Your cholesterol panel has four numbers. Only one of them is known to cause disease.

A standard lipid panel reports total cholesterol, LDL cholesterol, HDL cholesterol and triglycerides, and the popular shorthand — "bad" and "good" cholesterol — implies the two matter in symmetrical, opposite ways. They do not. The evidence that LDL causes atherosclerotic cardiovascular disease is about as strong as evidence in medicine gets; the evidence that raising HDL prevents it collapsed when it was tested directly.

That asymmetry is the single most useful thing to understand about the panel.

What the LDL evidence consists of

In 2017 the European Atherosclerosis Society convened a consensus panel to ask whether the association between LDL and atherosclerotic cardiovascular disease meets the criteria for causality rather than correlation [s1]. The panel assessed the totality of the evidence across four independent kinds of study: rare genetic mutations, prospective cohorts, Mendelian randomisation, and randomised trials of LDL-lowering drugs [s1].

Each pointed the same way. Rare mutations that impair LDL receptor function raise LDL cholesterol and produce a dose-dependent increase in cardiovascular risk; rare variants that lower it are associated with correspondingly lower risk [s1]. Separate meta-analyses of over 200 prospective cohort studies, Mendelian randomisation studies and randomised trials — together covering more than 2 million participants, over 20 million person-years of follow-up and over 150,000 cardiovascular events — showed what the panel called a remarkably consistent dose-dependent log-linear association between how much LDL cholesterol the vasculature is exposed to and the risk of disease, with the effect increasing the longer the exposure lasts [s1].

The panel's conclusion is unusually flat for a consensus document: the evidence "unequivocally establishes that LDL causes ASCVD" [s1]. The practical implication the panel drew is that any mechanism of lowering LDL particle concentration should reduce risk roughly in proportion to the absolute reduction achieved and the duration of that reduction, provided the drop in LDL cholesterol matches the drop in LDL particle number and the drug has no competing harms [s1].

What happened when the same tests were applied to HDL

Observational data on HDL looks, at first, just as convincing. In the studies pooled for a 2012 Mendelian randomisation analysis published in The Lancet, each standard-deviation increase in HDL cholesterol was associated with a lower risk of myocardial infarction, with an odds ratio of 0.62 (95% CI 0.58 to 0.66) [s2].

Then the researchers used genetics to ask whether that association was causal. Carriers of one variant in the endothelial lipase gene, LIPG Asn396Ser, present in 2.6% of the population, had HDL cholesterol 0.14 mmol/L higher than non-carriers with otherwise similar risk factors — a difference that, on the observational relationship, should have cut heart attack risk by 13% [s2]. It did not. The odds ratio for myocardial infarction in carriers was 0.99 (95% CI 0.88 to 1.11) [s2]. A broader genetic score built from 14 variants that affect HDL and nothing else gave the same answer: an odds ratio of 0.93 (95% CI 0.68 to 1.26) per standard-deviation increase [s2].

The same paper ran LDL as a positive control, and there the genetics and the epidemiology agreed — observational odds ratio 1.54 (95% CI 1.45 to 1.63) per standard deviation, genetic odds ratio 2.13 (95% CI 1.69 to 2.69) [s2]. The method works. HDL simply failed it.

The drug trial that closed the argument

Genetics is indirect. The direct test came from HPS2-THRIVE, which randomised 25,673 adults with established vascular disease, already on statin therapy, to extended-release niacin with laropiprant or placebo [s3]. Over a median 3.9 years the niacin group had HDL cholesterol an average of 6 mg/dL higher and LDL cholesterol 10 mg/dL lower than placebo — and no significant difference in major vascular events, which occurred in 13.2% of the niacin group and 13.7% of the placebo group (rate ratio 0.96, 95% CI 0.90 to 1.03, P=0.29) [s3].

Worse, the drug caused harm. Serious disturbances in diabetes control were 3.7 percentage points more common than with placebo, new diabetes diagnoses 1.3 points more common, and there were excesses of serious gastrointestinal, musculoskeletal and skin events, plus unexpected excesses of infection (1.4 points) and bleeding (0.7 points) [s3].

How to read the panel, then

A high LDL number is a marker of an exposure with an established causal role, and the relevant quantity is cumulative — how high, for how long [s1]. A high HDL number is a marker of something, but the two most rigorous tests available, natural genetic randomisation and a large randomised trial, both found that moving it did not move cardiovascular risk [s2] [s3]. That does not make HDL meaningless as a statistical predictor; it means the arrow does not point from HDL to disease in the way the label "good cholesterol" implies.

Neither of those facts tells any individual what their own numbers require. Absolute cardiovascular risk depends on age, blood pressure, smoking, diabetes, family history and kidney function as well as lipids, and guidelines set treatment thresholds on that combined risk rather than on a single value. What the evidence does settle is which number the arithmetic is actually about.