Two separable questions sit under the lipid lever, and two sources answer them. How much does lowering buy — the per-unit magnitude of event reduction (CTT 2010, the IPD meta-analysis of 26 statin RCTs). And which number to target — the metric that carries the causal signal (Marston 2022, apoB over LDL-C in ~430,000 people). This page holds the magnitude and the metric; the causal model underneath (why LDL/apoB causes disease, why the dose is cumulative) is the nucleus LDL ApoB and Cumulative Exposure. Structural sibling to Blood Pressure Lowering and Cardiovascular Events on the other big cardiometabolic lever.
The magnitude — CTT per 1.0 mmol/L LDL-C reduction
CTT scales every trial by the LDL-C reduction achieved at 1 year and reports the effect per 1.0 mmol/L (169,138 participants, 26 trials, median ~5 y):
| Outcome (per 1.0 mmol/L LDL-C reduction) | Effect |
|---|---|
| Major vascular events | RR 0.78 (95% CI 0.76-0.80) |
| All-cause mortality | RR 0.90 (0.87-0.93) |
| Coronary death | RR 0.80 |
| Non-fatal MI | RR 0.73 (-27%) |
| Coronary revascularisation | RR 0.75 (-25%) |
| Ischaemic stroke | RR 0.79 (-21%); any stroke -16% |
| Cancer incidence / non-vascular death | RR 1.00 / 0.97 (null) |
«similar proportional reductions in major vascular events per 1·0 mmol/L LDL cholesterol reduction were found in all types of patient studied (rate ratio [RR] 0·78, 95% CI 0·76-0·80; p<0·0001), including those with LDL cholesterol lower than 2 mmol/L» (Cholesterol Treatment Trialists’ Collaboration, 2010)
«all-cause mortality was reduced by 10% per 1·0 mmol/L LDL reduction (RR 0·90, 95% CI 0·87-0·93; p<0·0001)» (Cholesterol Treatment Trialists’ Collaboration, 2010)
Mortality moved, not just the marker. Unlike most of the surrogate cases the wiki holds, the LDL-C reduction transmitted all the way to all-cause death in randomised evidence — the validated-surrogate direction, and a worked instance of Surrogate Outcomes’s counter-exemplar. And with no excess cancer or non-vascular mortality even at low LDL, which retires the old low-cholesterol-harm hypothesis for the drug-lowered range.
By age — the same per-mmol effect to 75, a thinner and partly confounded signal above it [2026-10-06, CTT 2019]
CTT 2019 re-analysed the statin trials by age at entry (28 trials, 186,854 people, median 4.9 y; 14,483 older than 75). It is the same collaboration and largely the same trials as the 2010 table above (CORONA and HOPE-3 added), so it refines that table by age rather than corroborating it. Per 1.0 mmol/L LDL-C reduction, overall column 95% CIs, >75 column 99% CIs:
| Outcome | All ages (95% CI) | Age trend p | >75 (99% CI) |
|---|---|---|---|
| Major vascular events | 0.79 (0.77-0.81) | 0.06 (0.3 without HF/dialysis trials) | 0.87 (0.77-0.99); 0.82 (0.70-0.95) without HF/dialysis |
| Major coronary events | 0.76 (0.73-0.79) | 0.009 (0.01 without HF/dialysis) | 0.82 (0.70-0.96) |
| Coronary revascularisation | 0.75 (0.73-0.78) | 0.6 | 1.02 (0.75-1.40), 210 vs 209 procedures |
| Any stroke | 0.84 (0.80-0.89) | 0.7 | 0.89 (0.71-1.10) |
| Vascular death | 0.88 (0.85-0.91) | 0.004 (0.2 without HF/dialysis) | 0.95 (0.83-1.07); 0.86 (0.69-1.08) without |
| MVE, with vascular disease | 0.80 (0.77-0.82) | 0.2 (0.9 without HF/dialysis) | 0.85 (0.73-0.98) |
| MVE, no known vascular disease | 0.75 (0.71-0.80) | 0.05 (0.03 without HF/dialysis) | 0.92 (0.73-1.16) |
(Cholesterol Treatment Trialists’ Collaboration, 2019)
