On 29 July 2026 Caristo Diagnostics announced FDA authorisation for CaRi-Heart, the first device authorised in the United States to quantify coronary inflammation from a coronary CT angiogram. The software analyses the attenuation of the fat tissue surrounding the artery, derives an index called the FAI-Score and returns it alongside the report together with a ten-year cardiovascular mortality estimate. The regulatory route was De Novo, the one reserved for devices without an equivalent predicate on the market, which is the sign that the FDA treated the measurement as a new category.

It is the most recent episode in a shift that concerns the target of preventive cardiology itself: from stenosis, that is from how narrowed a vessel is, to the vessel wall and to what accumulates in it.

The change of target

For forty years the clinical question has been whether an artery was obstructed enough to limit flow, and the tools — coronary angiography, stress testing, ischaemia measurements — were built to answer it. The known problem is that a substantial share of heart attacks arises from plaques that do not obstruct: they sit in the wall, have a thin fibrous cap and a lipid core, and rupture without ever having narrowed the lumen significantly.

In April 2025 this reading reached The Lancet with a dedicated commission, The Lancet Commission on rethinking coronary artery disease: moving from ischaemia to atheroma (Lancet 2025;405(10486):1264-1312, online from 31 March). A figure circulates from that document which is worth quoting with its exact context, because elsewhere it gets attributed to imaging: the 8.7 million lives a year are the estimate of what would happen by eliminating or controlling behavioural and metabolic risk factors early in life by 2050 — smoking, hypertension, cholesterol, diet — with an 82.1% reduction in deaths. It is a counterfactual model, a theoretical ceiling, and not a benefit attributed to early detection by imaging.

On the inflammation side, on 29 September 2025 the American College of Cardiology published a dedicated scientific statement whose conclusion is unambiguous: “In aggregate, the evidence linking inflammation with atherosclerotic CVD is no longer exploratory but is compelling and clinically actionable”. The same document warns that not all trials of anti-inflammatory therapy in secondary prevention have succeeded, and that evidence is needed before broad recommendations on other agents.

What these tools measure

Different tools work on the same CT acquisition, and it helps to keep apart the things they measure.

  • Plaque volume and composition. Cleerly (latest clearance K242338, March 2025), HeartFlow Plaque Analysis (K250902, 18 July 2025) and Elucid PlaqueIQ segment the coronary tree and quantify calcified and non-calcified plaque. Cleerly’s authorised indication for use is the visualisation and measurement of plaque and stenosis: not risk prediction, and not therapy guidance.
  • Derived physiology. Fractional flow reserve computed from CT estimates the haemodynamic impact of a lesion without a catheter.
  • Perivascular inflammation. This is the measurement CaRi-Heart has just been authorised for: it concerns the fat tissue around the artery, whose attenuation changes in the presence of inflammation.

The evidence, and where it stops

Precision matters here, because the numbers circulating in communications are often stronger than those in the studies.

The ORFAN study, published in The Lancet in 2024, is cited for a cohort of 40,091 consecutive patients across eight UK hospitals. That cohort, however, serves to quantify how many events fall on patients without obstructive stenosis: the prognostic value of the FAI was assessed in a subcohort of 3,393 patients from two hospitals only, those with the longest follow-up. And the adjusted hazard ratios, per standard deviation increase of the index, sit between 1.25 and 1.45 depending on vessel and outcome, not at the double-digit values that appear in headlines.

On the plaque side, the study with the longest follow-up covers 536 patients followed for a median of 10.3 years: the highest stage of atheroma volume carries an adjusted hazard ratio of 3.57 against the two lowest. The DECIDE registry showed that receiving the quantitative analysis changes clinical management in 51.3% of 972 symptomatic patients, which is a measure of how much the report moves decisions, not of how much it improves outcomes.

