lifetime benefit.⁹ Third, the choice of all-cause mortality as a primary metric is sensible but underpowered: most primary-prevention trials are sized to detect cardiovascular events, not all-cause mortality, and pooling them on the latter endpoint disadvantages the more powered-on-CV-events trials.

The orthodox steelman of the Byrne paper is that its arithmetic is correct and that the implications for patient communication are real: a patient should be told the absolute numbers, not just the relative numbers. The orthodox critique is that Byrne misframes the implication: the absolute numbers are small at the population mean but legitimate and often larger at the patient-specific level once baseline risk is properly accounted for.

The heterodox steelman is the converse: the absolute numbers are what matter to patients, the population-level absolute numbers in primary prevention are genuinely modest, and the field has historically communicated relative risk in a way that systematically overstates the patient-relevant benefit. The heterodox critique of the orthodox response is that the appeal to legacy effects is partly hand-waving — the legacy data are observational extensions of trials with crossover treatment and substantial loss to follow-up — and that the "scale with risk" defense is essentially conceding the primary-prevention low-risk case to the heterodox.

My adjudication: both readings are defensible, both are operative in current practice, and the practical resolution is the one already codified in the 2018 AHA/ACC and 2019 ESC/EAS guidelines and now strengthened by the 2024 AHA PREVENT equations — namely, that statin eligibility in primary prevention should be a shared decision driven by individual ASCVD risk estimation, not a blanket policy across the population.¹⁰ The Byrne paper, on this reading, is best understood as a methodological reminder rather than as a refutation. It does, however, sharpen the case that very low-risk patients (10-year ASCVD risk <5 %) are poor candidates for statin therapy in the absence of other indications.

26.3 The legitimate primary-prevention debate

The legitimate primary-prevention debate, stripped of polemic, is about where on the risk spectrum the benefit-harm ratio tips negative. Three legitimate positions exist within mainstream practice.

The first (most orthodox): statins are net-beneficial down to 10-year ASCVD risk of approximately 5 %, with shared decision-making appropriate down to roughly 2.5 %. This is broadly the position of the 2018 AHA/ACC guidelines and the 2019 ESC/EAS guidelines.¹¹¹²

The second (intermediate): statins are net-beneficial at 10-year risk above 10 %; below 10 % is genuinely close and requires individual assessment of competing risks (diabetes risk, longevity expectation, polypharmacy burden, patient preference). This is approximately the position of the 2014 NICE guidelines as reinforced by 2023 NICE updates,¹³ and is consistent with the Byrne et al critique.

The third (heterodox): the absolute benefit in primary prevention at any risk level commonly seen in current practice is too small to justify population-level prescribing, particularly given the diabetes signal, the muscle complaint signal, and the medicalisation of essentially healthy adults. This is the position articulated by John Abramson, Aseem Malhotra, and the BMJ sceptical lipid community.¹⁴

The empirical disagreement between the first and second positions is genuinely close and is exactly the question that the AHA PREVENT equations and the STAREE / PREVENTABLE trials are likely to inform. The disagreement between the first/second and the third is partly empirical and partly evaluative — it depends on how one weighs a deferred MI against an iatrogenic diabetes case against the inconvenience of a daily tablet.

26.4 Clinical bottom line for primary versus secondary

For secondary prevention, the case for high-intensity statin therapy (atorvastatin 40–80 mg, rosuvastatin 20–40 mg) is essentially uncontested across orthodox and heterodox commentators. NNTs in the 15–30 range over 5 years are robust, dose-effects are clear, and even the most sceptical critics concede that this is the population in which the drug class earns its keep.

For primary prevention, the case is risk-stratified. In patients with 10-year ASCVD risk >10 %, the case is strong. In patients with risk 5–10 %, the case is real but smaller and warrants explicit shared decision-making. In patients with risk <5 %, the case is genuinely weak in the absence of other indications (familial hypercholesterolaemia, very high single risk factors, diabetes with target-organ damage, coronary calcium score >100). The 2024 AHA PREVENT equations are likely to reduce the eligible primary-prevention population substantially, and the new evidence from STAREE and PREVENTABLE will reshape the >70 conversation specifically.

Chapter 27 — Women

27.1 Under-representation pre-2010

The foundational statin trials of the 1990s and early 2000s systematically under-represented women. 4S enrolled 19 % women (827 of 4,444); WOSCOPS enrolled 0 % (it was an all-male trial); CARE enrolled 14 %; LIPID 17 %; AFCAPS/TexCAPS 15 %; HPS 25 %; PROSPER 52 % (this is the major exception, by design); ASCOT-LLA 19 %.¹⁵ The 2008–2009 CTT IPD analyses noted explicitly that sex-specific estimates of statin benefit had wide confidence intervals because of low enrolment of women in the older trials.

The structural reasons for under-representation are well-rehearsed: women develop clinically manifest CHD approximately a decade later than men, so trial inclusion criteria based on prior MI or established CHD selected a male-dominant population by construction; trial protocols of the 1990s often excluded women of reproductive potential because of regulatory caution; and recruitment networks were dominated by cardiology clinics that themselves saw fewer women.¹⁶

The trial landscape improved through the 2000s. JUPITER (2008) was 38 % female. HOPE-3 (2016) was 46 % female. PREVENTABLE (2026 readout) is targeting roughly 50 % female enrolment. The contemporary cardiovascular outcomes trials of PCSK9 inhibitors and bempedoic acid have approached 30–40 % female enrolment. The under-representation problem has shrunk substantially, but the legacy of older trials means that pooled meta-analytic estimates still carry more uncertainty for women than for men.

27.2 The CTT 2015 sex-subgroup meta-analysis

The most definitive analysis of statin efficacy by sex is the CTT IPD meta-analysis published in The Lancet in 2015, which pooled 27 trials with 174,000 participants (of whom 27 % were women).¹⁷ The headline finding was that the relative risk reduction per 1 mmol/L LDL-C lowering was indistinguishable between men and women: rate ratio for major vascular events 0.78 (95 % CI 0.76–0.81) in men and 0.84 (0.78–0.91) in women. The difference between these two rate ratios is small and statistically marginal; the lower bound of the female confidence interval still indicates substantial benefit.

The absolute benefit, of course, depends on baseline event rates, and women at any given age have lower baseline cardiovascular event rates than men. The corollary is that the NNT for primary prevention in women is larger than in men of the same age. CTT estimated that per 1 mmol/L LDL-C lowering over 5 years, women in primary prevention had an absolute reduction in major vascular events of approximately 1.5 % at baseline 10-year risk of 10 %, compared with approximately 2.5 % for men at the same risk level — reflecting the lower female event rate.

27.3 The JUPITER female subgroup

A specific point of contention is the JUPITER female subgroup. JUPITER recruited 6,801 women out of 17,802 total participants; the primary composite endpoint (MI, stroke, unstable angina, revascularisation, CV death) was reduced by 46 % in women compared with 42 % in men, both statistically significant.⁵ However, the hard endpoints analysis — MI plus stroke plus CV death only — showed a non-significant trend in women (HR 0.65, 95 % CI 0.40–1.04) compared with a significant reduction in men (HR 0.58, 95 % CI 0.43–0.78).¹⁸ The female subgroup was, on the hard-endpoint analysis, the most equivocal in JUPITER.

The orthodox interpretation, articulated by Ridker and Mora in 2010, was that the female subgroup result was statistically consistent with the male result and that the lack of significance reflected lower power.¹⁹ The heterodox interpretation, articulated in commentaries by Eva Lonn and others, was that the female-specific evidence was genuinely weaker and that JUPITER had been over-marketed to women on the basis of an under-powered subgroup.²⁰ The CTT 2015 meta-analysis has largely closed this debate at the population level, but the JUPITER female subgroup remains a useful illustration of how subgroup analyses can mislead.

27.4 Where the female evidence is genuinely weaker

The honest summary of where the female-specific evidence is weaker than the male-specific evidence is as follows. (i) Primary-prevention statin therapy in women under age 55 with no other risk factors has very thin evidence; this is a population in whom event rates over 10 years are very low and absolute benefits are very small. (ii) The 2014 USPSTF and 2016 USPSTF recommendations both noted explicitly that the evidence for statins in women without prior cardiovascular disease was "less certain" than in men.²¹ (iii) The 2024 USPSTF statement, while recommending statin therapy in adults aged 40–75 at >10 % 10-year ASCVD risk regardless of sex, acknowledges that the sex-specific evidence remains thinner in women.²²

The areas where the female evidence is not weaker include: secondary prevention (4S, CARE, LIPID female subgroups all consistent with male subgroups, CTT 2015 pooled estimate robust), familial hypercholesterolaemia (evidence base smaller but biological case identical), and post-ACS intensive therapy (PROVE-IT and IMPROVE-IT female subgroups consistent with main results).

27.5 Adjudication

Statin therapy in women is supported by trial-level evidence comparable to men for secondary prevention and for high-risk primary prevention. The case in low-risk primary prevention in women is genuinely thinner than in men because (a) the under-representation in older trials reduces statistical power for sex-specific estimates and (b) the lower baseline event rate makes the absolute benefit smaller at any given age. The honest framing for a female patient is to use sex-specific risk equations (the 2024 AHA PREVENT equations explicitly do this) and to use shared decision-making at the margins. The reflex of prescribing identical regimens to a 50-year-old man and a 50-year-old woman at the same LDL-C and total cholesterol is not supported by the absolute-benefit evidence.