- The heart-failure and dialysis trials drag the old-age estimates. Statins barely moved MVE in those four trials (RR 0.95 in each pair vs 0.77 elsewhere), 20% of the >75 group came from the HF trials, and «However, 1014 (53%) of 1906 vascular deaths among people older than 75 years occurred in the four heart failure or dialysis trials.» Without them: «Among people without heart or renal failure, we found little evidence of any diminution of benefit with increasing age on major vascular events (ptrend=0·3) or on vascular death (ptrend=0·2).» (Cholesterol Treatment Trialists’ Collaboration, 2019)
- No excess non-vascular death or cancer at any age: «Statin therapy had no effect at any age on non-vascular mortality, cancer death, or cancer incidence.» (Cholesterol Treatment Trialists’ Collaboration, 2019) All-cause mortality RR 0.91 (95% CI 0.88-0.93) overall; its age trend (p=0.04) did not persist once the HF/dialysis trials were excluded (p=0.1). Other harms by age (diabetes, cognition) were left to a separate CTT adverse-event analysis, not held. (Cholesterol Treatment Trialists’ Collaboration, 2019)
- The weak cell is primary prevention past 75, where the CI spans both the all-age effect and no effect; the authors judge «there were too few such older participants for reliable assessment in that group alone». The route (a)/(b) reading is on Baseline Risk and the Relative-Absolute Split; the start-a-statin consequence is on Statins for Primary Prevention and the Power of Zero CAC. (Cholesterol Treatment Trialists’ Collaboration, 2019)
- So the 2010 claim that the per-mmol effect held «in all types of patient studied» survives by age where the data are dense (to 75, and secondary prevention past 75), except a modest graded attenuation on major coronary events across all ages (cause unclear), and is unconfirmed, not refuted, for the oldest people without vascular disease. (Cholesterol Treatment Trialists’ Collaboration, 2010, inferred from 2019)
The shape — monotone, no threshold, multiplicative
- No knee, no plateau in the studied range. «There was no evidence of any threshold within the cholesterol range studied» (Cholesterol Treatment Trialists’ Collaboration, 2010). Benefit persisted starting below 2.0 mmol/L (RR 0.71, -29%) and «even among those reaching 1·8 mmol/L (70 mg/dL) or lower» further reduction still produced definite benefit (RR 0.63, 99% CI 0.41-0.95) (Cholesterol Treatment Trialists’ Collaboration, 2010). The RR per mmol «did not depend on the baseline LDL cholesterol concentration (trend p=0·2)».
- The reduction compounds multiplicatively: «a 2 mmol/L reduction would reduce the risk by about 40% (since the combination of risk ratios of 0·78×0·78 yields a risk ratio of about 0·6)» (Cholesterol Treatment Trialists’ Collaboration, 2010). This is the per-unit face of the nucleus’s cumulative-exposure thesis: successive mmol multiply, so more and longer both pay -> LDL ApoB and Cumulative Exposure.
- A clean confirmation of the corpus’s operative dose-response default — every reduction pays; the burden is on anyone claiming a knee to locate it. The persistence of benefit at low LDL also means there is no protective lower arm (no harm from low LDL in the drug-lowered range), the interventional parallel to The U-Shaped Association Artifact. This no-harm reading is OUTCOME-SPECIFIC — it holds for major vascular events, all-cause mortality and ischaemic/total stroke, but an observational inverse association runs the other way for the hemorrhagic (intracerebral) stroke subtype -> the counter-signal section below.