The most useful summary is the ACC’s own, in a scientific statement of 17 December 2025 devoted to quantitative plaque analysis: “there is insufficient evidence that QCPA improves patient management decisions and outcomes” and “there is no consensus on when or how to use and interpret QCPA in clinical practice”. The document asks vendors for published validation against invasive imaging or histology and for declared plaque volume thresholds. No randomised trial has so far shown that managing patients on the basis of these measurements reduces heart attacks or deaths.

Reimbursement and guidelines

The gap between evidence and adoption is documented by the payment codes. Since 1 January 2026 coronary plaque quantification has had a permanent Category I CPT code in the United States, 75577, with a national average payment of around a thousand dollars per study: it is the second AI imaging algorithm to obtain a Category I code, after the one for CT-derived flow reserve which came into force a year earlier. Actual Medicare coverage, though, is older, and dates to local determinations effective from 24 November 2024.

In Europe the picture differs. The 2024 ESC guideline on chronic coronary syndromes promotes CT angiography to a first-line test in many scenarios, but contains no recommendation on AI plaque analysis. In Italy the outpatient specialist care schedule introduced by the decree of 25 November 2024 provides no reimbursable service either for AI plaque quantification or for CT-derived physiology: anyone wanting to use them would do so outside the national health service perimeter.

There is also a constraint that will mature: these tools are medical devices, and an AI system used as a safety component of a product subject to third-party conformity assessment falls among the AI Act’s high-risk systems by the route of Article 6(1) — the same mechanism I described in an earlier article with regard to machinery.

Neighbouring technologies, the same question

The pattern repeats with other tools proposed for stratifying risk before disease manifests.

Polygenic scores for coronary artery disease have a solid population-level basis: the 2018 work that made them known identified 8% of people with a threefold risk relative to the rest. The predictive increment above classic risk factors remains disputed, however: a 2020 JAMA paper found no clinically useful improvement from adding a score to the risk equations, and a 2025 study, also in JAMA, computed 48 different scores for the same disease in more than 170,000 people, finding equivalent population performance but individual estimates that disagreed with each other. To this is added that accuracy decays along the continuum of genetic ancestries, and not only between discrete categories.

It is the same profile as imaging: a real signal at population level and a serious difficulty in translating it into the decision about the individual.

What to expect

Several awaited readouts have not arrived. The randomised trial testing a preventive strategy guided by AI plaque analysis in asymptomatic patients has produced no results. On the drug side, as of 30 July 2026 the outcome trial of the lipoprotein(a) antisense has no public results despite primary completion having been expected in June, and the trial of the anti-interleukin-6 antibody, which is the most direct test of the inflammatory hypothesis, has been complete since 9 June 2026 without the data being released. The outcome trial of the other anti-Lp(a) agent will not read out before 2028.

One result already in deserves attention because it runs against the enthusiasm. In March 2026, in a serial-CT substudy of 468 participants, a drug lowering triglycerides by 60% did not change non-calcified plaque volume at twelve months. Measuring plaque and shifting it are two distinct problems.

Limits

The 29 July authorisation rests on observational and prognostic evidence, not on a trial showing a benefit from managing patients according to that index; at the time of writing the De Novo decision number has not been made public and the record is not yet queryable in FDA databases, so the source is the company’s announcement. The associated payment codes are Category III, temporary and without guaranteed reimbursement.

The risk that mirrors the enthusiasm also applies. Measuring a widespread disease more finely across a large population means finding more of it, and without a shared intervention threshold the likely consequence is more therapy in people whose absolute risk remains low. It is exactly the concern the ACC expresses in asking for declared thresholds and consensus on interpretation.

Finally, almost all the evidence cited here comes from cohorts and registries, often built or funded by those who make the software. The measurements are reproducible and the underlying biology is plausible. What is missing is the step that counts most in medicine: the demonstration that deciding on the basis of those numbers makes people live longer.

Cover image: heart and lungs, plate from J.M. Bourgery and N.H. Jacob, “Anatomie de l’homme”, GuĂ©rin, Paris 1862 — reproduction by User:Royonx, CC BY-SA 3.0 — https://commons.wikimedia.org/wiki/File:Bourgery_%26_Jacob.jpg