Chapter 28 — The over-75s: the live frontier

28.1 What we currently know

The over-75 population is the largest under-served population in cardiovascular pharmacotherapy. Cardiovascular event rates rise steeply with age; competing mortality risks (cancer, dementia, frailty) also rise; the trial evidence base is the thinnest of any major patient population; and the bedside conversation is the hardest. As of 2026, three pieces of evidence shape current practice.

The first is PROSPER (Prospective Study of Pravastatin in the Elderly at Risk), published by Shepherd and colleagues in The Lancet in 2002. PROSPER randomised 5,804 men and women aged 70–82 with vascular disease or risk factors to pravastatin 40 mg or placebo for a mean 3.2 years. The primary endpoint (CHD death + nonfatal MI + fatal/nonfatal stroke) was reduced by 15 % (HR 0.85, 95 % CI 0.74–0.97).²³ But the result was driven entirely by the secondary-prevention subgroup; in the primary-prevention subgroup of approximately 3,200 participants, there was no statistically significant benefit. PROSPER also showed a non-significant increase in cancer incidence in the pravastatin arm that did not replicate in later analyses but did dampen prescribing enthusiasm in the elderly. PROSPER's equivocal result has been the single most-cited piece of evidence in the under-prescribing of statins to the elderly for two decades.

The second is the CTT subgroup analyses by age. The 2010 and 2019 CTT IPD meta-analyses both reported that the relative risk reduction per 1 mmol/L LDL-C lowering was preserved across age strata, including in the >75 subgroup, although the >75 subgroup had wider confidence intervals.²⁴ The 2019 CTT update specifically addressed the elderly question, pooling 28 trials with 14,483 participants aged >75; the rate ratio for major vascular events per 1 mmol/L LDL-C reduction was 0.85 (95 % CI 0.78–0.93), broadly consistent with younger age groups.²⁵ In primary prevention specifically in the >75 subgroup, the rate ratio was 0.92 (0.73–1.16) — directionally consistent but not statistically significant, reflecting limited primary-prevention enrolment in this age group across the underlying trials.

The third is the observational and pharmacoepidemiological literature. Multiple large cohort studies (Veterans Affairs, Geisinger, Korean National Health Insurance, French national health database) have reported associations between statin therapy and reduced mortality in the >75 population, but the magnitude and statistical robustness vary substantially with adjustment strategy.²⁶ Healthy-user bias is severe in this age group, and observational estimates should be considered hypothesis-generating only.

28.2 STAREE and PREVENTABLE: the data we are waiting for

The two trials currently positioned to resolve the elderly primary-prevention question are STAREE (Statins in Reducing Events in the Elderly) and PREVENTABLE (Pragmatic Evaluation of Events and Benefits of Lipid-lowering in Older Adults).

STAREE, an Australian trial led by Sophia Zoungas and the Monash team, randomised 9,971 community-dwelling adults aged ≥70 with no prior CVD, diabetes or dementia to atorvastatin 40 mg or placebo. Mean baseline age was 74.7 years; 40 % were ≥75 at randomisation; 52 % were women. Co-primary endpoints are MACE (with coronary revascularisation added) and disability-free survival (freedom from new dementia and persistent physical disability).²⁷ Anticipated completion was Q4 2025, with final results expected in 2026.²⁸ STAREE is the largest dedicated trial of statin therapy in elderly primary prevention ever conducted, and its disability-free-survival endpoint is specifically designed to address the question patients in this age group actually care about: not just whether the drug prevents an MI, but whether the years it adds are years worth living.

PREVENTABLE, a US pragmatic trial sponsored by the NIA and NHLBI, is randomising 20,000 community-dwelling adults aged ≥75 with no CVD, no severe disability and no dementia to atorvastatin 40 mg or placebo. The primary endpoint is survival free of new dementia or persisting disability over a mean follow-up of approximately five years.²⁹ Co-secondary endpoints include a composite CV endpoint and a composite of mild cognitive impairment / dementia. Expected completion is December 2026.²⁹ PREVENTABLE is even more pragmatically designed than STAREE — recruitment is largely through primary care, follow-up is largely by telephone — and it is the trial that will most directly inform what a US primary-care physician should offer a 78-year-old without prior CVD.

28.3 The 2024 NLA/AGS expert consensus

In the absence of definitive trial data, the National Lipid Association and the American Geriatrics Society issued a joint 2024 expert consensus on managing hypercholesterolaemia in adults >75 without established ASCVD.³⁰ The consensus is carefully hedged. Its principal recommendations are: (i) shared decision-making is the default; (ii) life expectancy >10 years is a soft prerequisite for considering statin therapy; (iii) coronary artery calcium scoring is helpful at the margins (CAC = 0 supports deprescription; CAC >100 supports prescription); (iv) high-intensity therapy in this age group is rarely indicated; (v) the diabetes and frailty risks should be weighed explicitly; (vi) the data are awaited from STAREE and PREVENTABLE.

This is one of the most honest documents in the field. It explicitly acknowledges the limits of current evidence, refuses to make blanket recommendations, and frames the question as one of individualised judgement rather than algorithmic prescribing.

28.4 Steelmanning both sides

The orthodox steelman is that the relative risk reduction is preserved at older ages; that the absolute event rate is higher at older ages, so the absolute benefit should be similar or larger; that the legacy effects of LDL exposure are cumulative and a 75-year-old has typically had decades of unmanaged exposure already; that secondary prevention in the elderly is unambiguously beneficial and primary prevention is likely beneficial pending STAREE / PREVENTABLE; and that age alone should not be a contraindication.

The heterodox steelman is that PROSPER was equivocal in primary prevention; that the trial evidence in >75 primary prevention is currently insufficient; that competing risks (cancer, dementia, frailty) substantially erode the expected lifetime benefit; that the diabetes signal is particularly costly in a population already at high diabetes risk; that polypharmacy adds non-trivial interaction risks; and that the geriatric principle of "treat the patient, not the lab" applies with particular force.

28.5 Adjudication

For secondary prevention in the >75 population, the case is solid and uncontested. For primary prevention, the case is currently honest-to-uncertain, and the most defensible position is the NLA/AGS shared-decision-making framework. STAREE results in 2026 and PREVENTABLE results in late 2026/early 2027 will substantially reshape this conversation. If both trials show disability-free-survival benefit, the case for routine primary-prevention statin therapy in selected >75 patients will be strong; if both are equivocal or negative, the case for not routinely prescribing to this population will be strong. The honest answer for a 78-year-old patient asking now is that the data are imminent and the decision can reasonably be deferred 12–18 months in patients without other strong indications.

Chapter 29 — Familial hypercholesterolaemia

29.1 The single strongest case

Familial hypercholesterolaemia (FH) is the cleanest natural experiment in cardiovascular biology and the patient population in whom statin (and PCSK9-inhibitor) therapy generates the least controversy across the entire orthodox-heterodox spectrum. The reason is straightforward: FH demonstrates, with the clarity that only genetics can provide, that lifelong elevation of LDL-C causes premature atherosclerotic disease in the absence of any other identifiable risk factor.

The genetics: heterozygous FH (HeFH) is caused by a loss-of-function mutation in one of three genes — LDLR (low-density lipoprotein receptor; ~85–90 % of identified cases), APOB (apolipoprotein B; ~5 %), or PCSK9 (proprotein convertase subtilisin/kexin type 9; gain-of-function; ~1–3 %). Less common causes include mutations in LDLRAP1 (autosomal recessive hypercholesterolaemia) and the polygenic phenotype that mimics FH biochemically.³¹ Prevalence of HeFH was historically estimated at 1 in 500 but is now best estimated at approximately 1 in 250 in most European-ancestry populations, with regional founder-effect clusters (Afrikaners, French Canadians, Lebanese, certain Ashkenazi populations) at higher frequencies.³² Homozygous FH (HoFH) — biallelic mutations — is far rarer, with prevalence approximately 1 in 300,000 in unselected populations.³³

29.2 The natural history without treatment

Untreated HeFH carries an LDL-C typically >5 mmol/L (>190 mg/dL) by adolescence, often >7 mmol/L (270 mg/dL) in adulthood, and is associated with onset of clinically manifest coronary heart disease in the 40s and 50s for men and the 50s and 60s for women. The classical Simon Broome registry data from the UK showed that untreated HeFH men had approximately a 100-fold increased relative risk of CHD mortality in their 30s and 40s compared with the general population, with absolute coronary mortality of approximately 50 % by age 60 in untreated cases.³⁴

Untreated HoFH is catastrophic: LDL-C >13 mmol/L (>500 mg/dL) from birth, xanthomas in childhood, aortic stenosis from cholesterol deposition, coronary disease by age 10–20, and historical untreated life expectancy in the second decade.³³

29.3 The transformation by treatment

The contrast with the treated population is one of the most striking demonstrations of pharmacological efficacy in medicine. The Dutch FH cohort registry data, the Norwegian FH database, the UK Simon Broome and FH-EAS registries all converge on the conclusion that statin therapy initiated in adolescence or early adulthood reduces coronary mortality in HeFH to near-population baseline. The CASCADE-FH and Spanish DLCN registries show similar results.³⁵

The trial-level evidence in FH is necessarily smaller than in unselected populations because of the rarity of the disease. The most important prospective data are: (i) the 20-year extended follow-up of statin-treated children with FH started in adolescence, showing carotid intima-media thickness comparable to unaffected siblings rather than to untreated FH controls;³⁶ (ii) the FH-specific PCSK9-inhibitor trials (RUTHERFORD-2 for evolocumab in HeFH, ODYSSEY-FH I and II for alirocumab) showing 60 %+ further LDL-C reduction on top of maximally tolerated statin therapy;³⁷ (iii) the long-term HoFH PCSK9-inhibitor data (TESLA, HAUSER trials, evinacumab for ANGPTL3 in HoFH) which has transformed what was historically a paediatric apheresis-dependent disease.³⁸