The hemorrhagic-stroke counter-signal — an observational low-LDL harm arm, subtype-specific and causally unsettled [2026-09-15, Ma]
CTT’s no-harm-at-low-LDL conclusion is the drug-lowered, composite-outcome reading. Ma’s 2019 dose-response meta-analysis of 12 prospective studies (476,173 participants, 7587 hemorrhagic-stroke cases) points the other way for one subtype: lower LDL-C associates with HIGHER hemorrhagic (intracerebral) stroke risk — per 10 mg/dL higher LDL-C, RR 0.97 (95% CI 0.95-0.98), i.e. the risk rises as LDL-C falls. (Ma et al., 2019) The signal concentrates below ~110 mg/dL descriptively, though the formal non-linearity test was NOT significant («We did not detect any significant non-linearity relation between LDL-C and hemorrhagic stroke (p = 0.09, Fig. 2)» (Ma et al., 2019)), so 110 is a descriptive break, not a located knee, and the primary model is a straight-line inverse association over the studied range (~39-153 mg/dL).
This is a DISTINCTION from CTT, not a tension — different exposure, different outcome, different estimand. Before reading it as a low-LDL harm arm that contradicts the page’s benefit magnitude, the parameter table (not-joined check (ii): different quantity):
| Parameter | CTT 2010 | Ma 2019 | Same quantity? |
|---|---|---|---|
| Design | RCT IPD meta-analysis (causal) | observational prospective-cohort MA (association) | No |
| Exposure | drug-lowering LDL-C (statin) | measured LDL-C level (largely naturally-occurring; a cited cohort had <1% lipid-lowering-drug use) | No |
| Outcome | major vascular events / all-cause mortality / any stroke | hemorrhagic (intracerebral) stroke only | No |
| Estimand | causal effect of lowering | risk discrimination / association | No |
| Sign at low LDL | benefit persists, no threshold, no excess harm | higher hemorrhagic-stroke risk | opposed, but not the same quantity |
So the two are compatible statements about different things: CTT says lowering LDL with a drug cuts composite events and mortality with no threshold; Ma says people who already have low LDL carry more hemorrhagic-stroke risk. The decisive appraisal move is that Ma’s signal is observational and the authors themselves stop at a biomarker claim, not causation — they flag the two artifact routes explicitly: reverse causation («low LDL-C concentrations could be due to certain pre-clinical pathologic changes of hemorrhagic stroke» (Ma et al., 2019)) and residual confounding («that residual confounding may contribute to the low LDL-C concentrations» (Ma et al., 2019)), concluding only that «low LDL-C could be a biomarker for increased hemorrhagic stroke risk» (Ma et al., 2019) and calling for MR + safety-monitored trials. This is a textbook case for The U-Shaped Association Artifact: a protective/harm arm read off observational data must survive a referent-correction or a genetic/objective check before it is believed, and neither has been applied here.
The causal question is a G-gap; the intervention arm has now landed and the genetic arm has not. (inferred from Ma et al., 2019) Does low (especially drug-lowered / very-low PCSK9i-range) LDL causally raise intracerebral hemorrhage, or is Ma’s association reverse-causation/frailty confounding? Ma’s observational data cannot say — but an interventional NMA can test the lowering leg directly.
