Genetic instruments for LDL cholesterol affect cardiovascular risk only through LDL and not through confounding pleiotropy
16 events · 8 assessments · 7 decisions
Reassessed
subclaim_change trigger: the supporting subclaim on MR-versus-statin-trial concordance was reworded to state the reconciliation explicitly and strengthened from supported (0.6) to verified (0.85), with Curator confirmation of no merge/split and correct SUPPORTS edges. Assessed the change as confirmatory and modestly strengthening for one of four supporting legs in the 'Robustness to pleiotropy' argument, not the load-bearing leg (which is cross-gene concordance). Status held at supported; the principled ceiling to verified (untestable exclusion restriction plus confirmed instrument pleiotropy) is unchanged. Confidence nudged 0.87 to 0.88, credence ~0.87. Re-recorded both argument evaluations to track the strengthened premise ('Robustness to pleiotropy' holds; 'Pleiotropy of lipid variants' holds_with_caveats). No structural change; decomposition already complete and adequate. Importance reaffirmed at 0.4, contestation 0.3. Light-touch pass appropriate given the confirmatory nature and settled-leaning subject; no external search needed to change the verdict. Not notifying dependents: status unchanged and the confidence nudge is immaterial to any dependent.
Reassessed: still Supported
verdict confidence 0.87 → 0.88 · credence 0.87
Reassessed
subclaim_change trigger: the multivariable-MR premise (conditioning on HDL and triglycerides leaves LDL's independent effect intact) moved from unassessed to verified (0.88). The prior parent assessment had flagged this as its main remaining yield, so the change is confirmatory. Kept status at supported and nudged confidence 0.85 to 0.87; credence ~0.86. Not verified because the exclusion restriction remains untestable in full and instrument pleiotropy on other lipids is verified. Assessed the apoB refinement carried with the trigger: conditioning on apoB attenuates LDL-c's independent effect, but this is a within-atherogenic-pathway question about which lipid measure is proximate (mediation/collinearity), not confounding through a non-lipid pathway, so it refines interpretation without bearing against this claim. Re-evaluated both named arguments against current premise standings (Robustness to pleiotropy: holds; Pleiotropy of lipid variants: holds_with_caveats). No structural change: the apoB point concerns a distinct question and did not warrant a new subclaim on this node. Reaffirmed importance at 0.4, contestation 0.35. Did not notify the single dependent (MR shows genetically lower LDL causally reduces CVD risk, verified via an assumes edge): the change is a same-status confidence nudge, not material.
Reassessed: still Supported
verdict confidence 0.85 → 0.87 · credence 0.86
Reassessed (status unchanged)
Triggered by the first assessment of the trial-agreement subclaim (supported, 0.6), which clarified that MR and randomized-trial per-unit LDL effect estimates diverge ~3-fold in raw magnitude but reconcile once cumulative exposure duration is accounted for. This resolves an item the prior parent assessment had carried as 'consistent though unassessed,' so it is confirmatory. Materially it is reassuring for the pleiotropy question specifically: an unexplained genetic-vs-trial divergence would be a candidate signature of a non-LDL pathway, whereas a divergence fully absorbed by the known lifelong-vs-short-term exposure mechanism leaves the residual consistent. Status held at supported, confidence 0.85, credence ~0.85. Updated the reader-facing assessment and reasoning trace to reflect that trial agreement is now assessed (in the triangulation sense, not per-unit equality). Re-evaluated the 'Robustness to pleiotropy' argument (holds) since its trial-agreement premise changed standing, and re-recorded the 'Pleiotropy of lipid variants' argument (holds_with_caveats) unchanged to confirm. Importance left at 0.4: methodologically consequential but not an actively contested crux within the LDL-causality literature. No dependent notification, as the parent status and confidence did not move. Remaining yield: formal assessment of the multivariable-MR premise, still unassessed.
Reassessed: still Supported
verdict confidence 0.85 · credence 0.85
Reassessed: still Supported
verdict confidence 0.85 · credence 0.85
Reassessed (status unchanged: supported); added one supporting subclaim; updated argument written form and both evaluations
Trigger: subclaim 111e0a7d (LDL-lowering variants frequently also affect triglycerides/HDL) newly VERIFIED. The prior assessment already treated this pleiotropy as real, so the change is confirmatory. Its material consequence is that the claim's support can no longer rest on pleiotropy being absent; it must rest on the pleiotropy being corrected for. To make that explicit I added a supporting subclaim under the 'Robustness to pleiotropy' argument — that multivariable MR conditioning on HDL and triglycerides shows LDL retains an independent effect (match_claim confirmed novel; created 0cf3b72c) — since that is the correction aimed squarely at the verified lipid-trait pleiotropy. Rewrote the 'for' argument written form to include it, and re-evaluated both arguments against the current premise standings (for: holds; against: holds_with_caveats). Verdict held at supported/0.85, credence ~0.85 for the operative reading; not verified because the exclusion restriction is untestable in full. Canonical form left unchanged: it fairly states the exclusion restriction as debated, and rewording toward 'methods adequately correct for pleiotropy' would be a different claim. Did not notify the sole dependent (567d349b, already verified on convergent evidence) because the verdict did not materially change. Considered linking related claim 1ba36585 (concordance across pleiotropy-robust methods) but it overlaps the existing directional-pleiotropy sensitivity subclaim, whereas the multivariable-MR claim is the sharper, on-point addition.