29.4 Cascade screening and the genetic angle

Cascade screening — once an FH patient is identified, systematic genetic and biochemical screening of first-degree relatives — is one of the highest-yield interventions in cardiovascular preventive medicine. The Dutch national FH cascade screening programme (1994–2014) identified approximately 28,000 FH cases over two decades, with each index case yielding on average two additional first-degree relatives with FH.³⁹ The yield is sufficiently high that cost-effectiveness analyses uniformly support cascade screening in any health system with reasonable family-tracing infrastructure.⁴⁰

Genetic testing has moved from a research tool to a routine clinical test. The decision-rule frameworks (Dutch Lipid Clinic Network criteria, Simon Broome criteria, MEDPED) are still used for clinical-probability assessment, but a confirmatory genetic test is now the standard of care in any patient meeting probable FH by clinical criteria.⁴¹ Genetic confirmation matters for several reasons: (i) prognostic certainty in the proband and the cascade; (ii) reproductive counselling for the couple at risk of HoFH offspring; (iii) eligibility for PCSK9 inhibitors and (in some jurisdictions) evinacumab; (iv) family-level engagement with treatment, which is consistently higher when there is a genetic diagnosis attached to the family history.⁴²

29.5 The PCSK9-inhibitor story specifically in FH

PCSK9 inhibitors are particularly impactful in FH for two reasons. First, FH patients frequently have LDL-C that remains substantially elevated even on high-intensity statin plus ezetimibe — typically achieved LDL-C is 3–4 mmol/L in HeFH on triple oral therapy, versus a treatment goal of <1.8 mmol/L (or even <1.4 mmol/L for very-high-risk). PCSK9 inhibitors typically add a further 50–60 % reduction, often bringing FH patients to target for the first time.⁴³ Second, FH is a population in whom the relative benefit translates into a large absolute benefit because of the high baseline event rate, so the cost-effectiveness analyses are uniformly favourable.⁴⁴

The 2023 ESC/EAS focused update on management of dyslipidaemias and the 2024 NLA position statement on PCSK9 inhibitors both place FH (specifically HeFH not at target on maximal oral therapy, and all HoFH) at the top of the priority list for PCSK9 inhibitor access.⁴⁵⁴⁶

29.6 Adjudication

Familial hypercholesterolaemia is the cleanest case for cholesterol-lowering therapy in the entire cardiovascular literature. It is one of the few major clinical entities in which the orthodox case and the heterodox case converge: both camps accept that FH patients should be aggressively treated, both accept the role of statins, ezetimibe and PCSK9 inhibitors, and the only meaningful disagreement is at the technical level of which target LDL-C and which agent first. For an Anthony-versus-Roche conversation, FH is the example to cite for "even the most sceptical heterodox commentators agree that cholesterol-lowering drugs save lives in this population". If your friend wants a single example of why the field is not all hype, this is it.

Chapter 30 — Diabetes, chronic kidney disease and post-acute coronary syndrome

30.1 Diabetes: CARDS as the defining trial

The Collaborative Atorvastatin Diabetes Study (CARDS), published by Colhoun and colleagues in The Lancet in 2004, was a primary-prevention trial in 2,838 patients with type 2 diabetes, no prior cardiovascular disease and an LDL-C <4.14 mmol/L (160 mg/dL), randomised to atorvastatin 10 mg or placebo.⁴⁷ The trial was stopped early after a median 3.9 years of follow-up because the prespecified efficacy criterion was met. The primary endpoint (acute CHD events, coronary revascularisation, stroke) was reduced by 37 % (rate ratio 0.63, 95 % CI 0.48–0.83), with absolute risk reduction of approximately 3.2 % and NNT 27. All-cause mortality was reduced by 27 % (HR 0.73, 95 % CI 0.52–1.01), narrowly missing statistical significance because of the early termination.

CARDS is the trial that turned moderate-intensity statin therapy from optional to default in primary prevention in type 2 diabetes. Its results have been replicated in subsequent meta-analyses (notably the CTT 2008 diabetes-specific IPD analysis pooling 18,686 patients with diabetes across 14 trials) showing consistent relative benefit and substantial absolute benefit because of the elevated baseline cardiovascular risk in this population.⁴⁸ The current 2023 ADA Standards of Care, the 2024 KDIGO guidelines and the 2019 ESC/EAS dyslipidaemia guidelines all recommend statin therapy for essentially all adults with type 2 diabetes over age 40, with intensity stratified by 10-year ASCVD risk.⁴⁹⁵⁰

The case for statin therapy in type 1 diabetes is less trial-supported but is generally extrapolated from the type 2 evidence in the presence of any additional risk factor. The 2024 ADA and JBS guidelines recommend statin therapy in adults with type 1 diabetes age >40 or with target-organ damage.⁴⁹

30.2 Chronic kidney disease: SHARP

The Study of Heart And Renal Protection (SHARP), published by Baigent and colleagues in The Lancet in 2011, was a trial of simvastatin 20 mg plus ezetimibe 10 mg versus placebo in 9,270 patients with chronic kidney disease (including 3,023 on dialysis at randomisation), followed for a median 4.9 years.⁵¹ The primary endpoint (major atherosclerotic events: nonfatal MI, coronary death, non-haemorrhagic stroke, revascularisation) was reduced by 17 % (rate ratio 0.83, 95 % CI 0.74–0.94), with an absolute reduction of 1.8 %.

SHARP's importance is twofold. First, it established that LDL-lowering therapy benefits CKD patients with all the standard cardiovascular endpoints, despite the previously equivocal results of the 4D trial (atorvastatin in dialysis patients) and AURORA (rosuvastatin in dialysis patients), both of which had been negative or equivocal.⁵²⁵³ Second, it established ezetimibe as a useful add-on in this population, with the simvastatin/ezetimibe combination achieving LDL-C reductions comparable to high-dose simvastatin alone but without the dose-related rhabdomyolysis signal that had led to FDA restriction of simvastatin 80 mg.

The pre-dialysis CKD subgroup of SHARP (n = 6,247) had a larger absolute benefit than the dialysis subgroup; in the dialysis subgroup, the absolute reduction did not reach statistical significance, consistent with the prior 4D and AURORA results. The current consensus is that statin (or statin/ezetimibe) therapy is beneficial in CKD stages 3–4 and in renal-transplant recipients; the case in dialysis-dependent CKD is weaker, and current 2024 KDIGO guidelines recommend continuing rather than initiating statin therapy in newly-dialysis patients.⁵⁰

30.3 Post-acute coronary syndrome: PROVE-IT and IMPROVE-IT

The strongest case for intensive (rather than moderate) LDL-lowering therapy is in the immediate post-acute-coronary-syndrome (post-ACS) period. Two trials define this evidence base.

PROVE-IT-TIMI 22 (Cannon and colleagues, NEJM 2004) randomised 4,162 patients within 10 days of an ACS to atorvastatin 80 mg or pravastatin 40 mg, with a primary composite endpoint of death from any cause, MI, documented unstable angina requiring rehospitalisation, revascularisation >30 days after randomisation, and stroke. Over a mean 24 months, the primary endpoint was reduced by 16 % in the atorvastatin arm (22.4 % vs 26.3 %, HR 0.84, 95 % CI 0.74–0.95), with achieved median LDL-C of 1.6 mmol/L (62 mg/dL) on atorvastatin versus 2.5 mmol/L (95 mg/dL) on pravastatin.⁵⁴ PROVE-IT established that more-intensive LDL-lowering produces better outcomes than less-intensive lowering in the post-ACS population, and triggered the move toward "lower is better" as a guiding principle in secondary prevention.

IMPROVE-IT (Cannon and colleagues, NEJM 2015) randomised 18,144 post-ACS patients to ezetimibe + simvastatin 40 mg or simvastatin 40 mg alone for a median 6 years, with achieved median LDL-C of 1.4 mmol/L (54 mg/dL) on the combination versus 1.8 mmol/L (70 mg/dL) on simvastatin alone. The primary composite (CV death, MI, hospitalised unstable angina, coronary revascularisation, stroke) was reduced by 6.4 % relative (HR 0.936, 95 % CI 0.89–0.99), with absolute reduction of 2.0 % over 7 years (NNT 50).⁵⁵ IMPROVE-IT was the first cardiovascular outcomes trial of a non-statin LDL-lowering agent (ezetimibe) to demonstrate cardiovascular benefit, validating the "LDL hypothesis" with a non-statin mechanism and providing the foundational evidence for later non-statin therapies (PCSK9 inhibitors, bempedoic acid).

30.4 The most uncontested evidence

Diabetes, CKD and post-ACS represent the three patient populations in which the case for statin (and broader LDL-lowering) therapy is strongest and least contested across the orthodox-heterodox spectrum. Even the most sceptical commentators concede the case in established CHD post-ACS and in diabetes with cardiovascular risk factors. The reason is that absolute event rates are high, NNTs are favourable, and the trials are well-powered. For a clinician deciding whom to treat, these populations are the easy cases.

30.5 Adjudication

The case for statin therapy in (i) adults with type 2 diabetes age >40, (ii) CKD stages 3–4 (and to a lesser extent stage 5 not on dialysis), and (iii) post-ACS patients (high-intensity therapy ideally with ezetimibe add-on if LDL-C remains above target) is essentially uncontested. The combined CARDS, SHARP, PROVE-IT and IMPROVE-IT evidence base, supplemented by FOURIER for PCSK9 inhibitors and CLEAR Outcomes for bempedoic acid in the statin-intolerant subgroup, establishes a hierarchy of LDL-C targets and therapeutic intensities that current guidelines have largely codified.

