The particle that carries cholesterol into an artery wall, not the cholesterol itself, drives heart disease — and that fact is about as settled as anything in nutrition and cardiology gets. The particle is tagged by a protein called apolipoprotein B (apoB): one apoB molecule rides on every atherogenic particle, so counting apoB counts the particles. Lower the count and you lower risk, in proportion to how much you lower it and how long you keep it down. The open questions are narrower than the headlines suggest: which number to read off a lab report, how much a given drop buys, and whether the way you lower it — food versus a pill — changes the payoff. This page walks that chain, from the causal model (LDL ApoB and Cumulative Exposure) to the decision.
apoB-containing particles cause atherosclerosis, and the evidence is unusually settled
The 2017 European Atherosclerosis Society consensus tested the LDL-heart-disease link against every major line of evidence at once — rare genetic mutations, prospective cohorts, Mendelian randomization (using inherited gene variants as a natural randomized experiment), and the LDL-lowering drug trials. It reached a verdict it stated in one sentence: «Consistent evidence from numerous and multiple different types of clinical and genetic studies unequivocally establishes that LDL causes ASCVD.» (Ference et al., 2017) The causal agent is not cholesterol mass but the apoB-containing particles that carry it — LDL, plus VLDL and its remnants, IDL, and Lp(a). (Ference et al., 2017)
Concordance makes the verdict strong: the natural genetic experiment and the intervention trials point the same way. This is the rare case where a blood marker’s transmission to the hard outcome is itself proven — the opposite of the usual surrogate, where a number moves and patients do not benefit. -> LDL ApoB and Cumulative Exposure, Surrogate Outcomes
The dose is cumulative: how low, times how long
Risk tracks the area under your apoB curve over a lifetime — level multiplied by years — not today’s snapshot. A modest reduction held for decades can beat a larger one started late. The drug trials show the per-unit face of this: each additional 1 mmol/L drop multiplies on the last, so more and longer both pay. -> LDL ApoB and Cumulative Exposure
The mechanism is agnostic to how you lower the particle count, but only under two conditions the consensus states plainly. The cholesterol drop must reflect a real drop in particle number, and the method must carry no competing harm on another pathway. (Ference et al., 2017) Those two provisos are not fine print — they separate a drug-lowered number from a diet-lowered one, as the rest of this page shows.
Count the particles: apoB beats LDL-C where the two diverge
For most people, standard LDL cholesterol (LDL-C) tracks particle count closely and is a fine number to watch. It fails in one identifiable group: the insulin-resistant, hypertriglyceridemic, metabolic-syndrome or diabetic person, whose small dense LDL packs more particles into a given amount of cholesterol. There, LDL-C under-states the real particle burden, and apoB — or non-HDL-C (total cholesterol minus HDL, already on every lipid panel) — is the number to trust. (Ference et al., 2017) -> LDL ApoB and Cumulative Exposure
The direct evidence is large and points three ways to the same answer. In ~430,000 people, entering the lipids together, only apoB independently predicted heart attack (adjusted HR 1.27 per standard deviation, 95% CI 1.15-1.40); LDL-C and triglycerides fell to non-significant once apoB was held fixed. (Marston et al., 2022) Multivariable Mendelian randomization reaches the same place genetically — apoB retained a robust effect (OR 1.92 per SD, 1.31-2.81) while the LDL-C estimate reversed to null. (Richardson et al., 2020) And an agnostic genetic search over 30 lipid measures, from a group with no overlap with the first two, picked apoB out on its own. (Zuber et al., 2020)
Read the convergence with one caution: Richardson’s MR lists Ference — author of the causal consensus above — as a co-author and shares an author with Zuber, so it corroborates that consensus rather than testing it afresh. Marston’s biobank-plus-RCT leg and Zuber’s non-overlapping group are the genuinely independent witnesses — two routes agreeing, not three unrelated ones.