The intervention arm — Li 2025: lowering LDL with drugs does not reproduce the observational ICH harm [2026-09-15, Li]
Li’s Bayesian network MA of 38 RCTs (271,411 individuals) of statins and PCSK9 inhibitors versus placebo, primary outcome ICH, is the interventional test the Ma section owed. The result: no agent significantly raises ICH — every per-agent credible interval crosses 1 (Li et al., 2025):
| Agent (vs placebo) | ICH OR (95% CrI) | Agent (vs placebo) | ICH OR (95% CrI) |
|---|---|---|---|
| Alirocumab | 0.84 (0.34–2.2) | Rosuvastatin | 1.4 (0.84–2.2) |
| Atorvastatin | 1.1 (0.65–1.7) | Pravastatin | 1.3 (0.84–1.9) |
| Bococizumab | 1.4 (0.51–3.7) | Simvastatin–Ezetimibe | 1.2 (0.63–2.3) |
| Fluvastatin | 0.58 (0.19–1.7) | Simvastatin | 0.91 (0.53–1.5) |
| Lovastatin | 0.20 (0.0064–1.6) | Evolocumab | 1.2 (0.48–2.8) |
«In our analysis, active therapy was associated with a nonsignificant increase in the risk of ICH.» (Li et al., 2025) «LDL-Cholesterol lowering agents (statins and PCSK-9 inhibitors) was not associated with a significant increased risk of ICH.» (Li et al., 2025)
The no-excess-ICH result held across every subgroup Li could cut — coronary-heart-disease and healthy/ no-CHD patients, primary and secondary prevention, and lipophilic vs hydrophilic statins alike — and the authors read the residual weak signal as off-target, not LDL-mediated: «any potential increase in ICH risk could be attributable to other non-LDL effects of statin therapy» (Li et al., 2025). Mortality fell under two agents (atorvastatin OR 0.82 [0.69–0.97], rosuvastatin 0.79 [0.65–0.95]; rest NS) (Li et al., 2025), so the net direction in the trials is favourable, not harmful.
Ma + Li: a DISTINCTION that resolves toward artifact, not a tension — different estimands (not-joined check (ii)). (Li et al., 2025; inferred from Ma et al., 2019)
| Parameter | Ma 2019 | Li 2025 | Same quantity? |
|---|---|---|---|
| Design | observational prospective-cohort dose-response MA | RCT Bayesian network MA (causal) | No |
| Exposure | measured LDL-C level (largely naturally occurring) | drug-lowering of LDL (statins + PCSK9i) vs placebo | No |
| Outcome | intracerebral (hemorrhagic) stroke | intracerebral hemorrhage | Yes |
| Estimand | risk association per 10 mg/dL LDL-C | causal OR of a lowering agent vs placebo | No |
| Result at low/lowered LDL | higher ICH (RR 0.97 per 10 mg/dL, 0.95–0.98) | no significant ICH excess (all CrIs cross 1) | — |
The exposures and estimands differ, so this is not a joined tension. But the two answer the decision
question together: a causal reading of Ma’s association predicts that lowering LDL should raise ICH, and
the interventional evidence does not reproduce that — the type-F composite (Li supplies the causal leg
Ma lacked) points to Ma’s low-LDL/ICH association being substantially non-causal-via-lowering (reverse
causation / frailty confounding, or a non-LDL statin channel), exactly the failure the U-shaped-artifact
rule predicts for an observational arm that survives no interventional check
-> The U-Shaped Association Artifact. Author lists are disjoint (Li, Wangwen vs Ma, Chaoran) and the
evidence streams independent, which is why the composite is not circular — but it is F, not
[E-independent] (the two reach different claims, not the same one).
Two honesty guards keep this from over-claiming a refutation (symmetric standards).
-
Li’s harm arm is a wide-CI null, not a tight no-effect. ICH is rare, so several CrIs stay consistent with a real increase (rosuvastatin up to 2.2x, bococizumab up to 3.7x); Li shows no significant excess, not proven safety, and moderate heterogeneity plus possible ICH publication-bias asymmetry sit on the estimate. (Li et al., 2025)
-
Estimand + horizon gap. Ma’s exposure is a lifelong LDL-C level; Li’s is a few years of drug-lowering. If low LDL raises ICH through slow structural vessel change (Ma cites microaneurysm / arterial-wall fragility), a short trial need not reproduce a lifetime effect — so the intervention null bounds, but does not fully close, the level-based causal question.
-
Causal/genetic arm — still the G-gap. A Mendelian-randomization check on whether the low-LDL/ICH signal is causal or confounded (Qin 2023, lipids + lipid-lowering drugs vs stroke subtypes; Frontiers, moderate tier, not held — cut, not awaited) would close the lifelong-exposure leg the trial horizon cannot. Named here as the genetic-check lever the artifact rule demands, not asserted.