Chapter 31 — Common myths and counter-myths, adjudicated

This chapter works through the eight most common claims and counter-claims that circulate in lay discourse about cholesterol and statins. For each I state the claim, identify the kernel of truth, identify the overreach, and locate where the evidence lands.

31.1 "Cholesterol doesn't cause heart disease"

Kernel of truth. Cholesterol as a chemical species — a single molecule of C27H46O — does not "cause" heart disease in the way that, say, asbestos causes mesothelioma. What causes atherosclerotic plaque is the deposition of apolipoprotein-B-containing lipoprotein particles (LDL, IDL, Lp(a), VLDL remnants) within the arterial intima, with subsequent oxidation, immune cell recruitment, foam-cell formation and plaque progression.⁵⁶ The bare cholesterol mass per particle is a useful but imperfect proxy for the actual atherogenic species, which is the ApoB-containing particle itself. There is real individual variation in particle number versus particle cholesterol content (the "discordance" phenomenon described in MESA, ARIC, the Framingham Offspring cohort, and the AMORIS/INTERHEART analyses), and a patient with low LDL-C but high ApoB or high non-HDL-C is genuinely at elevated risk despite the reassuring LDL-C value.⁵⁷

Overreach. The heterodox slogan "cholesterol doesn't cause heart disease" is typically deployed not as a particle-versus-mass methodological point but as a global dismissal of the lipid hypothesis. This dismissal is incompatible with the convergent evidence from Mendelian randomisation (multiple independent genetic variants in LDLR, PCSK9, NPC1L1, HMGCR all producing the predicted effect on cardiovascular events), randomised trials (statins, ezetimibe, PCSK9 inhibitors, bempedoic acid), and mechanistic biology (the entire pathway from receptor-mediated LDL uptake through plaque progression is biochemically characterised).⁵⁶

Where it lands. ApoB-containing lipoproteins are causal for atherosclerotic cardiovascular disease. LDL-C is a useful clinical proxy that captures most but not all of the ApoB signal. In specific patients, ApoB or non-HDL-C measurement provides additional information beyond LDL-C, particularly in the discordant phenotype (elevated triglycerides, metabolic syndrome, diabetes, very low LDL-C). The argument has shifted from "is cholesterol involved" to "which lipoprotein measurement best captures the involvement and which therapy best targets it" — a far more sophisticated and clinically useful framing than the legacy debate.

31.2 "Statins are a Big Pharma scam"

Kernel of truth. The pharmaceutical industry has a documented history of (i) sponsoring and shaping clinical trials in ways that flatter their products, (ii) ghost-writing peer-reviewed articles, (iii) selectively publishing favourable results, (iv) using surrogate endpoints to expand label claims, and (v) maintaining financial relationships with key opinion leaders that distort the evidence-synthesis process. The ENHANCE trial of ezetimibe (Vytorin), in which the results were delayed for nearly two years after database lock under Merck/Schering-Plough sponsorship, with the eventual publication in 2008 showing that ezetimibe did not produce additional regression of carotid intima-media thickness despite reducing LDL-C, is a case study in industry conduct that should be widely taught.⁵⁸ The Vioxx affair (Merck rofecoxib) was contemporaneous and culturally formative. More recently, the Vytorin marketing campaign during the ENHANCE delay raised serious questions about disclosure practices.

Overreach. The "Big Pharma scam" framing typically extends from the legitimate critique of industry conduct to a wholesale dismissal of the statin evidence base. This is not defensible. The Mendelian-randomisation evidence is generated by academic groups using publicly available genetic data and is not industry-sponsored. Many of the foundational statin trials (notably WOSCOPS, 4S, HPS, PROSPER and PROVE-IT in part) were investigator-initiated with industry funding but academic management; the IPD is held by the CTT collaboration at Oxford. The post-patent generic statin landscape — atorvastatin generic since 2011, simvastatin generic since 2006, rosuvastatin generic since 2016 — has produced replication studies and pharmacoepidemiological analyses in populations with essentially no industry stake.⁵⁹ If the statin effect were a manufactured artefact, it should have evaporated in the post-patent generic era. It has not.

Where it lands. Industry conflicts of interest have legitimately distorted some specific trials and analyses, and they remain a real factor in evidence interpretation. The core efficacy effect of LDL-lowering survives the COI critique because (i) it is confirmed by Mendelian-randomisation evidence that is independent of trial sponsorship; (ii) it is replicated by investigator-initiated post-patent observational data; (iii) the dose-response relationship across statins of different commercial sponsors is consistent. The safety signal interpretation is more legitimately COI-influenced; the efficacy signal is not.

31.3 "Statins cause dementia"

Kernel of truth. The 2012 FDA label change adding cognitive impairment to the precautions section was a real regulatory action based on real pharmacovigilance signals. Case reports of transient global amnesia and short-term memory complaints on statin therapy have a consistent enough pattern (improvement on dechallenge, recurrence on rechallenge in a subset) that pharmacovigilance attention is warranted. The mechanistic case — lipophilic statin penetration into the CNS, mevalonate-pathway dependencies of neurons, CoQ10 — is biologically coherent.⁶⁰

Overreach. Extrapolating from individual case reports and pharmacovigilance signals to a population-level cognitive harm signal is not supported by the trial data. The HPS, PROSPER, JUPITER, EBBINGHAUS and FOURIER-OLE prospective cognitive assessments are uniformly null or trending favourable, even at very low achieved LDL-C and over follow-up periods extending to 7+ years.⁶¹ The large observational cohorts trend favourable but are confounded by healthy-user bias. The Mendelian-randomisation evidence on genetically lower LDL-C and dementia risk is null-to-weakly-protective.

Where it lands. There is no convincing population-level signal that statins cause dementia. Rare individual susceptibility to subjective cognitive symptoms cannot be excluded and warrants the same dechallenge/rechallenge approach as for muscle symptoms. The FDA 2012 label is best interpreted as a clinical heads-up; it is not a finding of causation.

31.4 "LDL is fine if HDL is high"

Kernel of truth. HDL-C is inversely associated with cardiovascular risk in essentially every observational dataset ever assembled. The Framingham Offspring data, the EPIC cohort, the AMORIS dataset, the INTERHEART case-control analysis all show that low HDL-C is a robust risk marker.⁶²

Overreach. The inference that raising HDL-C by pharmacological means will reduce cardiovascular events. This hypothesis was tested directly and definitively by the CETP-inhibitor programme. The four major CETP inhibitor outcome trials — ILLUMINATE (torcetrapib, terminated for harm 2006), dal-OUTCOMES (dalcetrapib, Roche's drug, futility 2012), ACCELERATE (evacetrapib, futility 2015), and REVEAL (anacetrapib, modest benefit 2017 but commercial discontinuation) — collectively raised HDL-C substantially (often by 100%+) without producing the expected reduction in cardiovascular events.⁶³ The dal-OUTCOMES failure is directly relevant to the Roche conversation: it is the trial that ended Roche's serious participation in lipid-modifying therapeutics. The REVEAL result was modest and is best explained by the LDL-C reduction also produced by anacetrapib rather than by the HDL-C rise.

The Mendelian-randomisation evidence is consistent. Voight and colleagues, in The Lancet 2012, used genetic variants associated with HDL-C and with LDL-C to ask which was causally related to myocardial infarction. Variants altering LDL-C had the predicted effect on MI risk; variants altering HDL-C did not.⁶⁴ HDL-C is a risk marker, not a causal variable.

Where it lands. HDL-C is a useful risk marker. The "if HDL is high, LDL doesn't matter" inference is false: LDL-C and ApoB remain causal regardless of HDL-C level. The therapeutic strategy of raising HDL-C has been tested and failed. The current research focus has shifted from "HDL-C concentration" to "HDL functionality" (cholesterol efflux capacity, anti-inflammatory function), with mechanistic interest but no current therapeutic translation.

31.5 "Saturated fat is exonerated"

Kernel of truth. The dietary saturated-fat hypothesis has been substantially weakened by post-2010 observational and meta-analytic evidence. The PURE study (Prospective Urban Rural Epidemiology), published by Dehghan and colleagues in The Lancet in 2017, examined dietary patterns in 135,335 participants from 18 countries and reported that higher saturated fat intake was associated with lower total mortality and no association with cardiovascular events.⁶⁵ Earlier meta-analyses by Siri-Tarino and colleagues (2010) and the Cochrane review by Hooper and colleagues (2020, updating 2015) showed at most a modest cardiovascular signal from saturated-fat reduction, much smaller than the 1970s–80s consensus had suggested.⁶⁶ The "diet-heart" hypothesis as originally framed by Ancel Keys, in the form of "saturated fat raises cholesterol raises heart disease", has been seriously revised.

Overreach. Concluding that saturated fat is "exonerated" entirely, or that dietary fat composition is irrelevant to cardiovascular risk. The mainstream cardiology consensus (AHA 2021 dietary guidance, ESC 2021 cardiovascular prevention guidelines) continues to recommend replacing saturated fat with polyunsaturated fat on the basis of the LDL-C effect and the residual signal in meta-analyses.⁶⁷ The PURE result is methodologically contested (food frequency questionnaire data, large cross-country confounding) and its interpretation remains under active debate.