Read this correctly. None of it says LDL-C is causally inert — apoB never travels without cholesterol, which is why lowering LDL-C works. It says apoB is the necessary element and the better number to measure and target where the two come apart. Current US guidance now agrees, adding an apoB check for exactly this group. -> LDL Lowering and Cardiovascular Events
One live disagreement: particle number or cholesterol content?
Marston makes a finer claim: particle number carries the risk, independent of the cholesterol each particle holds. An independent Mendelian-randomization study (Helgadottir) separated the two using discordant gene variants and reaches the opposite: CAD risk tracks the cholesterol content (non-HDL-C), not the particle count. Helgadottir attributes the apoB signal to confounding of two nearly interchangeable traits. (Helgadottir et al., 2022) This does not touch the settled verdict above — both camps agree apoB and non-HDL-C beat LDL-C in the discordant group. It contests only which of those two to prefer as the secondary target, and it is genuinely unresolved. The practical reading: non-HDL-C is free on every panel and captures most of what a separate apoB draw would. -> ApoB Particle Number vs Cholesterol Content
Lowering the number with a drug cuts hard events by about a fifth per unit, with no floor
The statin trials settle the magnitude. Pooling 26 trials and 169,138 people, each 1.0 mmol/L (about 39 mg/dL) reduction in LDL-C cut major vascular events by ~22% (RR 0.78, 95% CI 0.76-0.80) and all-cause mortality by 10% (RR 0.90, 0.87-0.93). (Cholesterol Treatment Trialists’ Collaboration, 2010) Two features carry the decision:
- Mortality moved, not just the marker. The reduction transmitted all the way to death in randomized evidence — and with no excess cancer or non-vascular death even at low LDL, which retires the old low-cholesterol-harm worry for the drug-lowered range.
- No knee, no floor in the studied range. There was «no evidence of any threshold within the cholesterol range studied»; benefit persisted even below 1.8 mmol/L (70 mg/dL). (Cholesterol Treatment Trialists’ Collaboration, 2010) Successive drops multiply — a 2 mmol/L reduction cuts risk by roughly 40% — so lower and longer both keep paying.
The relative effect is constant across baseline-risk groups, which lets you scale it to a person without any subgroup guesswork. The same ~22% is a small absolute gain for someone at low risk and a large one for someone at high risk. -> LDL Lowering and Cardiovascular Events
Age mostly holds that rule, with one gap. A later CTT pooling of 28 statin trials (186,854 people, 14,483 of them over 75) found the proportional cut in major vascular events shrank only slightly with age, a trend short of significance (p=0.06). The oldest were over-represented in four heart-failure and dialysis trials where statins did little; set those aside and the trend faded further (p=0.3). (Cholesterol Treatment Trialists’ Collaboration, 2019)
People over 75 with known vascular disease benefited much like younger ones (RR 0.85 per 1 mmol/L LDL-C reduction, 99% CI 0.73-0.98). Without vascular disease, benefit tended to shrink with age (trend p=0.05; p=0.03 without the heart-failure and dialysis trials), which the authors call a weak trend. There the over-75 estimate was RR 0.92 (99% CI 0.73-1.16) and 71-75 was 0.84 (0.70-1.01), and the authors judge there were too few such participants to assess that group reliably. (Cholesterol Treatment Trialists’ Collaboration, 2019)
The authors still expect a larger absolute benefit in older people, because untreated risk climbs steeply with age. That is their projection from risk arithmetic, not counted events: they declined to estimate absolute effects from the trials’ own event numbers, which they judged unrepresentative of any current population. The projection assumes the relative effect holds into old age, which is least certain in exactly the over-75 group without vascular disease. (Cholesterol Treatment Trialists’ Collaboration, 2019) -> Baseline Risk and the Relative-Absolute Split
The one place low LDL might not be benign
An observational signal points the other way for a single stroke subtype: people who already have low LDL-C carry somewhat higher hemorrhagic (bleeding) stroke risk. But this is an association, not a lowering experiment, and the authors themselves stop at a biomarker claim, flagging reverse causation and confounding. (Ma et al., 2019) The interventional test settles it for the drug route: pooling 38 trials (271,411 people) of statins and PCSK9 inhibitors, lowering LDL with drugs did not significantly raise brain hemorrhage in any agent or subgroup. (Li et al., 2025) So the observational bleeding-stroke arm is most likely not caused by lowering, and it does not justify withholding treatment from someone whose ischemic and vascular risk dominates. One residual piece — the lifelong-genetic-exposure question — remains open. -> LDL Lowering and Cardiovascular Events
It matters whether diet or a drug does the lowering
Here the two provisos above bite. The ~22%-per-mmol figure is a drug result, and a diet-lowered number does not automatically inherit it.
