Lipoprotein(a) is an LDL-like particle in which one apolipoprotein B100 molecule is covalently linked to apolipoprotein(a) (Erqou et al., 2009). Its concentration is largely genetically set (by the LPA gene), stable within a person over years, and not meaningfully moved by the diet/activity/weight levers this wiki is built around. So it is not itself a modifiable exposure. It earns a page for a different reason: it is a baseline-risk stratifier that changes the absolute payoff of the lipid-lowering levers a person can pull -> see the decision hook below.
The observational association (what ERFC established)
The Emerging Risk Factors Collaboration pooled individual participant data on 126 634 people in 36 prospective studies, over 1.3 million person-years, having excluded anyone with known CHD or stroke at baseline (Erqou et al., 2009). Higher usual Lp(a) tracked higher vascular risk, and the association was broadly continuous in shape (no threshold located; a potentially curvilinear rise at the high end is noted, not resolved).
Effect sizes are stated per 3.5-fold higher usual Lp(a) concentration (ie, per 1 SD), after regression-dilution correction:
| Outcome | Adjusted risk ratio per 1-SD (3.5-fold) higher usual Lp(a) | 95% CI |
|---|---|---|
| CHD (age/sex-adjusted) | 1.16 | 1.11-1.22 |
| CHD (further adj. for lipids + conventional risk factors) | 1.13 | 1.09-1.18 |
| Ischemic stroke | 1.10 | 1.02-1.18 |
| Nonvascular mortality (aggregate) | 1.01 | 0.98-1.05 |
| Cancer deaths | 1.00 | 0.97-1.04 |
| Nonvascular deaths other than cancer | 1.00 | 0.95-1.06 |
Two features carry the decision, not the point estimates:
- The effect is modest and vascular-specific. ERFC’s own summary: «Under a wide range of circumstances, there are continuous, independent, and modest associations of Lp(a) concentration with risk of CHD and stroke that appear exclusive to vascular outcomes.» (Erqou et al., 2009) The three nonvascular rows are flat (CIs straddle 1.00) — the risk is confined to vascular disease, not a general frailty marker.
- It survives adjustment for the standard lipids. The CHD RR falls only from 1.16 to 1.13 after adjusting for non-HDL cholesterol and other conventional risk factors — Lp(a) adds risk information beyond a standard lipid panel. Overall it is only about one-quarter as strong as non-HDL cholesterol for CHD (Erqou et al., 2009).
The decision hook — a route-(a) baseline-risk multiplier
Because Lp(a) is essentially unmodifiable by lifestyle, its value is not a new lever but a baseline-risk shift (transportability route (a): absolute benefit scales with baseline risk even when the relative effect of the available intervention is unchanged, with no subgroup claim required). ERFC states the consequence directly: because the Lp(a)-CHD association is similar at every level of non-HDL cholesterol, «the absolute benefits of cholesterol lowering should be greater if Lp(a) concentration is high (or when absolute risk is high for some other reason).» (Erqou et al., 2009)
So the decision-change is: a known-high Lp(a) up-weights the apoB/LDL-lowering lever LDL ApoB and Cumulative Exposure — the same diet, and the drug comparators (statins, ezetimibe), buy more absolute risk reduction in a person whose absolute risk is raised by high Lp(a). It does not tell that person to do anything Lp(a)-specific with lifestyle, because there is nothing lifestyle does to Lp(a). This keeps Lp(a) firmly a Layer-2 prognostic input, not a Layer-1 exposure of its own.
The causal leg — Mendelian randomization (now held)
ERFC is an observational IPD meta-analysis, so its RRs are risk associations, corrected for regression dilution but still open to residual confounding. The causal magnitude is a separate question that source cannot answer, and it flags the direction itself: the LPA-gene variant studies, taken with these findings, are «consistent with the existence of a causal relationship» and «interventions capable of much more powerful and specific Lp(a) lowering than currently available may be required to demonstrate any vascular benefits in randomized trials.» (Erqou et al., 2009)
That causal leg is now held, via a Mendelian-randomization / instrumental-variable study of three disjoint Copenhagen cohorts (CCHS prospective, CGPS cross-sectional, CIHDS case-control), using the LPA KIV-2 size polymorphism as the genetic instrument — the number of KIV-2 repeats correlates inversely with Lp(a) level and explained 21% and 27% of Lp(a) variation in the two general-population samples (Kamstrup, 2009). The instrument avoids reverse causation and confounding by construction (genotype is fixed at conception and did not differ by measured cardiovascular risk factors), so a KIV-2-genotype/MI association consistent with the genotype’s effect on Lp(a) is evidence of a causal Lp(a)->MI relationship (Kamstrup, 2009).