Operative reading now: the LDL-lowering benefit on composite CV events and mortality is causal and robust; the subtype-specific low-LDL hemorrhagic-stroke association (Ma) is not reproduced by drug-lowering in RCTs (Li), so it is most likely non-causal-via-lowering and certainly does not license withholding LDL-lowering from a person whose absolute ischaemic/vascular risk dominates. The one residual leg is the genetic (lifelong-exposure) check.
The metric — Marston: target apoB particle number, not cholesterol content or particle type
CTT scales by LDL-C because that is what the statin trials measured. Marston asks whether LDL-C is even the right number to carry the risk. In UK Biobank (n=389,529, untreated) plus two statin RCTs (FOURIER, IMPROVE-IT; n=40,430), each lipid is entered simultaneously:
- apoB, non-HDL-C and TG each predict MI alone, but only apoB survives mutual adjustment — «when assessed together, only apoB was associated (adjusted hazard ratio [aHR] per 1 SD, 1.27; 95% CI, 1.15-1.40; P < .001). Similarly, only apoB was associated with MI in the secondary prevention cohort.» (Marston et al., 2022). Non-HDL-C fell to aHR 1.09 (NS) and TG to 1.00 (NS) once apoB was held constant.
- Particle TYPE stops mattering once particle NUMBER is fixed — adjusting for apoB, the TG/LDL-C ratio (a proxy for TG-rich vs LDL particles) was flat (aHR 1.04, 0.99-1.09, P=.12): «for a given concentration of apoB-containing lipoproteins, the relative proportions of particle subpopulations may no longer be a predictor of risk» (Marston et al., 2022).
- Verdict: «risk of MI was best captured by the number of apoB-containing lipoproteins, independent from lipid content (cholesterol or TG) or type of lipoprotein (LDL or TG-rich)» (Marston et al., 2022). Where apoB is unavailable, «non-HDL-C in particular is the preferred surrogate for apoB, as it incorporates TG-rich lipoproteins in addition to LDL» — LDL-C is the weakest of the three because it misses the remnant particles.
The genetic version of the same answer — Richardson MVMR [2026-08-06]. Marston’s apoB-supremacy is
observational; multivariable Mendelian randomisation reaches it as a natural experiment. Instrumenting
LDL-C, TG and apoB from a UK Biobank GWAS (up to 441,016) against CARDIoGRAMplusC4D (60,801 CHD cases):
each is atherogenic alone, but entered together “only apolipoprotein B … retained a robust effect, with
the estimate for LDL cholesterol … reversing” — apoB OR 1.92 (1.31-2.81), LDL-C 0.85 (0.57-1.27, P=0.44),
TG weakened. (Richardson et al., 2020) So the metric
verdict (target apoB, not LDL-C content) is now backed by RCT-scaling (CTT), observational multivariable
(Marston), and genetic multivariable MR (Richardson) alike. Note: this does not discredit LDL-C’s
causal role — apoB is necessary and always accompanies cholesterol; a cholesterol change concordant with
apoB still moves risk (which is why CTT’s LDL-C scaling works for statins). Causal model + the
not-independent caveat: LDL ApoB and Cumulative Exposure.
CONTESTED — the particle-number-not-content leg (Helgadottir 2022). Marston’s verdict is that risk is best captured «independent from lipid content (cholesterol or TG)» (Marston et al., 2022) — i.e. count, not content. An independent MR (deCODE/Danish; no Ference/Ala-Korpela/Marston authors) that separates apoB from non-HDL-C with 82/235 discordant variants reaches the opposite: «the genetic effects on CAD risk are proportional to effects on non-HDL-C, but not to effects on apoB» (Helgadottir et al., 2022), attributing the observational apoB signal to confounding of two ~0.9-correlated traits. This leaves the apoB-over-LDL-C metric verdict intact (both agree) but contests whether the right secondary target is apoB (particle number) or non-HDL-C (cholesterol content, already on every panel) -> full joined issue: ApoB Particle Number vs Cholesterol Content. Held contested, not resolved (one MR vs Marston + the number-camp lineage).