The cleanest demonstration comes from a recovered double-blind trial, the Minnesota Coronary Experiment. Replacing saturated fat with corn oil dropped serum cholesterol by 13.8% versus 1.0% in controls, yet produced no mortality benefit. A fall in total cholesterol «can represent widely different biochemical phenomena», and the trial measured neither particle number nor a competing off-target effect. (Ramsden et al., 2016) A companion trial saw cholesterol fall further in the intervention arm yet mortality rise. (Ramsden et al., 2013) Neither refutes apoB causality; both land inside its provisos. The practical lesson: a diet-induced cholesterol change is not interchangeable with a drug-induced apoB change. -> LDL ApoB and Cumulative Exposure
That said, real dietary levers on apoB exist — they are just modest and route-specific:
- The Portfolio pattern (nuts, plant protein, viscous fibre, plant sterols) lowered LDL-C 17% and apoB 15% in pooled trials, GRADE high — a magnitude near a starting-dose statin. But no trial measured events; the benefit is borrowed from the LDL-causality fabric, and it is the diet’s inventors reporting it. (Chiavaroli et al., 2018) -> Portfolio Dietary Pattern and LDL Cholesterol
- Protein source shifts the number more than most food swaps. In controlled feeding, both red and white meat raised LDL-C and apoB relative to plant protein, independent of saturated fat; meat color did not matter. (Bergeron et al., 2019) Across 36 trials, the sign of red meat’s lipid effect flips with what replaces it — worse than plant protein, better than fish or refined carbohydrate — so the substitute has to be named for the swap to carry a decision. (Guasch-Ferré et al., 2019) -> Lean Red Meat and Atherogenic Lipoproteins
No head-to-head trial has randomized a diet-lowered versus drug-lowered apoB to hard events — a genuine gap, not a settled equivalence.
Whether to act is a baseline-risk decision, and imaging can defer it
Because the relative effect is constant, the whole decision turns on absolute risk. Estimate a person’s 10-year cardiovascular risk (SCORE2 in Europe, PREVENT in the US), then read it against a threshold. Below roughly 7.5% ten-year risk, a primary-prevention adult is not a statin candidate by USPSTF, and the drug question closes before any imaging. (US Preventive Services Task Force, 2022) -> SCORE2 Baseline Risk and the ESC Treatment Thresholds, Statins for Primary Prevention and the Power of Zero CAC
In the grey zone, a coronary calcium (CAC) scan re-stratifies sharply. Among statin-eligible adults, a score of zero carried 4.2 events per 1,000 person-years versus 11.2 with any calcium — and a zero score reclassified about half of candidates out of clear benefit (estimated 10-year number needed to treat 64 at CAC=0 versus 28 at CAC>100). (Nasir et al., 2015) Two honesty guards: CAC=0 is low risk, not no risk, and no trial has randomized treatment by calcium score, so a zero belongs in a shared decision, not an automatic rule. Familial hypercholesterolemia and a very high Lp(a) override a low score.