- Genetic effect exceeds observational — the prediction CONFIRMED. On instrumental- variable analysis, genetically-elevated Lp(a) gave an HR for MI of 1.22 (95% CI 1.09-1.37) per doubling of Lp(a), while the observational Cox HR per doubling of plasma Lp(a) in the same cohort was 1.08 (95% CI 1.03-1.12) — the IV estimate was nominally larger than the observational one (Kamstrup, 2009). Both figures are on the same per-doubling scale, so this comparison is internal to the study; do not read either against ERFC’s per-3.5-fold (1-SD) RRs above, which are a different exposure contrast. The prior directional expectation — that a lifelong, undiluted, unconfounded genetic exposure would show a larger effect than a single-measurement observational association — was recorded before the evidence landed and is borne out (the lifelong-exposure and small-isoform-harm explanations for the gap are the source’s own (Kamstrup, 2009)).
- What it establishes, and what it does not. «Conclusion These data are consistent with a causal association between elevated lipoprotein(a) levels and increased risk of MI.» (Kamstrup, 2009) The authors are explicit that this is consistency-with-causality, not closure of the treatment question: final proof of causality still requires randomized clinical trials demonstrating reduced MI risk in response to Lp(a)-lowering therapy (Kamstrup, 2009). The loop therefore stays partly open: causal existence is established; whether lowering Lp(a) reduces events is not — no Lp(a)-lowering outcome trial is held. The dose-response MR that sizes how much lowering would be needed is now held -> see the magnitude leg below.
- Transportability caveat. Kamstrup studied only white individuals of Danish descent, and Lp(a) levels differ markedly between ethnic groups, so the causal magnitude may not transport to other populations (Kamstrup, 2009).
The magnitude leg — how much lowering is needed (type-F refinement)
Kamstrup establishes that Lp(a) causes MI; it does not size how much lowering would buy a clinically meaningful benefit. That magnitude is now held, via a pooled Mendelian-randomization analysis: individual-participant data from 5 studies (48 333 participants of European descent, including 20 793 with CHD) plus external validation from summarized data of 48 studies (CARDIOGRAMplusC4D: 62 240 CHD cases, 127 299 controls) (Burgess et al., 2018). A 43-variant LPA genetic score (explaining 51-63% of Lp(a) variance) was the instrument, and the exposure-outcome association was linearly proportional to the absolute change in Lp(a) mass concentration rather than to the proportional change — so the decision-relevant metric is mg/dL lowered, not percent lowered (Burgess et al., 2018).
The headline magnitude. Each 10-mg/dL lower genetically-predicted Lp(a) was associated with a 5.8% lower CHD risk (OR 0.942; 95% CI 0.933-0.951); external replication gave 5.2% (OR 0.948; 0.941-0.955) (Burgess et al., 2018). Using the same per-10-mg/dL genetic-score design for LDL-C, each 10-mg/dL lower LDL-C gave a 14.5% lower CHD risk (OR 0.855; 0.818-0.893). So mg-for-mg, Lp(a) is roughly a quarter as potent as LDL-C: a 38.67-mg/dL (1-mmol/L) LDL-C reduction has the same association with CHD risk as a 101.5-mg/dL (95% CI 71.0-137.0) Lp(a) reduction (external validation 109.1 mg/dL; 89.0-133.1) (Burgess et al., 2018). The Lp(a) effect is independent of LDL-lowering: stratifying on the statin-target HMGCR variant (rs12916) left the per-10-mg/dL Lp(a) OR nearly identical across the three genotype groups (0.945 / 0.939 / 0.945; P = .79), and the same held for PCSK9- and ezetimibe-mimicking variants (Burgess et al., 2018).
The Lp(a)-vs-LDL-C potency ratio is a within-source comparison, which is why it is safe to state:
| Parameter | Lp(a) (Burgess) | LDL-C (Burgess) | Same quantity? |
|---|---|---|---|
| Contrast | per 10-mg/dL genetically-predicted lower | per 10-mg/dL genetically-predicted lower | YES — identical scale/design |
| Instrument | 43-variant LPA score | LDL-C genetic score | both genetic scores, same MR method |
| Outcome | CHD (odds ratio) | CHD (odds ratio) | YES |
| Estimate | OR 0.942 (0.933-0.951) | OR 0.855 (0.818-0.893) | comparable -> 2.63 mg/dL Lp(a) per 1 mg/dL LDL-C |
The ratio holds because both estimates come from Burgess on the identical per-10-mg/dL genetic-score contrast against CHD. What must NOT be equated numerically are the cross-source Lp(a) figures: Burgess reports a per-10-mg/dL OR, Kamstrup a per-doubling HR (1.22 for MI), and ERFC a per-3.5-fold / 1-SD RR (1.16 for CHD) — three different exposure contrasts on two different outcome scales, so their point estimates do not sit on a common axis and are not compared here.
Why this is type-F, not a third independent (E) route. Burgess shares authors with both held
sources — Kamstrup, Tybjaerg-Hansen and Nordestgaard with the Kamstrup MR, and Di Angelantonio,
Danesh and Butterworth (via EPIC-CVD) with the ERFC observational IPD
(Burgess et al., 2018) — and it
re-uses the same Copenhagen cohorts and EPIC-CVD data. So it is not an independent convergence:
it bounds and operationalizes the already-held causal claim with a dose-response magnitude
(amplify-attenuate refinement), and is marked type-F, NOT [E-independent].