Synthesis — the two answers reconcile; they do not clash
(inferred from Cholesterol Treatment Trialists’ Collaboration, 2010; Marston et al., 2022)
A naive reading pits them against each other: CTT scales the entire causal magnitude by LDL-C, while Marston finds LDL-C carries no independent signal beyond apoB. The parameter table shows why this is a distinction, not a tension (not-joined check (ii): different quantity, different question):
| Parameter | CTT 2010 | Marston 2022 | Same quantity? |
|---|---|---|---|
| Design | RCT IPD meta-analysis | prospective cohort (+ cited Mendelian randomization) | No |
| Estimand | causal effect of lowering LDL-C (treatment contrast) | baseline risk discrimination among correlated lipids | No |
| Lipid unit | achieved LDL-C reduction (mmol/L) | 1-SD of each lipid, mutually adjusted | No |
| What “LDL-C” does | the scaling proxy | drops to NS after apoB adjustment | — |
CTT’s LDL-C-scaled magnitude and Marston’s apoB-supremacy are compatible because LDL-C and apoB are tightly correlated in the general untreated population — ρ >= 0.95 in Marston’s untreated primary-prevention cohort (Marston et al., 2022) — and a statin lowers both. So for the concordant majority, CTT’s «per 1.0 mmol/L LDL-C» is a stand-in for the apoB-particle reduction it accompanies — the causal quantity Marston identifies and the nucleus names. The decision-relevant seam, with its proviso made explicit:
- The equivalence is strongest where the lipids concord, and that is NOT the treated/impaired case. Statin treatment and the discordant metabolic states (metabolic syndrome, diabetes, hypertriglyceridemia — small-dense-LDL) drive LDL-C and apoB apart, and Marston found apoB supreme even in its statin-treated secondary cohort. So CTT’s ρ>=0.95-backed LDL-C proxy holds for the concordant, general-population case and weakens exactly where lipid-lowering or metabolic impairment has begun — the reason to read the particle number, not the cholesterol, once treatment or discordance is in play -> LDL ApoB and Cumulative Exposure.
- Use apoB (or non-HDL-C) to decide and monitor; expect roughly CTT’s magnitude per equivalent apoB reduction (exactly in the concordant; approximately when read off LDL-C, under-counting where they discord). The two sources answer which number and how much, and only together give the whole decision.
Independence note: not a type-E robustness convergence. Ference (nucleus) co-authors both Marston and
Richardson, and CTT is the RCT evidence family the nucleus’s consensus already rests on — one research
programme across all three. The value here is A/F (magnitude × metric assembled into one decision structure;
Marston refines what CTT’s LDL-C scaling is; Richardson adds the genetic-MR leg), not independent
triangulation — so none of the three is tagged [E-independent].
Decision relevance
- The relative effect is large and monotone; the absolute benefit still scales with baseline risk. A ~22% RRR per mmol is a small absolute gain in a low-risk person and a large one in a high-risk person — the same treat-on-absolute-risk logic as the BP lever (Baseline Risk and the Relative-Absolute Split, Statins for Primary Prevention and the Power of Zero CAC). CTT’s constant RR across every baseline stratum is what licenses stratifying on baseline risk without any subgroup claim (route a).
- A high Lp(a) is one such baseline-risk multiplier. Because the Lp(a)-CHD association holds at every level of non-HDL cholesterol, a known-high Lp(a) raises a person’s absolute risk and so raises the absolute benefit the same LDL/apoB reduction buys — a route-(a) shift needing no subgroup claim, since nothing lifestyle does moves Lp(a) itself. (Erqou et al., 2009) -> Lipoprotein(a) and Cardiovascular Risk
- Lower and longer both pay (no threshold, multiplicative) — but “worth it” is a net-effect call: the rhabdomyolysis excess (4 vs 1 per 10,000) was confined to 80 mg simvastatin, and the whole-strategy trade-off, not the LDL number, is the unit of decision.