Once a person is above the treatment threshold, the two major guideline families have converged on aggressive LDL-C goals — a goal of <55 mg/dL (1.4 mmol/L) in secondary prevention (Blumenthal et al., 2026), the same very-high-risk target the 2025 ESC/EAS focused update carries forward (Mach et al., 2025). But both families rest on the same trial base, so this is shared warrant, not two independent witnesses — and the specific cut is a reasoned stopping point on a monotone curve, not a trial-certified optimum. -> LDL Lowering and Cardiovascular Events
Lp(a): a fixed particle that raises the stakes
Lp(a) is an apoB-containing particle set almost entirely by genetics — stable over life, unmoved by diet, exercise, or weight. It is not a lever. It earns attention as a risk multiplier: higher Lp(a) tracks higher vascular risk (~16% per each roughly 3.5-fold increment for coronary disease), confined to vascular outcomes. (Erqou et al., 2009) Because that extra risk holds at every cholesterol level, a high Lp(a) raises the absolute payoff of the LDL/apoB-lowering levers a person can pull — reason to lower apoB harder, not reason to do anything Lp(a)-specific with lifestyle.
A causal leg now backs that risk-multiplier reading, with a dose to match. A Mendelian-randomization study (the LPA KIV-2 genetic instrument, which avoids reverse causation by construction) found genetically-raised Lp(a) carried an MI hazard of 1.22 (95% CI 1.09-1.37) per doubling — larger than the observational estimate, and «consistent with a causal association between elevated lipoprotein(a) levels and increased risk of MI». (Kamstrup, 2009)
A pooled MR then sized the lever: risk scales with the absolute mg/dL change, and mg-for-mg Lp(a) is only about two-fifths as potent as LDL-C (~0.38) — a 1-mmol/L (38.67 mg/dL) LDL-C drop matches roughly a 101.5-mg/dL (95% CI 71.0-137.0) Lp(a) drop for the same CHD reduction (each 10 mg/dL lower ~ 5.8% lower CHD risk, OR 0.942). (Burgess et al., 2018) That large absolute requirement is the point: it is why lowering Lp(a) enough to matter is hard, and why the causal-existence finding still does not close the treatment question — no Lp(a)-lowering outcome trial is yet held. -> Lipoprotein(a) and Cardiovascular Risk
The statin’s real costs are smaller than its reputation
Two fears drive people off statins, and the trial evidence sizes both.
Muscle symptoms are mostly not the drug. In ~155,000 people in blinded trials, statins raised any muscle pain or weakness by a hair (RR 1.03), with the small excess confined to year one — only about 1 in 15 such reports was actually caused by the drug. (Reith et al., 2022) A dedicated n-of-1 trial in people who had already quit statins for side effects found that 90% of the symptom burden a statin produced was also produced by an identical placebo — a nocebo effect, the symptom real but not statin-specific. (Wood et al., 2020) Half those patients successfully restarted.
The supplement people reach for does not fix it. Coenzyme Q10 is the popular remedy, and the mechanistic story is tidy — statins do lower circulating CoQ10. But the one held meta-analysis (Kovacic 2025, seven RCTs, n=389) reports a pain reduction that is statistically significant yet fragile: a weighted mean difference of -0.96 on a 0-10 scale, with an upper confidence bound of -0.03. The result turns non-significant when any of several single trials is dropped. (Kovacic et al., 2025)
The tell is where the benefit sits: it concentrates in trials that did not require a history of statin muscle symptoms to enrol, while the trials that did require reported symptoms show no significant effect. (Kovacic et al., 2025) That discordance is consistent with a nocebo or non-specific-pain explanation, read through the held SAMSON evidence above; Kovacic reads it instead as dilution masking a small real effect, and the subjective-only endpoint cannot separate the two. CoQ10 is cheap and safe, so a time-limited personal trial costs little, but it must not delay or replace an effective statin -> Coenzyme Q10 Supplementation for Statin Muscle Symptoms.