Reassessed: still Supported
verdict confidence 0.85 · credence 0.85
Reassessed
Triggered by subclaim_change: the supporting subclaim on concordance of per-unit LDL-coronary effect estimates across distinct LDL genes moved from unassessed to VERIFIED (0.9). The prior assessment already treated this concordance as borne out, so the change is confirmatory, strengthening the leading leg of the 'for' argument without altering the balance. Kept status supported; nudged confidence 0.82 to 0.85 because the strongest supporting premise is now independently verified, credence ~0.87. Not raised to verified: the exclusion restriction is an idealization that cannot be established with certainty, and individual-variant pleiotropy is real (captured by the 'against' subclaim). Also closed a standing gap by recording evaluations for both named arguments (both previously null): 'Robustness to pleiotropy' holds, resting mainly on the now-verified concordance premise; 'Pleiotropy of lipid variants' holds_with_caveats, establishing possibility of bias only, blunted by concordance and drug-target/MVMR corrections. Left importance at 0.4 (foundational MR assumption, consequential but largely uncontested among experts). No dependent notification: status unchanged and confidence shift is immaterial. Did not add or restructure subclaims; decomposition remains apt.
Reassessed (status unchanged) and backfilled argument evaluations
Trigger: subclaim c9205783 ("MR sensitivity analyses find no evidence of directional pleiotropy") was assessed SUPPORTED at 0.85 with caveats (MR-Egger low power; real non-directional pleiotropy handled by multivariable/outlier-robust methods). This confirms the treatment the existing assessment already assumed, so it is absorbed without a status change: the claim remains SUPPORTED. I nudged confidence 0.82→0.83 to reflect the now-formal subclaim status and refreshed the reader-facing assessment and reasoning trace to make the three convergent support lines and the bounded pleiotropy caveat explicit.\n\nThe substantive gap addressed this pass was that both named arguments had written forms but no evaluations. Recorded them anchored to this verdict: 'Robustness to pleiotropy' holds (weight rests on cross-gene concordance and trial agreement; null pleiotropy tests add support but have limited power); 'Pleiotropy of lipid variants' holds_with_caveats (valid threat that makes the strict claim an idealization, but bounded by corrective methods, so it explains supported-not-verified without contradicting).\n\nImportance left at 0.4: a notable methodological premise (the MR exclusion restriction) with one verified dependent and low-to-moderate live contestation; not central. No dependent notification: status unchanged and the confidence nudge is immaterial to the verified MR causal-effect claim that presupposes this one.
Reassessed: still Supported
verdict confidence 0.83 → 0.85 · credence 0.87
Reassessed: still Supported
verdict confidence 0.82 → 0.83 · credence 0.85
Structured and assessed
First pass on the MR exclusion-restriction (no-confounding-pleiotropy) claim for LDL-CVD. Decomposed into a 'for' argument (Robustness to pleiotropy, pre-existing) grouping three novel subclaims — no directional pleiotropy in sensitivity analyses, concordance across distinct LDL genes, and agreement of genetic with randomized-trial estimates — and a new 'against' argument (Pleiotropy of lipid variants) holding one novel subclaim on the widespread pleiotropy of lipid variants. All four dependencies confirmed novel by the Matcher and created; wrote both arguments' forms. Set importance 0.4 (contestation 0.4): a methodological crux under a verified parent, but bounded consequence since the LDL-causes-CVD conclusion rests on multiple independent lines. Two web searches confirmed the state of the literature (MR-Egger/weighted-median null for directional pleiotropy; NPC1L1/HMGCR/PCSK9 concordance; drug-target MR attributing variant CVD effects entirely to LDL). Assessed 'supported' (confidence 0.82, credence 0.85): the operative claim that the estimate is not driven by confounding pleiotropy is unusually well evidenced, but the strict 'only through LDL' is an idealization given real variant-level pleiotropy and the untestability of the assumption in full.
Assessed Supported
verdict confidence 0.82 · credence 0.85
This is the key assumption that lets Mendelian randomization read the genetic LDL–cardiovascular association as causal: that variants used as instruments raise or lower cardiovascular risk through LDL alone, not through side effects on other pathways (horizontal pleiotropy). The available evidence supports it unusually well for LDL, though the assumption can never be fully proven because some pleiotropy is undetectable in principle. Three lines converge in its favor. Formal sensitivity analyses that are designed to detect pleiotropy (MR-Egger, weighted-median and related methods) find no evidence of directional pleiotropy for the LDL–coronary relationship. Instruments in genes that lower LDL by entirely different mechanisms (for example HMGCR, PCSK9 and NPC1L1) yield concordant per-unit effects on coronary risk, a convergence that gene-specific pleiotropy would be unlikely to produce; drug-target analyses of PCSK9 and HMGCR variants attribute their cardiovascular effect entirely to LDL lowering. And the genetic per-unit estimates agree with the effect seen in randomized statin, ezetimibe and PCSK9-inhibitor trials, an external check that the signal runs through LDL. The genuine caveat is that many individual lipid variants are pleiotropic, affecting triglycerides, HDL, ApoB or Lp(a), so the strict "only through LDL" statement does not hold variant by variant. The field addresses this with cis (drug-target) instruments and multivariable MR, and the LDL causal signal survives those corrections. The credible reading is therefore that the LDL–CVD estimate is not driven by confounding pleiotropy, rather than that no instrument has any off-target effect. The assessment would weaken if a specific off-LDL pathway were shown to carry a material share of the genetic effect.
Claim entered the graph