ApoB, LDL-C or non-HDL-C — which should I be tracking, and what is discordance?

I have three numbers on my report that all claim to measure roughly the same risk and they do not agree about how I am doing. LDL-C 2.6 mmol/L, which is described in the comment field as near target. Non-HDL-C 4.1 mmol/L, which by the usual rule of "LDL target plus 0.8" would be well above target. ApoB 1.18 g/L, which every source I can find describes as clearly elevated.

My triglycerides are 3.4 mmol/L, which I suspect is the reason the three numbers disagree, but I do not understand the mechanism. I would like to know:

• Which of the three is the best measure, and on what grounds — cost, availability, evidence, or something else?
• What "discordance" formally means and how to tell whether I have it.
• Whether the disagreement between my three numbers is telling me something, or is just an artefact of how they are computed.

I have no interest in collecting more numbers for their own sake. I want to know which one I should be looking at in five years' time, and whether the other two are then redundant.

ApoB is the best of the three on mechanistic and genetic-epidemiological grounds, non-HDL-C is the best value because it is free and already on your report, and calculated LDL-C is the weakest — and your case is a clean example of why. You have discordance, it is real, and the ApoB number is the one telling you the truth.

What each one measures

• LDL-C is the mass of cholesterol carried inside LDL particles, per litre. It says nothing about how many particles that cholesterol is distributed across.
• Non-HDL-C is the cholesterol in every atherogenic lipoprotein: LDL, VLDL, IDL, remnants, chylomicron remnants and Lp(a). Computed as total cholesterol minus HDL-C.
• ApoB is a count. Every atherogenic particle — LDL, VLDL, IDL, remnant, Lp(a) — carries exactly one molecule of apolipoprotein B-100. So an ApoB concentration is a direct measure of particle number, in a fixed stoichiometry, with no assumptions.

That last property is the whole argument. Atherogenesis is driven by particles entering and being retained in the arterial wall, and the rate of entry scales with particle number rather than with the cholesterol cargo each particle happens to be carrying. Two people with identical LDL-C can carry it in 900 large cholesterol-rich particles or 1400 small depleted ones, and the second person has the higher risk despite the identical LDL-C.

Why your three numbers disagree

Because at a triglyceride of 3.4 mmol/L your LDL particles are triglyceride-enriched and cholesterol-depleted, and your remnant pool is large. Two things follow:

First, the particle count required to carry 2.6 mmol/L of LDL cholesterol is higher than it would be at normal triglycerides. Hence ApoB 1.18 g/L against an LDL-C that looks acceptable.

Second, the Friedewald calculation is subtracting a triglyceride-derived estimate of VLDL cholesterol that is too large at high triglycerides, so your calculated LDL-C is biased downwards. Work it: your VLDL-C term is 3.4 ÷ 2.2 = 1.55 mmol/L. Your non-HDL-C of 4.1 minus your calculated LDL-C of 2.6 gives exactly that 1.5. The equation has assigned 1.55 mmol/L of cholesterol to VLDL by assumption, not by measurement, and if the true figure is 1.2 then your real LDL-C is nearer 2.9.

So the disagreement is not an artefact in the sense of being meaningless. It is a signal, and the signal is: cholesterol-based measures are understating your atherogenic particle burden.

Formal discordance

Discordance means being in different percentile categories on two measures for the same person. Formally you compare percentile ranks in a reference population; informally, look for one of these patterns:

• LDL-C at target, ApoB not. Your case. Associated with high triglycerides, insulin resistance, metabolic syndrome, and with post-weight-loss states. It is the more common direction and the one that matters, because it means risk is being underestimated.
• ApoB at target, LDL-C not. Fewer, larger, cholesterol-rich particles. Less common, and generally the reassuring direction.
• Non-HDL-C high with LDL-C at target. Points at the remnant pool. Your remnant cholesterol is non-HDL-C minus LDL-C = 4.1 − 2.6 = 1.5 mmol/L, which is substantial. Remnant cholesterol has independent evidence as a causal risk factor and is free to compute.
• Everything at target with high Lp(a). Lp(a) contributes to both non-HDL-C and ApoB but at a low particle count, so a very high Lp(a) can hide inside otherwise acceptable numbers. This is a genuine gap and is the reason to measure it once.

Thresholds, so the numbers mean something

Approximate correspondences, all of which need individualising to your actual risk by someone who knows your history:

• ApoB 1.0 g/L (100 mg/dL) sits roughly at the population 50th to 60th percentile in Western cohorts. Around 0.8 g/L is often quoted for people at raised risk, and 0.65 g/L or lower for those at very high risk.
• Non-HDL-C targets are conventionally set at the LDL-C target plus 0.8 mmol/L (30 mg/dL). So an LDL-C target of 1.8 corresponds to a non-HDL-C target of 2.6. Your 4.1 is a long way from that.
• ApoB in g/L multiplied by 100 gives mg/dL. Your 1.18 g/L is 118 mg/dL.