New-onset diabetes is real, but small. Unlike the muscle fear, this is a genuine pharmacological effect: statins raise new diabetes diagnoses by ~10% at low/moderate intensity and ~36% at high, an absolute excess near 1.2 per 1,000 person-years at standard doses. (Reith et al., 2024) It works by nudging glycemia up a fraction, so it lands mostly on people already near the diabetes threshold: about 62% of the excess cases arose in the top quarter of baseline glycaemia. (Reith et al., 2024) The cardiovascular benefit already nets out any cardiovascular harm that runs through higher glucose; the diagnosis itself stays a separate, small cost, which the CTT authors judge greatly outweighed by the fall in major vascular events. (Reith et al., 2024) Neither harm changes the start-or-continue decision for someone with a real indication. -> Statin Muscle Symptoms and the Nocebo Effect, Statins and New-Onset Diabetes
Eggs and dietary cholesterol: close to a non-lever
The cholesterol in food is a weak lever on the cholesterol in blood for most people. A medium egg carries about 180 mg of cholesterol, but absorption and the body’s own synthesis largely compensate, so dietary cholesterol barely moves serum LDL except in a minority of “hyper-responders.” (Godos et al., 2020) The apoB lever runs mainly through other routes — genetics and saturated fat, not ingested cholesterol. For cardiovascular purposes eggs are close to a non-lever for the general population, and the full appraisal, including the strata and cautions, lives on its own page. -> Eggs Dietary Cholesterol and Cardiovascular Risk
Lipids are one axis of cardiovascular risk, not the whole of it
Lowering apoB answers the lipid axis, not the whole cardiovascular question. Inflammation is a separate, additive lever: in the CANTOS trial, canakinumab cut recurrent cardiovascular events «independent of lipid-level lowering» in statin-treated patients whose LDL was already controlled. (Ridker et al., 2017) So a person at goal on apoB can still carry residual inflammatory risk that lipid-lowering does not touch. Blood pressure is the parallel large lever, handled on its own axis. -> Inflammation as a Modifiable Lever, Blood Pressure
A second outcome lies beyond the artery: higher LDL-C in midlife tracks higher later dementia incidence (about 8% per 1 mmol/L, midlife-specific), which adds to the reasons to lower it without changing the cardiovascular verdict. (Livingston et al., 2024) -> Dementia Prevention and Modifiable Risk Factors
What to do with this
Rank the levers by size, and the lipid axis sorts itself quickly:
- Get the absolute risk first. A 10-year risk estimate, and a calcium scan in the grey zone, decides whether the lipid lever is worth pulling at all. Below the treatment threshold, a fraction-of-a-percent absolute benefit rarely justifies a lifelong drug.
- Read the right number. LDL-C is fine if you are lean and metabolically healthy; ask for apoB or non-HDL-C if you are insulin-resistant, hypertriglyceridemic, or diabetic, where LDL-C under-reads the danger.
- If the lever is worth pulling, lower and keep it lowered. The benefit is proportional with no floor in the studied range, and a drug’s per-unit effect is the one actually measured on hard events. A diet-lowered number is a real but modest and less-certain version of the same lever.
- Do not let the small costs decide it. Muscle complaints are mostly nocebo; the diabetes risk is real but small: its cardiovascular effects are already counted in the benefit, and the diagnosis itself is a minor cost the benefit outweighs.
- Spend little worry on eggs. Dietary cholesterol is a minor input to the number that matters.
The loop stays open: nothing here has been graded against a realized outcome for a specific person. The direction and the mechanism are settled; the weighting and the baseline risk you bring to it are personal.
Evidence box
Question ’What is the relationship between apoB-containing lipoproteins and atherosclerotic cardiovascular disease — how settled is it, which number best captures a person”s risk and where do the numbers diverge, how much does moving that number change hard events, does it matter whether diet or a drug does the moving, and for whom is acting warranted?‘ Evidence included 28 sources — 15 gold, 13 high Overall certainty High (see Rating Certainty of Evidence) Source-selection note All sources are gold or high tier. Last updated 2026-10-06 · Independently reviewed: No · Full edit history