This recently published Risk-Weighted ApoB (RW-ApoB) metric is an attractive advance in thinking about the relationship between Lipids and ASCVD. It is an advance in part because it is derived from commonly available existing metrics. However, I find that the model has several structural problems that limit its validity and may inflate its apparent predictive superiority. I will summarize these only briefly here.
-
Mechanistic conflation: The formula treats “LDL atherogenicity” as its reference unit, when the actual mechanism of arterial injury is ApoB-mediated proteoglycan retention which is a property of the ApoB molecule itself, shared identically across all three particle classes. The correct reference unit is therefore ApoB retention capacity, not an LDL particle class effect.
-
The ApoB discount: The formula’s ApoB coefficient (0.736) is less than 1.0, meaning it discounts raw ApoB for individuals with low TG and Lp(a). This is mechanistically indefensible — low TRL and Lp(a) add no excess risk, but they do not reduce the baseline retention risk of ApoB particles already present. This coefficient is in fact a calibration artifact, scaled to match population medians, and should not be interpreted as an atherogenicity weight. This is largely a scaling problem which is easily addressed.
-
Excluded middle and systematic bias: The MESA validation cohort excluded subjects with ApoB < 50 mg/dL — precisely the subpopulation where RW-ApoB diverges most from raw ApoB and where the formula’s error terms would be largest. Additionally, the derivation cohort (UK Biobank) used non-fasting TG while all three validation cohorts used fasting samples; since TG drives the TRL term which carries a 4.5× atherogenicity multiplier, this systematic pre-analytical difference propagates amplified error across cohorts. Fixed-factor Lp(a) unit conversion (mg/dL ==> nmol/L) in FOS and SCAPIS introduces further error amplified by the 6.5× Lp(a) weight. None of these error propagation pathways were subjected to sensitivity analysis.
A more important missing variable, however, is arterial wall inflammation. We know that two individuals with near-identical lipid profiles can have markedly different atherosclerotic outcomes and that the divergence can be explained by inflammatory metrics. This strongly suggests that systemic inflammatory state is a primary permissive condition for lipid-mediated arterial injury, and not merely an additive co-risk. This is supported by the CANTOS, COLCOT, and LoDoCo2 trials, each of which reduced MACE through anti-inflammatory mechanisms without lipid modification. The appropriate inflammation biomarker is not hsCRP (several reasons including being non-specific and confounded by systemic conditions but possibly material when low) but myeloperoxidase (MPO), which is mechanistically upstream it is the leukocyte-derived enzyme that generates hypochlorous acid at the vessel wall, creating oxLDL and initiating the foam cell cascade. MPO predicts incident MACE independently of hsCRP (HR ~1.71 after full adjustment), and its companion marker Lp-PLA2 captures the downstream intra-plaque amplification step. Together they bracket the arterial oxidation-inflammation pathway.
To enhance the RW-ApoB, I propose a two-stage revision. First, correct the mechanistic base so that ApoB is an inviolable floor, excess atherogenicity from TG and Lp(a) particles is added above it rather than redistributed via a sub-unity coefficient. Second, apply inflammation as a multiplicative modulator rather than an additive term, reflecting its role as a permissive gate rather than an independent parallel risk source. The proposed Risk-Adjusted ApoB (RA-ApoB) takes the form:
RA-ApoB = [ApoB + Δ_TRL + Δ_Lp(a)] × f(MPO)
where Δ_TRL and Δ_Lp(a) represent only the incremental excess atherogenicity above the ApoB retention baseline (using multipliers of 3.5 and 5.5 respectively, i.e., the published 4.5× and 6.5× minus 1.0), and f(MPO) is a calibrated function anchored at 1.0 at population median MPO, approaching a non-zero floor (~0.25) at very low inflammation to preserve the FH edge case, and scaling upward at elevated MPO. The elasticity coefficient [b] and floor value require empirical calibration; MESA contains the necessary measurements and would be a natural starting cohort.
One circularity concern deserves acknowledgment: some arterial inflammation may itself be initiated by retained lipid particles, particularly at extreme lipid burdens (as in familial hypercholesterolemia). Our framework treats this as a positive feedback effect confined to the high-lipid tail of the distribution, not as invalidating the primary directionality. A testable prediction follows: the inflammation modulator should explain disproportionately more variance in the middle of the lipid distribution and less at the extreme high-lipid tail. That asymmetry, if confirmed, would validate the biological model.
I invite comment on this model as I am still developing it. Two questions are on my mind at the moment: (1) whether the MPO literature is strong enough to justify it as the primary inflammation input versus a composite score; if not, the fallback may be , and (2) what the right study design looks like to empirically fit [b] and the floor. The MESA dataset is an obvious starting point given it has ApoB, TG, Lp(a), and inflammatory markers with ~12 years of adjudicated CHD outcomes. If this model is not sufficiently predictive at end-points, we may need to modify the MPO parameter to reflect MPO and Lp-PLA2. I think not, though, because Lp-PLA2 is downstream.