Why the null Lp(a)-lowering trials do not overturn causality
Burgess reconciles the causal genetics with the disappointing lowering trials. The completed RCTs (niacin, CETP inhibitors, PCSK9 inhibitors) lowered Lp(a) by only 20-35%, and «The median Lp(a) concentration among participants enrolled in these trials was approximately 12 to 20 mg/dL», so «a 30% reduction in Lp(a) concentration would translate into only a 3- to 6-mg/dL absolute reduction» — far below the ~100-mg/dL absolute change the MR implies is needed (Burgess et al., 2018). The predicted event reduction scales with absolute lowering: ~100 mg/dL -> ~22-25% CHD reduction over 3-5 years, 80 mg/dL -> ~18-20%, 50 mg/dL -> ~10-12%, so «only persons with very high Lp(a) concentrations are likely to benefit substantially from therapies that reduce Lp(a) concentration» (Burgess et al., 2018).
So the null trials are consistent with under-dosing, not with Lp(a) being non-causal: «The clinical benefit of lowering Lp(a) is likely to be proportional to the absolute reduction in Lp(a) concentration» (Burgess et al., 2018). The decision-change is for trial design and screening (enroll people with baseline Lp(a) >= 90-100 mg/dL; target 80-90% lowering), not yet for treatment.
But the loop is STILL open, and Burgess does not close it. The ~22-25% event reduction is a predicted trial estimate — modeled by porting the LDL-C statin-trial relationship onto Lp(a) under the explicit assumption that the two lipoproteins have similar cumulative CHD associations over time — not an observed outcome (Burgess et al., 2018). No Lp(a)-lowering outcome trial has yet achieved a ~100-mg/dL reduction and shown event reduction, and an MR is not an RCT. Two further limits: the estimate is agnostic to mechanism (if Lp(a)-lowering drugs act partly through non-Lp(a) channels, smaller absolute reductions might suffice) (Burgess et al., 2018), and all participants were of white European ancestry (Burgess et al., 2018) — so the magnitude may not transport to other ancestries (the source does not list this among its own stated limitations).
Two independent routes to the same causal claim [E-independent]
The Lp(a)->vascular-risk claim is now reached by two genuinely independent routes: ERFC’s observational IPD meta-analysis (Cambridge/Danesh school, 36 pooled prospective studies) and Kamstrup’s genetic instrumental-variable analysis (Copenhagen/Nordestgaard lab, three Danish cohorts). The author lists are disjoint, and the schools, methods (observational pooling vs genetic IV), and datasets do not overlap, so this is type-E independent-backing robustness, not a same-pipeline echo — the convergence raises confidence that the association reflects a real causal effect rather than shared observational confounding. Corroborated by Kamstrup et al. (JAMA 2009, chunk 01). (The now-held Burgess 2018 MR is NOT a third E route: it shares Kamstrup/Tybjaerg-Hansen/Nordestgaard with Kamstrup and Di Angelantonio/Danesh/Butterworth with ERFC, and re-uses the same cohorts — so it lands as the F-refinement on magnitude above, not as independent backing.)
Confidence held at medium, re-judged. The causal-existence claim is now well-warranted: two
independent gold/high routes (observational IPD + genetic IV) converge, which strengthens the
warrant over the single-source opener. But the page’s confidence spans the whole
decision, and the binding cap is the open treatment loop — no Lp(a)-lowering intervention is shown
to reduce events — so the warrant does not yet support a validated treatment target, and
medium is retained rather than raised. Burgess 2018 adds the dose-response magnitude (a type-F
refinement, not a third independent route) and explains why the null lowering-RCTs are uninformative,
but it does not lift the cap: its 22-25% event-reduction figure is a predicted trial estimate,
not an observed one, so medium still holds.
Vascular-specificity — a companion contrast, NOT independent corroboration
Lp(a)‘s confinement to vascular outcomes (null for cancer and other nonvascular death) contrasts with the broader marker C-reactive protein. That contrast is a useful specificity note, but it comes from the same collaboration (the ERFC CRP IPD, erfc2010crp), so it is not independent (type-E) backing — a shared body/analysis pipeline defeats independence by construction. Treat it as a same-source companion observation, not as convergent evidence . -> Surrogate Outcomes (Lp(a) as a risk marker vs a steering target).
What this page does NOT establish
- The nucleus of a
lipoprotein-acluster now holds three sources — the observational association (gold-tier IPD), the genetic causal leg (Kamstrup MR), and the dose-response magnitude (Burgess MR, type-F). The shape is now partly resolved (linear in absolute Lp(a) over the studied range; the >200-mg/dL 3-to-4-fold lifetime-risk figure is an extrapolation, not directly observed). What remains open, above all, is any Lp(a)-lowering treatment benefit (henceconfidence: medium). - The loop is open. No Lp(a)-lowering intervention is shown here to reduce events; the page establishes a prognostic association, a causal effect on MI, the absolute-lowering magnitude needed, and why the existing null trials under-dosed — but not a validated treatment target. The predicted 22-25% event reduction at ~100-mg/dL lowering is modeled, not observed. Causal existence plus a dose-response estimate is still not treatment efficacy.