hs-CRP: One High Reading Means Retest, Not Panic

The full story

A high-sensitivity CRP that comes back above 2 mg/L looks like a verdict. It arrives in the same column as your cholesterol, with a reference range beside it and a small flag in the margin, and the natural response is to start reorganising your life around it. The correct response, most of the time, is to book a second test.

Not because the assay is bad. The assay is excellent. It is because hs-CRP is one of the twitchiest analytes in routine use, and nearly everything worth knowing about interpreting it follows from that single fact.

What one reading buys you

The prognostic association is not in dispute. A 2010 individual-participant meta-analysis in the Lancet pooled 160,309 people without vascular disease from 54 prospective studies, capturing 27,769 outcomes. Each standard deviation higher log CRP, which is a three-fold higher concentration, carried a coronary heart disease risk ratio of 1.63 after adjustment for age and sex.

Then watch it shrink. Adjust for conventional risk factors and it falls to 1.37. Adjust further for fibrinogen and it falls to 1.23. The same analysis found an association of similar magnitude with non-vascular mortality, 1.54, which the authors flag as making CRP's specificity for vascular disease unclear. A marker that predicts everything is mostly telling you something general about how unwell you are. MedlinePlus states the clinical version plainly: CRP tells you how much inflammation is in the body, not what is causing it or where it is.

The noise is not a rounding error

In a 2001 study in Clinical Chemistry, 113 adults were measured five times at quarterly intervals across a year. Only 63 percent of first and second hs-CRP measurements landed in the same quartile, though 90 percent were within one quartile of each other. That is close to total cholesterol at 60 percent, and log-transforming the data did not rescue it. Put bluntly: a single hs-CRP misclassifies your risk quartile roughly a third of the time.

A 2007 study in the Scandinavian Journal of Clinical and Laboratory Investigation quantifies the same problem differently. Across serial samples from 38 healthy subjects, within-subject biological variation was 50 percent for hs-CRP, against 8.2 percent for osteopontin and 8.8 percent for soluble RANKL measured in the same people. In 300 Danish blood donors with a median age of 43, median hs-CRP was 1.0 mg/L with a 95th percentile of 12 mg/L. The distribution has a long right tail, and you can land in it by accident.

The most quietly unsettling dataset is the smallest. A 2009 report in the Israel Medical Association Journal took 23 patients with stable coronary disease on unchanged statin therapy, with no change in symptoms, health or medications, and measured hs-CRP twice within 14 days, a mean of 7 days apart. It fell by more than 1.0 mg/L in 10 of them and rose by more than 1.0 mg/L in 4. The changes were unrelated to white cell count or to any measurable cytokine. Nothing happened to these people. Their inflammation number moved anyway.

And yet it tracks over decades

Here is the paradox that makes hs-CRP genuinely useful rather than merely noisy. A 2009 analysis in the American Journal of Medicine followed 2,409 Framingham Heart Study Offspring participants across three exams spanning 20 years, using the same assay method throughout. Between the first two exams, averaging 16 years apart, somewhere between 23 and 47 percent of men and 27 and 49 percent of women stayed in the same CRP quintile, and a further 24 to 51 percent moved only one quintile. Both the long-term and the four-year variability of CRP were significantly smaller than that of plasma cholesterol in the same people.

So your CRP wanders week to week, and your underlying CRP level is a stable feature of you across decades. Those statements are compatible, and together they define the measurement problem: you are trying to estimate a stable quantity through a noisy channel. The fix is not a better laboratory. It is more than one sample.

What the evidence does not show

It does not show that CRP is the thing to lower. The cleanest test of that is genetic. A 2011 Mendelian randomisation analysis in the BMJ covered 47 studies and 194,418 participants, of whom 46,557 had coronary heart disease. CRP gene variants that raise CRP concentration by up to 30 percent per allele were unrelated to coronary heart disease, giving a risk ratio of 1.00 per standard deviation of genetically higher log CRP, significantly discordant from the 1.33 seen for measured CRP in the same framework. If the molecule caused the disease, a lifetime of genetically higher levels should have caused more of it.

The authors' conclusion: CRP itself is unlikely to be even a modest causal factor for coronary heart disease.

That distinction is not pedantry, and the drug trials prove it. A 2019 randomised trial in the New England Journal of Medicine gave low-dose methotrexate to 4,786 patients with prior heart attack or multivessel coronary disease plus diabetes or metabolic syndrome. It did not lower interleukin-1-beta, interleukin-6 or CRP, and it did not reduce cardiovascular events, while raising liver enzymes, lowering leucocyte counts and increasing non-basal-cell skin cancers. Anti-inflammatory intent was not enough. The drug had to actually move the marker.

Two trials that did move it did work, with caveats attached. A 2017 randomised trial in the same journal treated 10,061 post-heart-attack patients with an interleukin-1-beta antibody that lowered hs-CRP substantially more than placebo without changing lipids at all, and reduced the primary composite by 15 percent. That is real evidence that lowering inflammation lowers events. But all-cause mortality was unchanged, fatal infection was more common, and the lower of the two doses missed entirely. A 2020 randomised trial of low-dose colchicine in 5,522 patients with chronic coronary disease cut the primary composite from 9.6 to 6.8 percent over a median 28.6 months, with death from non-cardiovascular causes numerically higher. And the celebrated 2008 statin trial that enrolled 17,802 apparently healthy people on the basis of an hs-CRP of 2.0 mg/L or more reduced the primary composite by 44 percent, but it lowered LDL cholesterol by half at the same time, so it cannot isolate an inflammation-lowering effect at all.

The parameter that matters here is the sampling protocol

For most tests, the spec that matters is the assay. For this one it is when the blood is drawn. Two readings taken at least two weeks apart, when you are well and free of any intercurrent infection, gets you far closer to your actual level than one reading ever will, and it costs a second test rather than a second opinion.

It is also worth knowing what raises CRP for reasons that have nothing to do with your arteries. MedlinePlus lists bacterial and viral infection, and separately notes smoking, environmental toxins, insomnia, depression and obesity. A head cold the week before your draw is not a cardiovascular finding.

Two more interpretive anchors. The 2 mg/L threshold that keeps appearing is not a biological cliff; it is the enrolment criterion two large trials used to select people likely to benefit from treatment. And a double-digit result sits out near the 95th percentile of healthy blood donors, which makes it a reason to look for something acute rather than to compute a decade of vascular risk from it.

The honest summary is that hs-CRP is a real risk marker that is not a real drug target, produced by an excellent assay measuring a restless biology. That combination rewards patience and punishes single data points. If your first number came back high, the useful next move is not a supplement stack or a diet overhaul justified by one measurement. It is a second measurement, taken on an ordinary week when nothing else is going on, and a conversation with someone who can put it next to the rest of your risk factors, where it belongs.