Which to track in five years

ApoB, with non-HDL-C as the free running check between ApoB draws. The evidence base for ApoB as the superior discriminator is strongest in the genetic literature: Mendelian randomisation analyses comparing variants that lower triglycerides against variants that lower LDL-C found that the association with coronary risk was proportional to the change in ApoB, not to the change in either lipid measure separately [1]. That is close to the cleanest available demonstration that the particle count is the causal quantity.

The practical counterargument for LDL-C is not that it is a better measure. It is that every outcome trial and every guideline threshold was built on it, so a clinician has decades of calibrated experience with LDL-C numbers and comparatively little with ApoB. That is a real reason to keep reporting it and not a reason to prefer it.

ApoB's practical virtues, since they are usually undersold: standardised against an international reference material, analytical CV of roughly 3 to 5%, no fasting requirement, no equation, valid at any triglyceride concentration, and typically inexpensive. There is very little argument against measuring it beyond local availability.

edited 17 Feb 2026 by dana_wexler — clarified the distinction between purity and content

Worth separating two questions that get merged: which measure best estimates risk, and which best tracks a response to treatment. The answer is not the same, and this matters specifically after large weight loss.

For risk estimation at a single point in time, the argument in the answer above holds and ApoB wins.

For tracking a change, the relevant property is how much of the observed difference is signal. Compare reference change values, computed as 2.77 × sqrt(CVa² + CVi²):

• ApoB: CVa about 4%, CVi about 7%. sqrt(16 + 49) = sqrt(65) = 8.1. RCV = 22%.
• Total cholesterol: CVa about 2.5%, CVi about 6%. sqrt(6.25 + 36) = 6.5. RCV = 18%.
• Non-HDL-C: propagates the errors of two measures, so somewhat worse than total cholesterol alone, roughly 20 to 25%.
• Triglycerides: CVa about 4%, CVi about 21%. sqrt(16 + 441) = 21.4. RCV = 59%.
• Calculated LDL-C: inherits error from three measurements plus the model error of the divisor assumption. Effectively the noisiest of the cholesterol-based measures, and the noise grows with triglycerides.

So for detecting a change of a given size, total cholesterol and ApoB are the tightest, and calculated LDL-C is the loosest. Combined with the fact that calculated LDL-C is also the most biased during a triglyceride shift, there is not much left to recommend it as a serial monitoring tool.

Practical consequence for a weight-loss course: an ApoB fall from 1.18 to 1.05 g/L is an 11% change, inside the 22% RCV, and therefore not established from two draws alone. A fall to 0.90 is 24% and clears it. This is worth knowing before you conclude that a 10% improvement has happened; it may have, and you cannot demonstrate it with two measurements.

The way round it is the same as everywhere else: three or more points over enough time, one lab, standardised conditions, and read the slope rather than the last pair.

A brief note on what none of the three tells you, since the question asked what becomes redundant.

All three measures are concentration measures of a circulating pool. What determines an artery's fate is cumulative exposure — concentration integrated over time — and none of your numbers contains a time axis. Someone with an ApoB of 1.18 g/L at 32 and someone with the same value at 62 have very different accumulated exposures, and the same intervention has very different expected value in the two cases.

This has two practical implications.

First, a single panel is a poor input to a risk decision on its own, which is why risk calculators take age, sex, blood pressure, smoking and diabetes status alongside a lipid value, and why a coronary calcium score or a carotid ultrasound sometimes settles a decision that no lipid panel can. If the question is whether to start lipid-lowering therapy, imaging that shows whether the cumulative exposure has already deposited anything is often more decisive than another lipid measurement.

Second, a lipid improvement achieved late does not undo earlier exposure, though it does stop adding to it. Framing a post-weight-loss ApoB fall as "risk reversed" is wrong; "risk accrual slowed" is right. That is not a reason to be gloomy about it — slowing accrual is the entire mechanism by which lipid-lowering works — but it does explain why absolute benefit in the trials depends so heavily on baseline risk and duration rather than on the size of the lipid change alone.

None of this is advice about whether you personally need treatment. With triglycerides at 3.4 and ApoB at 1.18 g/L there is a genuine conversation to be had with a clinician about secondary causes, family history and whether pharmacological lipid lowering is warranted, and it is not a conversation a lipid panel can have on its own.