Where it lands. This is genuinely a contested area where the orthodox-heterodox split remains unresolved. The conservative summary is: (i) replacing saturated fat with refined carbohydrate is probably neutral or slightly harmful; (ii) replacing saturated fat with polyunsaturated fat probably modestly reduces cardiovascular events via LDL-C and other mechanisms; (iii) replacing saturated fat with monounsaturated fat (the Mediterranean pattern) probably reduces cardiovascular events through multiple mechanisms; (iv) the overall dietary pattern matters more than any single macronutrient. Critically, this question is not the same question as statin efficacy. Even if dietary saturated fat turned out to be cardiovascularly neutral, the case for pharmacological LDL-lowering in established disease would be unaffected.

31.6 "Mendelian randomisation settles it"

Kernel of truth. Mendelian randomisation (MR) using genetic variants in LDLR, PCSK9, HMGCR, NPC1L1 and other loci provides exceptionally clean evidence that lifelong genetically determined LDL-C is causal for cardiovascular events, with a per-unit effect substantially larger than the per-unit pharmacological effect (because the genetic exposure is lifelong rather than initiated mid-life).⁶⁸ This is one of the strongest pieces of evidence in cardiovascular biology and is what tips the causality argument decisively in favour of the lipid hypothesis.

Overreach. Treating MR as the entirety of the relevant evidence. MR establishes that lower LDL-C causes lower cardiovascular event risk in expectation. It does not by itself answer (i) whether a particular pharmacological agent has off-target effects that change the harm-benefit calculation; (ii) at what age and in whom intervention is optimal; (iii) what the diminishing returns of progressively lower LDL-C look like in practice; (iv) how the population-level cost-effectiveness shakes out. MR is necessary for the causality argument and insufficient for the pharmacotherapy argument.

Where it lands. MR settles the causality question robustly. It does not settle the harm-benefit ledger for any specific pharmacological intervention. The intellectual move "MR therefore statins" is too quick; the correct move is "MR therefore the target is real, and randomised trials of specific agents therefore tell us about specific drugs hitting that real target". The two pieces of evidence are complements.

31.7 "Run-in periods in trials hide real-world tolerability"

Kernel of truth. Multiple major statin trials used placebo (4S, JUPITER) or active-drug run-in periods, which by construction excluded patients who could not tolerate the drug short-term from the randomised population. This means that the trial-level AE rates underestimate real-world tolerability when the drug is prescribed to an unselected population.⁶⁹ As discussed in Chapter 25, ASCOT-LLA's blinded-versus-unblinded reanalysis demonstrates this empirically: AE rates rise substantially when the same drug is given in an unblinded setting.

Overreach. Inferring from the run-in critique that the trials are systematically misleading about efficacy as well as tolerability. Run-in periods affect intent-to-treat efficacy estimates only marginally because the run-in exclusion is typically 5–10 % of screened participants and is largely independent of the efficacy outcome. The CTT efficacy estimates survive sensitivity analyses that adjust for run-in selection.

Where it lands. The run-in critique is correct and should be acknowledged as a real limitation of trial-level tolerability estimates. It does not invalidate the efficacy findings. Population-level prescribing expectations should be tempered: real-world 12-month discontinuation rates of 25–50 % are not surprising and partly reflect the gap between trial and real-world tolerability. For an Anthony-versus-Roche conversation, this is exactly the kind of point that demonstrates substantive engagement with the evidence: it is a legitimate methodological critique that orthodox commentators should concede, and it shifts how one should communicate with patients, without changing the underlying efficacy story.

31.8 "Generic statins are essentially free, so just take them"

Kernel of truth. The marginal cost of generic atorvastatin or rosuvastatin is, in most healthcare systems with generic price competition, on the order of cents per day. The drug-acquisition argument against statin therapy is weak.

Overreach. Inferring that cost-near-zero is the same as benefit-greater-than-zero. The "just take them" argument ignores the genuine questions of (i) NNT in low-risk patients (a treatment with NNT >200 over 5 years is providing genuinely small absolute benefit at the population level, even if the marginal cost is low); (ii) NNH for diabetes, muscle symptoms and rare serious events, which scale per patient-year regardless of drug price; (iii) opportunity cost in the patient's pill burden, daily attention budget, and clinical relationship time; (iv) the patient-autonomy and informed-consent dimension — the patient has the right to a numerate communication of expected harms and benefits, and "it's cheap" is not that communication; (v) the medicalisation dimension — population-level prescribing of pharmaceuticals to people without disease shapes the social meaning of health and shifts both individual and system focus away from non-pharmacological interventions.

Where it lands. The "cheap, so why not" argument is rhetorical rather than substantive. The right question is always "what is the expected net benefit to this patient and is it large enough to be worth their time, attention and (small but nonzero) iatrogenic risk?" In high-risk patients the answer is yes. In low-risk primary-prevention patients the answer is contested and depends on individual values. The drug being inexpensive is a relevant but small input to the decision; it is not, by itself, an argument.

31.9 Synthesis: where these myths leave the substantive debate

If one strips away the polemic from both sides, the substantive disagreements in the cholesterol-statin debate are reducible to a small number of empirical questions, each of which is partly resolved and partly open.

1. Is LDL-C causal for ASCVD? Yes, robustly, by MR + trial + mechanism convergence. (Closed.)
2. Do statins reduce cardiovascular events in established CHD? Yes, robustly, with NNT 15–30 over 5 years. (Closed.)
3. Do statins reduce cardiovascular events in high-risk primary prevention (>10 % 10-year ASCVD risk)? Yes, with NNT 30–60 over 5 years. (Closed.)
4. Do statins reduce events in low-risk primary prevention (<5 % 10-year ASCVD risk)? Yes in relative terms but at NNT >100 over 5 years, where the benefit-harm ledger is genuinely close. (Contested; partly resolved by shared decision-making.)
5. What is the real prevalence of statin-attributable adverse effects? Muscle: 1–5 % pharmacologically specific, with much larger nocebo overlay. Diabetes: 9 % relative, on-target. Cognitive: no population signal. Rhabdomyolysis: rare. (Closed at the population level; rare individual susceptibility cannot be excluded.)
6. Is the primary-prevention case strong in the over-75 population without other risk factors? Unknown, awaiting STAREE 2026 and PREVENTABLE late-2026. (Open.)
7. Do non-statin LDL-lowering agents (ezetimibe, PCSK9 inhibitors, bempedoic acid, inclisiran) provide cardiovascular benefit proportional to LDL-C lowering? Yes, validated by IMPROVE-IT, FOURIER, CLEAR Outcomes; inclisiran ORION-4 read-out July 2026. (Largely closed; inclisiran specific outcome still pending.)
8. Is Lp(a) a separate causal lipoprotein for which targeted therapy is justified? MR says yes; outcome trials of pelacarsen, olpasiran and lepodisiran are pending. (Open.)

This list is, I think, the right summary of the field. It is the list I would put in front of a Roche scientist and ask which items they thought I had wrong. It is also the list against which any specific heterodox claim should be evaluated: which of items 1–8 is the claim about, and what is the heterodox argument adding or subtracting?

Footnotes — Part V

Part VI — Adjudication and Calibrated Position

Chapter 32. Bradford-Hill applied

Bradford Hill's nine "viewpoints" on causal inference, set out in 1965 in his Presidential Address to the Royal Society of Medicine¹, remain the cleanest framework for evaluating whether an exposure–outcome relationship is causal. They are not a checklist — Hill himself was emphatic that they are aids to judgment, not algorithmic criteria — but they are useful here because the cholesterol-CHD relationship is one of the most thoroughly examined in medicine.

Strength of association

Untreated familial hypercholesterolaemia produces a 20-fold elevation in coronary risk by age 50.² Lp(a) >150 nmol/L approximately doubles risk.³ At population level the log-linear relationship between LDL-C and major vascular events is shallow per unit but cumulative over lifetime: a 1 mmol/L lifetime-lower LDL-C exposure correlates with ~50% lower CHD risk in Mendelian-randomisation analyses, compared with ~22% per mmol/L of pharmacological reduction in the CTT meta-analyses — a divergence explained by exposure duration, which itself supports rather than undermines causality.⁴

Consistency

The relationship has been observed across high-income and middle-income populations, men and women, age strata from 35 to 85, and four mechanistically distinct LDL-lowering drug classes (statins, ezetimibe, PCSK9 monoclonal antibodies, bempedoic acid). The PCSK9-LoF Mendelian-randomisation finding is consistent with the FOURIER pharmacological finding. The HMGCR-variant MR is consistent with the statin RCTs.

Specificity

ApoB-containing lipoprotein retention in arterial intima is biochemically specific. The atherogenic particles are LDL, Lp(a), and remnant lipoproteins; HDL is not retained. The specificity of the cholesterol-CHD relationship strengthened, rather than weakened, as the mechanism was refined.

Temporality

Genetic exposure precedes disease by definition; this is the structural advantage of MR. LDL-lowering interventions precede event reduction in every well-conducted trial. There is no plausible reverse-causation pathway.

Biological gradient

The dose-response is one of the cleanest in cardiovascular medicine: each 1 mmol/L of LDL-C reduction produces approximately 22% reduction in major vascular events, across populations, drug classes, baseline LDL strata, and follow-up durations.⁵

Plausibility

The mevalonate-LDLR-foam-cell-plaque pathway is biochemically and histologically delineated end-to-end. The remaining uncertainty is about magnitude of contribution from inflammation, plaque vulnerability biology, and Lp(a) — not whether the LDL pathway operates.

Coherence

Biology (Brown & Goldstein 1985 Nobel), genetics (Abifadel 2003, Cohen & Hobbs 2006), pharmacology (Endo 1976 mevastatin discovery), epidemiology (Framingham 1948–), and clinical trials (4S 1994 onward) all align on the same model. There is no major coherent counter-model that explains the same data better.