- The muscle-harm side of that ledger is small and front-loaded
[2026-09-07, Reith/CTT]. The CTT Collaboration’s IPD meta-analysis (~155,000 in double-blind statin RCTs) sizes the harm the net-effect call needs: per 1000 people, an LDL-C reduction of 1 mmol/L for 5 years «might cause 11 (generally mild) episodes of muscle pain or weakness, but prevent 50 major vascular events in those with pre-existing vascular disease … and 25 major vascular events in those without». (Reith et al., 2022) Serious myopathy is rarer still — «RR 1·74; 95% CI 1·11–2·74 … an absolute excess of 0·08 (0·01–0·18) per 1000 person-years», ~50x rarer than ordinary aches and mostly year-1. (Reith et al., 2022) So the harm column of the LDL-lowering ledger is small versus the benefit column, and the tolerability and deprescribing decision it drives lives on Statin Muscle Symptoms and the Nocebo Effect. - New-onset diabetes is the OTHER harm-column entry, and unlike muscle symptoms it is a real
pharmacological effect
[2026-09-07, Reith/CTT]. The same CTT Collaboration’s IPD meta-analysis of glycaemia finds statins raise new diabetes diagnoses by a small, intensity-graded amount (low/moderate RR 1.10, high RR 1.36), concentrated in people already near the diagnostic threshold — a genuine drug effect, not a nocebo artifact, but still small in absolute terms. Reith’s own net-effect call is that «the absolute benefits of statin therapy greatly outweigh any excess risks of diabetes associated with the small increase in glycaemia they induce». (Reith et al., 2024) The magnitude, mechanism, and the stratum where the harm concentrates live on Statins and New-Onset Diabetes. (Same CTT body as this page’s benefit magnitude and the muscle entry above -> agreement across the three is NOT independent type-E backing.) - Drug route only. CTT is statin-lowering; the magnitude does not transfer to a dietary LDL-C change (the MCE/Sydney disconnects -> Surrogate Outcomes, Linoleic Acid and Cardiovascular Disease).
A dietary LDL lever, and why its magnitude does not inherit CTT [2026-08-29, Landry]
Landry’s 2024 umbrella quantifies one dietary LDL route: in 31 observational studies (Benatar), «vegan dietary patterns may be associated with significantly lower LDL-cholesterol concentrations of −0.49 mmol/l (−0.62, −0.36) (p < 0.0001) compared to omnivorous diets, though heterogeneity was high (I2=92 %)» (Landry et al., 2024). Two guards stop this being read as a ~11% event reduction off CTT’s 0.78/mmol:
- Observational and healthy-adherer confounded, GRADE low — and the randomised evidence in presumably-healthy adults found no significant effect: vegetarian+vegan combined LDL «−0.13 mmol/l (−0.37, 0.12)» (4 RCTs, NS) (Landry et al., 2024). So the −0.49 is the confounded observational estimate, not a trial effect.
- By this page’s own drug-route-only rule, a dietary LDL change does not transmit to events at CTT’s statin-derived rate (the MCE/Sydney disconnects). A modest, uncertain dietary apoB/LDL lever — cross-linked from Vegetarian Dietary Patterns and Mortality, where it is one of several small pleiotropic channels — not a statin-magnitude one. (inferred from Landry et al., 2024)
The treat-to-target number — the two guidance families converge, but the NUMBER is not the DIRECTION [2026-08-06]
CTT gives the magnitude per unit; the guidelines translate it into a level to treat to. The decision-relevant fact for 2026 is that the two major families now converge on the most aggressive target, and that convergence must be read correctly: it raises the guidance null for direction, not for the specific number.
