Two documents about the same 84 people went up three days apart. One of them contains the only routine measurement that tracked coronary inflammation in that cohort. It is not the one with the headline.
On 22 October, Biograph announced the completion of a US study using Caristo Diagnostics’ CaRi-Heart technology, reporting that coronary inflammation was “common (42.9%) in asymptomatic, low-risk adults”. The underlying preprint, posted to medRxiv on 20 October, is linked from the release and is 21 pages long. Read against each other, the release’s numbers hold up better than the house style of this site would predict. What separates them is a single word, and a variable the release never names.
The two documents
The study is cross-sectional: 84 consecutive asymptomatic adults at two Biograph sites, mean age 53.7, 63.1% men, each scanned with coronary CT angiography and profiled on a wide panel of blood markers. Perivascular fat attenuation was scored by Caristo’s CaRi-Heart platform into an age- and sex-standardised percentile per vessel, and “elevated” was pre-specified as any one of three vessels at or above the 75th percentile. Caristo’s product received FDA De Novo authorisation under DEN250042 in July 2026, so this is a cleared device being characterised in a new population rather than a prototype.
The authors are all Biograph employees and Biograph sponsored the study, which the preprint states in its competing-interest declaration. They also disclose using a large language model for “literature synthesis, initial text drafting, and language refinement” across four sections, while stating that the design, statistics and data were their own.
Four claims, scored
Taking every checkable factual claim in the release in turn, against the preprint:
“42.9%” — holds. The preprint reports elevated FAI-Score in 36 of 84 participants, “42.9% of participants (36/84; 95% CI 32.8% - 53.5%)”. The release gives the point estimate without the interval, which on 84 people runs from roughly a third to over half.
“17.9% … very high inflammation” — holds. 15 of 84, at or above the 90th percentile.
“About 47% of participants with no detectable coronary plaque exhibited elevated coronary perivascular inflammation” — holds. The preprint has 47.2%, 25 of 53.
FDA authorisation — holds, and is verifiable outside either document.
Four for four on the arithmetic. Against the pattern this site usually finds in vendor material, that is worth saying plainly: nobody rounded anything in their own favour.
The one word that does not hold
The release says routine bloodwork and clinical tools “showed no correlation with the presence of coronary inflammation and failed to identify which patients had inflamed arteries”.
For the binary outcome, that is what the paper found. But hs-CRP — one of the four markers the release names — did correlate with inflammation when the score was treated as a continuous variable rather than a yes/no: “hs-CRP was also inversely correlated with FAI-Score (Spearman ρ=−0.27, p=0.014)”. Inversely. Higher systemic inflammation went with lower coronary inflammation score, significantly, and the effect survived into a multivariable model at p < 0.001.
The authors do not oversell this. They report that removing two participants with extreme hs-CRP values “most likely due to illness” weakened it to ρ=−0.22, p=0.053, and they write that the association “requires replication and should not be interpreted as protective”. That caveat is the correct one and it is in the paper. What the release does is flatten a significant inverse correlation into “no correlation”, which is a different statement about the same data.
What the enumeration could not reach
The release’s sentence enumerates what it checked: cholesterol markers, blood sugar, hs-CRP, a risk calculator. All four are blood draws or a calculator fed by them. And across that list, the claim is fair.
The paper, though, ran “96 secondary association tests”, and exactly one survived correction across all of them. It was body mass index. Median BMI was 27.1 kg/m² among participants with elevated FAI-Score versus 24.4 among those below the threshold, “AUC = 0.71, Mann-Whitney p=0.001, all-test FDR q=0.028” — described in the paper as “the only secondary association to meet the all-test false discovery rate threshold”.
The word “BMI” does not appear in the press release. Neither does “body mass index”. The release’s claim about blood tests is true, and the thing that worked was not a blood test, so the enumeration never had to mention it. Both documents are honest; only one of them tells you that the cheapest measurement in the clinic was the one that tracked the signal, and at AUC 0.71 it tracked it only modestly — this is weak discrimination, not a screening test.
Two further things the release leaves out cut in Biograph’s own favour, which is why they are worth naming. Its full regression models, using ApoB, hs-CRP, HbA1c, age, sex and BMI together, reached a cross-validated AUC of 0.71 for elevated inflammation and 0.52 — chance — for the very high category. And the paper’s sharpest single sentence is absent from the release entirely: “all 15 individuals with very high FAI-Score had CAD-RADS 0”, meaning every participant in the most inflamed group had no detectable luminal narrowing at all, at Fisher’s exact p=0.018. That is a stronger version of the story the release was trying to tell, and the release does not carry it.
What neither document establishes
The preprint’s limitations section is blunter than the release’s single “more research is needed” line. The cohort is “a convenience sample of consecutive, self-referred asymptomatic adults” at one organisation’s two sites. The design “cannot establish that FAI-Score elevation precedes plaque formation”. And: “These data therefore do not yet support population-wide CCTA screening or treatment based solely on FAI-Score elevation.”
Also unresolved, and not a criticism of either document: the percentiles are relative to Caristo’s reference distribution, so “42.9% above the 75th percentile” is a statement about this cohort against that reference, not a count of diseased arteries. I did not obtain the reference-population construction — the methods paper behind it sits behind a publisher paywall that blocks automated access — so how much of the 42.9% reflects cohort composition rather than disease is not something I can check here.
Which document to read
Nothing here is a caught exaggeration. The release’s four numbers are the paper’s four numbers. But a reader who stops at the release learns that blood tests failed and does not learn that a tape measure did better than any of them, that the model for severe inflammation performed at chance, or that every one of the 15 most inflamed participants had clean arteries. Three days separate the two documents and 21 pages separate their contents. The release is linked to the preprint directly, which is more than many do. It is still the shorter document that makes the claim and the longer one that tells you where the walls are.