Experiment

The intervention-reversal criterion is the strongest form of evidence. Statins, ezetimibe, PCSK9 inhibitors, and bempedoic acid all lower LDL and all reduce events. Genetic loss-of-function carriers of PCSK9 have lifelong low LDL and lifelong low CHD. The CETP-inhibitor class lowered LDL only modestly while having off-target effects and failed; this does not refute the LDL hypothesis any more than a failed antibiotic refutes the germ theory.

Analogy

Pathogen-disease analogies (lipid particles as the "atherogenic agent," macrophages as the "responders") are not load-bearing here because we have direct evidence.

Verdict on Bradford-Hill

LDL-C (more precisely, ApoB-containing lipoproteins) is causally related to atherosclerotic cardiovascular disease. This is not the contested claim. The contested claims are quantitative — how much benefit, in whom, for what duration, at what harm.

Chapter 33. NNT and NNH by risk stratum — the practical table

The single most important table in this document. Values are approximate and synthesised from CTT meta-analyses, individual trials, and 2024–2026 reappraisals; they should be read as orders of magnitude, not point estimates.

The two columns that most often go missing in public conversation are NNH and the evidence weight. Restoring them changes the conversation because they show that the disagreement between camps is not "do statins work" — both camps largely accept that they do for somebody — but "for whom does the benefit-harm calculation clearly favour treatment, and for whom is the answer genuinely contested?" The two disputed cells are low-risk primary prevention and healthy elderly without ASCVD pending STAREE/PREVENTABLE. Most of the public argument is about those two cells, even when participants don't realise it.

Chapter 34. The calibrated landing position

What is essentially settled

LDL-C (more precisely, ApoB-containing lipoproteins) is causally and cumulatively related to atherosclerotic cardiovascular disease. The evidence comes from genetics (FH, Mendelian randomisation), epidemiology, mechanism, pharmacology, and four mechanistically distinct intervention classes. The strength of evidence in 2026 is higher than for almost any other modifiable risk factor in medicine.

Statins reduce major vascular events in patients with established ASCVD, in patients with diabetes plus risk factors, in patients with familial hypercholesterolaemia, and in high-risk primary prevention. In these populations the magnitude of benefit clearly exceeds the magnitude of identifiable harm. Conversation with a Roche-employed friend, or anyone else, that pretends otherwise is unwinnable on the evidence.

What is reasonably contested

The absolute benefit of statins in low-risk primary prevention is small (ARR 0.4–1.6% over 5 years for major endpoints). Reasonable people can argue that this is worth taking; reasonable people can argue that it isn't. The disagreement is not about whether the drugs work — they do — but about whether the absolute magnitude justifies routine prescription in patients whose 10-year ASCVD risk is low, particularly given (a) the run-in periods in pivotal trials that may understate real-world tolerability and (b) the genuine if modest new-onset diabetes signal.

Statin therapy in the healthy elderly without prior ASCVD is empirically contested and will be substantially clarified by STAREE (late 2025 / 2026) and PREVENTABLE (December 2026). In this population, the legitimate questions are: does lifetime-cumulative-LDL-exposure logic apply when the remaining lifetime is short; does the dementia/disability composite primary endpoint reveal benefits or harms not captured in MACE-focused trials; do the population-specific harms (diabetes, frailty interactions, polypharmacy interactions) attenuate or eliminate the CV benefit? PREVENTABLE in particular will be the most consequential single trial for this question.

The CTT IPD access controversy is a legitimate methodological complaint regardless of one's view on statin efficacy. The substantive answer the CTT analyses produce may be approximately correct; the epistemic process is closed and unreproducible. A mature consensus camp acknowledges this; a mature heterodox camp does not over-extrapolate from it.

Industry capture has demonstrably distorted specific trials (ENHANCE delayed publication; JUPITER early-stopping decision; bococizumab development trajectory) and specific analyses (the CTT publication-only data access). The core LDL-causality and statin-benefit findings nevertheless survive the COI critique because they are confirmed by investigator-initiated trials (4S started life largely independently of Merck), post-patent generic prescriptions (no commercial incentive to maintain the consensus), and four mechanistically distinct drug classes targeting the same pathway.

What remains genuinely uncertain

• Whether very low LDL targets (<1.4 mmol/L, <55 mg/dL) confer additional benefit at low ARR cost, and whether very low LDL has consequences not yet detected at population level. FOURIER-OLE through 7+ years is reassuring but not definitive.
• Whether Lp(a)-targeted therapy will produce CV outcome benefit proportionate to LDL-targeted therapy, or whether Lp(a)'s pathology is sufficiently different that the translational link will be weaker. Lp(a)HORIZON, OCEAN(a), and ACCLAIM-Lp(a) readouts 2025–2027.
• Whether new-onset-diabetes signal is fully on-target (HMGCR-mediated, MR-supported) or whether some statin molecules have molecule-specific diabetogenic effects.
• Whether modern adherence-corrected effectiveness estimates align with trial-derived efficacy estimates after run-in adjustments. The Schubert 2025 adherence meta-analysis suggests yes; the Byrne 2022 reappraisal suggests less.
• Whether the AHA PREVENT equations (2024) correctly recalibrate risk for population-level decisions, and what the implications are for the ~15.8 million US adults reclassified.

The honest one-sentence position

The case for statin therapy in secondary prevention, familial hypercholesterolaemia, and high-risk primary prevention is overwhelming; the case in low-risk primary prevention and the healthy elderly is empirically thinner, ethically more contestable, and properly resolved by shared decision-making informed by patient-specific ARR/NNH calculations rather than by population-level mandate.

That sentence will not satisfy a partisan of either camp. That is the point.

Chapter 35. What would change my mind — pre-committed updating

Pre-commitment to falsification is the cleanest defence against motivated reasoning. The following updates are stated in advance, not retrofitted to whatever the evidence eventually shows.

If I would shift toward the heterodox camp

I would weaken my confidence in the population-level statin-benefit thesis if:

1. STAREE reads out negative or null on co-primary endpoints, particularly disability-free survival. A null result in 9,971 healthy elderly would substantially undermine the lifetime-cumulative-LDL extrapolation in this population.

2. PREVENTABLE reads out negative on its dementia/disability primary endpoint despite achieving the targeted LDL reduction. This would either reveal a previously unappreciated cognitive harm, or — more methodologically interesting — would suggest that LDL reduction in late life does not translate to functional benefit even when CV events are reduced.

3. An independent reanalysis of CTT IPD (if Oxford CTSU ever opens it) showed that major-vascular-event reductions are not robust to alternative trial-inclusion criteria or to adjustment for the run-in selection effect.

4. A high-quality, blinded N-of-1 replication of SAMSON in 1000+ patients identified a specific-pharmacological muscle-harm phenotype affecting >10% of patients, not the <2% the current trials and SAMSON suggest.

5. A primary-prevention trial in low-risk patients (10-year ASCVD <7.5% by PREVENT) showed null effect on hard endpoints, falsifying the down-extension of the CTT log-linear relationship.

If I would shift toward the consensus camp's strongest form

I would strengthen my confidence (already high) if:

1. STAREE and PREVENTABLE both read out positive on CV endpoints, particularly with reassuring or absent dementia/disability harms. This would close the elderly-statin debate.

2. Lp(a)HORIZON, OCEAN(a), and ACCLAIM-Lp(a) all read out positive on hard CV endpoints. This would confirm that targeting any genetically-validated atherogenic lipoprotein produces CV benefit, sealing the lipid-causality case.

3. ORION-4 reads out strongly positive for inclisiran on hard CV endpoints, validating the siRNA approach at population scale.

4. A long-term cognitive safety study (extending FOURIER-OLE-style methodology to 10+ years) confirms no signal of accelerated dementia or cognitive decline.

Conversion criteria flagged in advance

If items 1 and 2 from the consensus-strengthening list both hold, the rational position is unambiguous statin/non-statin LDL lowering for essentially all adults whose 10-year ASCVD risk exceeds some threshold to be determined empirically by NNT/NNH analysis.

If items 1 and 2 from the heterodox-shifting list both hold, the rational position is significantly more conservative routine prescription, with shared decision-making becoming the standard of care for most primary-prevention contexts. This would be a substantial shift from current guidelines and should be acknowledged as such.

If neither set of conditions holds — the most likely outcome at this writing — the calibrated landing position above remains the rational stance.

Footnotes for Part VI

Appendix A — Debate Playbook (Conversation-ready Cards)

For Anthony's actual conversation. Quick reference, not new claims. Each card states a likely opening from the friend, the strongest steelman response, and the calibrated counter-position.

Card 1 — "The evidence for statins is overwhelming."

Steelmanned version (concede this): In secondary prevention, FH, and high-risk primary prevention it genuinely is. RCT, MR, mechanism, four drug-class convergence, post-patent generic confirmation. This is not a contested space.

Calibrated counter: "Agreed for those populations. The contested space is low-risk primary prevention and the healthy elderly — that's where STAREE and PREVENTABLE in 2025–2026 will substantially clarify the question. Where do you land on the 50% reduction in statin-eligible US adults from the 2024 AHA PREVENT equations?"

The move: don't argue against the strong cases. Move the conversation to the genuinely contested cases.

Card 2 — "LDL is not really causal, it's just a marker."

Steelmanned version (concede the kernel): Cholesterol mass per se is an imperfect proxy; ApoB particle number is more biologically precise; observational confounding does exist; some individuals with high LDL never develop CHD.

Calibrated counter: "ApoB-containing lipoproteins are causal. Cohen and Hobbs 2006 showed PCSK9 loss-of-function carriers have 28% lower LDL and 88% lower CHD over 15 years — that's MR, not observation. Ference's lifetime-exposure work, the 2017 EAS consensus, Brown and Goldstein's Nobel-prize biochemistry. Four drug classes hitting the same target reduce events. The question is not 'is LDL causal' but 'in whom does pharmacological reduction produce net benefit'."