The very-high-risk secondary-prevention target is the same in both families.
| Parameter | ESC-EAS (2019, reaffirmed 2025) | ACC-AHA 2026 | Same quantity? |
|---|---|---|---|
| Very-high-risk LDL-C goal | <1.4 mmol/L (55 mg/dL) + ≥50% reduction | «a goal of LDL-C <55 mg/dL (1.4 mmol/L)» + ≥50% reduction | Yes |
| High-risk / primary-high goal | <1.8 mmol/L (70 mg/dL) | «goal of LDL-C <70 mg/dL (1.8 mmol/L)» | Yes |
| non-HDL-C secondary goal | — (not set in the 2025 text) | <85 mg/dL (2.2 mmol/L) | ACC-only |
(Blumenthal et al., 2026) — ESC-EAS 2025 states «the LDL-C treatment goals … for persons in each risk category have not changed from the 2019 ESC/EAS Guidelines» (Mach et al., 2025); the ESC numeric figure itself lives in that update’s Fig. 1 (not in the OCR’d text — do not source the numbers to the 2025 document; they are the reaffirmed 2019 targets).
The delta that makes this a finding: ACC-AHA restored numeric goals. The 2018 US guideline had abandoned treat-to-target; 2026 puts it back — «LDL-C and non-HDL-C treatment goals are back to guide LLT» (Blumenthal et al., 2026) — and lands on the ESC numbers. So the US and European families, which diverged on whether to name a target for a decade, now agree on the target itself.
But this is shared-source agreement, and the two families are not two separate witnesses. Both rest on
the same trial base — CTT for the per-mmol magnitude (ESC states it as its «average 20 % proportional
… reduction … per each mmol/L» (Mach et al., 2025); the
ACC document rests on CTT 2010 throughout) plus FOURIER/ODYSSEY/IMPROVE-IT/CLEAR. Two guideline committees
reading the same trials to the same number is convergence of shared warrant, not a second route — so this
carries no [E-independent] weight.
The load-bearing honesty — the DIRECTION is evidenced, the specific NUMBER is a reasoned extrapolation. A target is legitimate only if its transmission to a patient-important outcome is itself evidenced. Here:
- Direction (lower LDL/apoB → fewer events, monotone, no threshold) is high-certainty — CTT’s «no evidence of any threshold», benefit persisting «even among those reaching 1·8 mmol/L (70 mg/dL) or lower», and the concordant genetic MR (Richardson). The guidelines’ own claim that the per-unit benefit is mechanism-agnostic is the same view: «similar reductions in the risk of CVD as statins … per unit decrease in LDL-C» (Mach et al., 2025).
- The specific cut (why <55 and not <50 or <70) has no head-to-head trial. No RCT randomized patients to target <55 vs <70 mmol/L; the numbers are derived from achieved-LDL strata in the outcome trials and the monotone curve, extrapolated to a round threshold. Under a monotone no-threshold curve a target is a pragmatic stopping point, not an evidenced optimum — reducing past it merely keeps paying (or stops being worth the added agent/cost/adherence burden), which is a net-effect judgment, not a measured knee. So the convergence licenses «lower, for the high-risk» with confidence; it does not certify 1.4 mmol/L as the right number.
(inferred from Blumenthal et al., 2026; Cholesterol Treatment Trialists’ Collaboration, 2010; Mach et al., 2025)
Limits
- CTT magnitude is a ~5-year per-mmol effect from drug trials; the lifetime cumulative effect is larger by the nucleus’s own logic but is modelled, not measured here.
- Marston is observational (with cited MR) and shares lineage with the nucleus’s consensus — large and consistent, but not independent of it; measured on conventional panels, not NMR; not enriched for severe hypertriglyceridemia.
- The guidelines are WHAT-TO-DO documents; their effect claims are borrowed. ACC-AHA and ESC-EAS are named here for the recommendation (the target, the escalation), never as evidence of the magnitude — the magnitude is CTT’s/Marston’s, which the guidelines themselves rest on (ACC-AHA even reproduces Marston’s aHR 1.27). A guideline’s restatement of CTT is not a second confirming study.
- The open loop stands: no operation here grades either against a realized patient outcome.