Card 3 — "Statins cause widespread muscle damage."

Steelmanned version (concede the kernel): Observational reports of 10–30% rates. Real cases of rhabdomyolysis (cerivastatin/Baycol withdrawn 2001). Pharmacogenomic susceptibility (SLCO1B1*5, SEARCH 2008). CoQ10 depletion is a real biochemical phenomenon. The trials' run-in periods selected out susceptible patients.

Calibrated counter: "SAMSON, NEJM 2020 — N-of-1 with statin/placebo/no-tablet months in 60 patients who already self-reported intolerance. Found 90% of symptom intensity attributed to nocebo, only ~10% specifically pharmacological. ASCOT-LLA blinded vs open-label phase showed the same pattern — muscle symptoms rose four-fold once patients knew which treatment they were on. That doesn't mean no real harm exists — pharmacogenomic susceptibility and idiosyncratic reactions are real. It means most muscle complaints in routine practice are not specifically the drug, and the clinical implication is rechallenge with a different molecule at lower dose before abandoning therapy."

Card 4 — "Industry funded all the trials so you can't trust them."

Steelmanned version (concede the history): ENHANCE delay (Merck/Schering-Plough 2007–08). Vioxx parallel. Lipitor as biggest-selling drug ever. CTT IPD access controversy. Ghost-writing case studies (Healy, Sismondo). Run-in periods that bias AE reporting. JUPITER's early-stopping decision while AstraZeneca controlled the data. The Bayer/cerivastatin/Baycol withdrawal story. These all happened.

Calibrated counter: "Industry capture has demonstrably distorted specific trials and analyses. The core finding survives that critique because: (a) 4S was investigator-initiated with Merck funding only; (b) statins have been generic for over a decade with no commercial incentive to maintain the consensus; (c) Mendelian-randomisation evidence (genetic variants) is by definition uncontaminated by industry; (d) four mechanistically distinct drug classes (statins, ezetimibe, PCSK9-mAb, bempedoic acid) hit the same target through different molecules with different sponsors and produce convergent results. Industry capture is real and worth fighting. But it's the wrong tool to dismiss the core finding."

Card 5 — "What about the residual cardiovascular risk after statin therapy?"

Steelmanned version (concede the gap): Even on maximum-intensity statin therapy, residual CV event risk is substantial. PCSK9 add-on trials show meaningful additional benefit, suggesting LDL targets in current guidelines are not at the floor. Lp(a) is largely untouched by statins. Inflammation contributes (CANTOS, colchicine).

Calibrated counter: "Right — and this is where the action is for the next decade. Inclisiran outcomes (ORION-4, July 2026 readout). The Lp(a) trials (pelacarsen, olpasiran, lepodisiran). Colchicine for inflammation. CETP-inhibitor revival via obicetrapib? The residual-risk question doesn't refute LDL causality; it tells you LDL alone is necessary but not sufficient."

Card 6 — "PREVENT changes everything."

Steelmanned version (concede the implication): The 2024 AHA PREVENT risk equations recalibrate ASCVD risk and reclassify approximately half of US adults to lower categories. Approximately 15.8 million adults may no longer meet statin-eligibility thresholds under the old pooled-cohort equations.

Calibrated counter: "Significant guideline upheaval is coming. The interesting question is whether this represents better calibration (the old equations over-estimated risk and over-prescribed) or under-calibration (population CV risk really is lower now because of better blood pressure control, smoking declines, and screening, and the calibration is appropriately catching that). Both views are defensible. What it doesn't change: secondary prevention, FH, and high-risk primary prevention all remain unambiguous indications."

Card 7 — "What's your honest position?"

Calibrated direct answer: "In secondary prevention, familial hypercholesterolaemia, and high-risk primary prevention, the case is overwhelming and I take statins seriously as standard of care. In low-risk primary prevention and the healthy elderly without ASCVD, I think absolute benefit is small enough that shared decision-making with patient-specific NNT/NNH calculations should be the standard rather than population-level mandate. I'd be substantially updated by STAREE in 2025 and PREVENTABLE in late 2026 — those will be the most consequential trials of the next two years for the contested cells of the matrix."

That's the line that gets respect from a Roche scientist. It signals you've read the actual evidence, you know what the named trials are, you know what's being adjudicated, and you have a calibrated position — not a tribal one.

Card 8 — Roche-specific aside

If the conversation goes meta about industry, you can mention specifically: "Roche actually got out of the LDL game in 2012 when dal-OUTCOMES failed for dalcetrapib. The CETP class is essentially extinct on the back of ILLUMINATE, dal-OUTCOMES, ACCELERATE, and REVEAL. Roche's lipid commercial alignment now is diagnostics — measuring LDL, ApoB, Lp(a) accurately — which dovetails with the consensus camp's measurement-and-target philosophy rather than with statin mass-prescription. So your interests there are not what someone at Pfizer or Amgen would have. Where do you think the diagnostic market is going with Lp(a) becoming actionable?"

That move (a) shows you know Roche's actual portfolio, which is a credibility signal, and (b) moves the conversation onto ground where you and your friend share interest in the science rather than position from camps.

Appendix B — Erratum, Methods Note, and Bibliography Index

Erratum, in response to the external review panel

The external review panel (lipidologist, trialist, methodologist, pharmacology lecturer, clinical pharmacist) scored this document 74/100 and flagged the following items that an attentive reader should keep in mind. These are not silently fixed; they stand as acknowledged limitations.

On steelmanning balance. The CoQ10-depletion hypothesis in Chapter 1/21 is dispatched on the basis of two under-powered supplementation RCTs (Taylor 2015; Banach 2015), which is the strongest available counter-evidence but not the strongest formulation of the hypothesis. The strong-form CoQ10 thesis would predict harm independent of replacement-trial power, and the document could have engaged this version more squarely. JUPITER in Chapter 7 is summarised in a way that emphasises the early-stopping critique and underweights the significant all-cause-mortality reduction that is the consensus camp's strongest single piece of JUPITER evidence. The Bradford-Hill "strength of association" criterion in Chapter 32 leads with familial hypercholesterolaemia (a 20-fold effect) rather than with the much smaller per-mmol/L population-level effect that most of the document's hard claims actually rest on.

On the NNT/NNH-by-risk-stratum table (Chapter 33). The table is presented with bare integers where the underlying evidence supports ranges, not point estimates. Readers should treat the values as orders of magnitude rather than precise predictions. A more honest version of the table would widen the cells to intervals (e.g. "~50–80" rather than "~50") and add an evidence-weight annotation per cell distinguishing strong-trial estimates from extrapolated ones.

On the asymmetry in Chapter 35. The "what would change my mind" chapter lists conditions that would shift the document toward each camp but does not list conditions under which the LDL-causality core itself would be invalidated. The implicit assumption is that LDL causality is essentially settled, which is the document's substantive position. Stating that assumption explicitly rather than embedding it in the falsification architecture would have been more rigorous.

On pharmacology gaps. The document covers SLCO1B1 (the SEARCH pharmacogenomic finding) but does not engage CYP3A5*3, COQ2 polymorphisms, or ABCG2 in the SAMS discussion. The asymmetric-AE-capture chapter does not engage the heterodox critique of MedDRA coding choices in trial AE classification.

On the bibliography. The master document's citations are present as inline Markdown footnotes throughout each chapter file. A consolidated, deduplicated, end-of-document Vancouver bibliography in formal numbered form is not produced. The Bibliography Index below partially compensates by listing the canonical references most cited throughout, but is not a substitute.

Methods note (compact form — full version in 02-Methods-and-orchestration.md)

Document produced by multi-agent orchestration in approximately 45 minutes of wall-clock time. Eight grounding web searches preceded research dispatch. Five parallel research agents produced approximately 65,000 words of footnoted research content from briefs of 800–1,200 words each. Synthesis layer (front matter, adjudication chapter, debate playbook, this erratum) written by orchestrator directly. External review panel ran as a separate independent agent with no role in the writing. Approximately nine [CITATION NEEDED] flags remain unresolved and are transparent rather than silently filled.

Bibliography Index — canonical references most cited throughout

This is an index, not a full Vancouver bibliography. References are listed by topic rather than alphabetically, to support a reader looking up the source for a specific claim.

Foundational biochemistry - Brown MS, Goldstein JL. A receptor-mediated pathway for cholesterol homeostasis. Science. 1986;232(4746):34-47. (Nobel lecture review.) - Endo A. The discovery and development of HMG-CoA reductase inhibitors. J Lipid Res. 1992;33(11):1569-82. (PMID: 1464741.) - Abifadel M, Varret M, Rabès JP, et al. Mutations in PCSK9 cause autosomal dominant hypercholesterolemia. Nat Genet. 2003;34(2):154-6. (PMID: 12730697.) - Cohen JC, Boerwinkle E, Mosley TH Jr, Hobbs HH. Sequence variations in PCSK9, low LDL, and protection against coronary heart disease. N Engl J Med. 2006;354(12):1264-72. (PMID: 16554528.)

LDL causality / Mendelian randomization - Ference BA, Yoo W, Alesh I, et al. Effect of long-term exposure to lower LDL cholesterol beginning early in life on the risk of coronary heart disease: a Mendelian randomization analysis. J Am Coll Cardiol. 2012;60(25):2631-9. (PMID: 23083789.) - Ference BA, Ginsberg HN, Graham I, et al. 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(32):2459-72.

First statin era - Scandinavian Simvastatin Survival Study Group (4S). Randomised trial of cholesterol lowering in 4444 patients with coronary heart disease. Lancet. 1994;344(8934):1383-9. (PMID: 7968073.) - Shepherd J, Cobbe SM, Ford I, et al. (WOSCOPS). Prevention of coronary heart disease with pravastatin in men with hypercholesterolemia. N Engl J Med. 1995;333(20):1301-7. (PMID: 7566020.) - Sacks FM, Pfeffer MA, Moye LA, et al. (CARE). The effect of pravastatin on coronary events after MI in patients with average cholesterol levels. N Engl J Med. 1996;335(14):1001-9. (PMID: 8801446.) - LIPID Study Group. Prevention of cardiovascular events and death with pravastatin in patients with coronary heart disease and a broad range of initial cholesterol levels. N Engl J Med. 1998;339(19):1349-57. (PMID: 9841303.) - Downs JR, Clearfield M, Weis S, et al. (AFCAPS/TexCAPS). Primary prevention of acute coronary events with lovastatin in men and women with average cholesterol levels. JAMA. 1998;279(20):1615-22. (PMID: 9613910.)

Mega-trial era - Heart Protection Study Collaborative Group. MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin in 20,536 high-risk individuals. Lancet. 2002;360(9326):7-22. (PMID: 12114036.) - Sever PS, Dahlöf B, Poulter NR, et al. (ASCOT-LLA). Prevention of coronary and stroke events with atorvastatin in hypertensive patients. Lancet. 2003;361(9364):1149-58. (PMID: 12686036.) - Cannon CP, Braunwald E, McCabe CH, et al. (PROVE-IT TIMI 22). Intensive versus moderate lipid lowering with statins after acute coronary syndromes. N Engl J Med. 2004;350(15):1495-504. (PMID: 15007110.) - LaRosa JC, Grundy SM, Waters DD, et al. (TNT). Intensive lipid lowering with atorvastatin in patients with stable coronary disease. N Engl J Med. 2005;352(14):1425-35. (PMID: 15755765.) - Ridker PM, Danielson E, Fonseca FA, et al. (JUPITER). Rosuvastatin to prevent vascular events in men and women with elevated CRP. N Engl J Med. 2008;359(21):2195-207. (PMID: 18997196.)

CTT meta-analyses - 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(9753):1670-81. (PMID: 21067804.) - Cholesterol Treatment Trialists' (CTT) Collaboration. Efficacy and safety of statin therapy in older people: a meta-analysis of individual participant data from 28 randomised controlled trials. Lancet. 2019;393(10170):407-15. (PMID: 30712900.)

Non-statin LDL lowering - Kastelein JJ, Akdim F, Stroes ES, et al. (ENHANCE). Simvastatin with or without ezetimibe in familial hypercholesterolemia. N Engl J Med. 2008;358(14):1431-43. (PMID: 18376000.) - Baigent C, Landray MJ, Reith C, et al. (SHARP). The effects of lowering LDL cholesterol with simvastatin plus ezetimibe in patients with chronic kidney disease. Lancet. 2011;377(9784):2181-92. (PMID: 21663949.) - Cannon CP, Blazing MA, Giugliano RP, et al. (IMPROVE-IT). Ezetimibe added to statin therapy after acute coronary syndromes. N Engl J Med. 2015;372(25):2387-97. (PMID: 26039521.)

PCSK9 monoclonal antibodies - Sabatine MS, Giugliano RP, Keech AC, et al. (FOURIER). Evolocumab and clinical outcomes in patients with cardiovascular disease. N Engl J Med. 2017;376(18):1713-22. (PMID: 28304224.) - Schwartz GG, Steg PG, Szarek M, et al. (ODYSSEY OUTCOMES). Alirocumab and cardiovascular outcomes after acute coronary syndrome. N Engl J Med. 2018;379(22):2097-107. (PMID: 30403574.) - O'Donoghue ML, Giugliano RP, Wiviott SD, et al. (FOURIER-OLE). Long-term evolocumab in patients with established atherosclerotic cardiovascular disease. Circulation. 2022;146(15):1109-19. https://www.ahajournals.org/doi/10.1161/CIRCULATIONAHA.122.061620

Inclisiran - Raal FJ, Kallend D, Ray KK, et al. (ORION-9). Inclisiran for the treatment of heterozygous familial hypercholesterolemia. N Engl J Med. 2020;382(16):1520-30. - Ray KK, Wright RS, Kallend D, et al. (ORION-10 and ORION-11). Two phase 3 trials of inclisiran in patients with elevated LDL cholesterol. N Engl J Med. 2020;382(16):1507-19. - ORION-4 outcomes trial: NCT03705234, primary endpoint readout July 2026.

Bempedoic acid - Nissen SE, Lincoff AM, Brennan D, et al. (CLEAR Outcomes). Bempedoic acid and cardiovascular outcomes in statin-intolerant patients. N Engl J Med. 2023;388(15):1353-64. https://www.nejm.org/doi/full/10.1056/NEJMoa2215024

SAMS / nocebo - Howard JP, Wood FA, Finegold JA, et al. (SAMSON). Side effect patterns in a crossover trial of statin, placebo, and no treatment. N Engl J Med. 2021;383:2182-2184 / J Am Coll Cardiol 2021. https://www.jacc.org/doi/10.1016/j.jacc.2020.10.056 - Gupta A, Thompson D, Whitehouse A, et al. (ASCOT-LLA blinded vs open-label). Adverse events associated with unblinded, but not with blinded, statin therapy in ASCOT-LLA. Lancet. 2017;389(10088):2473-81. (PMID: 28476290.)

New-onset diabetes - Sattar N, Preiss D, Murray HM, et al. Statins and risk of incident diabetes: a collaborative meta-analysis of randomised statin trials. Lancet. 2010;375(9716):735-42. (PMID: 20167359.) - Swerdlow DI, Preiss D, Kuchenbaecker KB, et al. HMG-coenzyme A reductase inhibition, type 2 diabetes, and bodyweight: evidence from genetic analysis and randomised trials. Lancet. 2015;385(9965):351-61. (PMID: 25262344.)

Atherosclerosis biology - Williams KJ, Tabas I. The response-to-retention hypothesis of early atherogenesis. Arterioscler Thromb Vasc Biol. 1995;15(5):551-61. (PMID: 7749869.) - Ridker PM, Everett BM, Thuren T, et al. (CANTOS). Antiinflammatory therapy with canakinumab for atherosclerotic disease. N Engl J Med. 2017;377(12):1119-31. (PMID: 28845751.) - Nidorf SM, Fiolet ATL, Mosterd A, et al. (LoDoCo2). Colchicine in patients with chronic coronary disease. N Engl J Med. 2020;383(19):1838-47. (PMID: 32865380.)

Heterodox critique landmarks - Ravnskov U. The Cholesterol Myths. NewTrends Publishing; 2000. - Diamond DM, Ravnskov U. How statistical deception created the appearance that statins are safe and effective in primary and secondary prevention of cardiovascular disease. Expert Rev Clin Pharmacol. 2015;8(2):201-10. (PMID: 25655639.) - DuBroff R, de Lorgeril M. Cholesterol confusion and statin controversy. World J Cardiol. 2015;7(7):404-9. (PMID: 26225201.) - Byrne P, Demasi M, Jones M, Smith SM, O'Brien KK, DuBroff R. Evaluating the association between low-density lipoprotein cholesterol reduction and relative and absolute effects of statin treatment: a systematic review and meta-analysis. JAMA Intern Med. 2022;182(5):474-81.

Pending trials (2024–2026) - STAREE — Atorvastatin 40mg in 9,971 Australians ≥70, dual primary endpoints, results late 2025/2026. https://www.ahajournals.org/doi/10.1161/JAHA.124.036357 - PREVENTABLE — Atorvastatin 40mg in 20,000 US adults ≥75, primary endpoint survival free of new dementia or persistent disability, expected completion December 2026. https://www.preventabletrial.org/ - ORION-4 — Inclisiran outcomes, primary readout July 2026. - Lp(a)HORIZON (pelacarsen) — readout mid-2025. - OCEAN(a) (olpasiran) — readout 2027. - ACCLAIM-Lp(a) (lepodisiran) — enrolling.

Bradford-Hill foundational - Hill AB. The environment and disease: association or causation? Proc R Soc Med. 1965;58(5):295-300. (PMID: 14283879.)

Glossary (abbreviated)

• ApoB — Apolipoprotein B, the major structural protein of all atherogenic lipoproteins (LDL, VLDL, IDL, Lp(a)). One ApoB per particle.
• ARR — Absolute risk reduction.
• CTT — Cholesterol Treatment Trialists' Collaboration, Oxford CTSU.
• CHD — Coronary heart disease.
• FH — Familial hypercholesterolaemia.
• HMGCR — 3-Hydroxy-3-methylglutaryl-CoA reductase, the rate-limiting enzyme statins inhibit.
• IPD — Individual participant data.
• LDL-C — Low-density lipoprotein cholesterol.
• Lp(a) — Lipoprotein(a), genetically determined LDL-like particle with apo(a).
• MACE — Major adverse cardiovascular events.
• MR — Mendelian randomisation.
• NNT — Number needed to treat.
• NNH — Number needed to harm.
• PCSK9 — Proprotein convertase subtilisin/kexin type 9; degrades LDLR.
• RCT — Randomised controlled trial.
• RR / HR — Relative risk / Hazard ratio.
• SAMS — Statin-associated muscle symptoms.
• SLCO1B1 — Gene encoding the OATP1B1 hepatic transporter; *5 polymorphism associated with simvastatin myopathy.