A coronary calcium scan adds little to the new risk calculator except near the treatment line
Table of Contents
1.0 Summary
The 2026 ACC/AHA multisociety dyslipidemia guideline swapped the old American risk calculator for a newer one called PREVENT, and told clinicians to consider a coronary artery calcium scan, a CT scan that measures calcified plaque in the heart arteries, when the decision about starting cholesterol medication is still unresolved. A JAMA analysis tested whether the scan actually improves on PREVENT. It used 6,098 adults aged 45 to 79 with no cardiovascular disease at the start, enrolled at six US sites in the Multi-Ethnic Study of Atherosclerosis, and followed them for 10 years, during which 366 of them (6%) had a heart attack or stroke. Adding the calcium score moved the model’s ability to rank who would have an event from 0.73 to 0.75, a gain of 0.02. The scan made a difference in one place: among people PREVENT put at borderline risk, meaning a 3% to under 5% chance of an event in 10 years, the observed 10-year event rate ran from 1.9% when the calcium score was zero to 14.3% when it was 300 or higher. The scans in this cohort were done on a research protocol between 2000 and 2002, MESA data helped build the PREVENT equations in the first place, and the gain in ranking was minimal in women and not statistically significant in Chinese participants.
2.0 Key Points
- In 6,098 MESA participants aged 45 to 79 without cardiovascular disease at baseline, adding the coronary artery calcium score to the PREVENT-ASCVD equations raised the C statistic from 0.73 (95% CI 0.70 to 0.75) to 0.75 (95% CI 0.73 to 0.77), a change of 0.02 (95% CI 0.01 to 0.03), with a categorical net reclassification improvement of 0.095 (95% CI 0.053 to 0.137) [1].
- Among participants at borderline risk, 10-year cardiovascular events occurred in 1.9% of those with a calcium score of 0, 3.9% with a score of 1 to 99, 7.4% with 100 to 299, and 14.3% with 300 or higher [1, 2].
- In that borderline group the scan correctly moved 50% of the 40 people who went on to have an event into a higher category and correctly moved 61% of the 1,075 who did not into a lower one, while moving 33% of those who had an event into a lower category incorrectly [1].
- Across the whole cohort the trade was less favorable: of the 366 participants who had an event, 18 (4.9%) were moved up correctly and 39 (10.7%) were moved down incorrectly, while of the 5,732 without an event, 1,097 (19.1%) were moved down correctly and 224 (3.9%) up incorrectly [1].
- The proportion of participants with a calcium score of 0 fell steadily across risk categories, from 77.6% at low risk to 56.2% at borderline, 41.4% at intermediate and 22.4% at high risk, and the improvement in ranking was minimal in women (net reclassification 0.075, versus 0.112 in men) and not statistically significant in Chinese participants [1].
3.0 Evidence base
American risk estimates ran for years on the pooled cohort equations, which overestimate risk by roughly a factor of two. The 2026 ACC/AHA multisociety dyslipidemia guideline replaced them with the PREVENT equations and set four categories by the estimated chance of a heart attack or stroke over 10 years: low is under 3%, borderline is 3% to under 5%, intermediate is 5% to under 10%, and high is 10% or more. The guideline gives a class 1 recommendation, its strongest, to start at least a moderate intensity statin at intermediate risk. It says a moderate intensity statin may be reasonable at borderline risk, and it gives a second class 1 recommendation to consider a calcium scan for selected adults at borderline or intermediate risk when the decision about lipid-lowering therapy is still uncertain [1].
Those recommendations carried over evidence collected under the old calculator. Studies that added calcium scoring to the pooled cohort equations reported much larger gains, about 0.09 in the C statistic and a net reclassification of 0.19 for coronary heart disease. A calculator that overstates risk leaves more room for a test to correct it, so the older numbers may not survive the switch [1].
Huang and colleagues tested the question directly in the Multi-Ethnic Study of Atherosclerosis, a longitudinal cohort at six US sites. They included 6,098 adults aged 45 to 79 with no cardiovascular disease when they enrolled. Mean age was 61.4 years (SD 9.6), 52% were women, and mean estimated 10-year risk by the PREVENT base equations was 6.4% (SD 4.8%). Calcium scans were done at baseline, between 2000 and 2002, on a research CT protocol, and 49% of participants had a score above zero. Over 10 years, 366 participants (6%) had an adjudicated fatal or nonfatal heart attack or stroke [1].
Two numbers describe what the scan added. The C statistic measures how well a model ranks the people who will have an event ahead of those who will not, where 0.5 is a coin flip and 1.0 is perfect. It went from 0.73 (95% CI 0.70 to 0.75) to 0.75 (95% CI 0.73 to 0.77) when the calcium score was added, a change of 0.02 (95% CI 0.01 to 0.03). Net reclassification counts how often a test moves future patients into the category that matches what happened to them, subtracting the moves in the wrong direction. It came to 0.095 (95% CI 0.053 to 0.137). Adding the optional kidney marker to PREVENT instead, the urine albumin to creatinine ratio, produced the same 0.02 change. Calibration, which compares predicted risk against observed risk and is exact at 1.0, was 1.09 (95% CI 0.93 to 1.25) for PREVENT alone and 0.92 (95% CI 0.81 to 1.03) with the calcium score added [1].
The reclassification arithmetic across the whole cohort deserves a closer look, because the scan moved people in both directions. Among the 366 participants who had an event, adding the calcium score moved 18 (4.9%) up into a higher category and 39 (10.7%) down into a lower one. Among the 5,732 who had no event, it moved 1,097 (19.1%) down correctly and 224 (3.9%) up incorrectly. Applied to everyone, the test spared a large number of people a statin conversation and sent twice as many future patients down as it sent up [1].
Sorted by PREVENT category, the pattern is clearer. At low risk (n = 1,885), 10-year event rates were 1.0% with a calcium score of 0 and 2.4% above 0, against 1.3% overall. At borderline risk (n = 1,115) the split was 1.9% and 5.7% against 3.6% overall. At intermediate risk (n = 1,788) it was 4.9% and 8.6% against 7.0%. At high risk (n = 1,310) it was 6.8% and 15.2% against 13.4%. The share of people with no coronary calcium at all fell across those same categories, from 77.6% at low risk to 56.2% at borderline, 41.4% at intermediate and 22.4% at high [1].
Borderline risk is where the scan changed the answer. The 10-year event rate there was under 3% without any coronary calcium and above 5% with any, which straddles the threshold at which treatment is recommended. Among the 40 borderline participants who went on to have an event, 20 (50%) were correctly reclassified to a higher category, 13 (33%) were incorrectly reclassified lower, and 7 (18%) stayed put. Among the 1,075 borderline participants who had no event, 654 (61%) moved correctly to a lower category, 277 (26%) moved incorrectly higher, and 144 (13%) stayed. Reclassification in this subgroup was 0.085 (95% CI 0.052 to 0.117) [1, 2].
The improvement was not even across the cohort. Net reclassification was 0.075 in women and 0.112 in men, and the change in the C statistic was not statistically significant in Chinese participants [1].
4.0 Where it fits in practice
The study’s own conclusion is narrow: calculate risk with PREVENT first, then consider a calcium scan for the people at borderline or possibly intermediate risk, because those are the patients whose category the scan can actually move. Nilay Shah of Northwestern University, one of the investigators, put it plainly in an interview: ordering a scan in every adult does not improve on the PREVENT calculator [2].
Two practical cautions came from commentators on the paper. Melis Sahinoz, a cardiovascular fellow at Vanderbilt, noted that a calcium score reflects calcified plaque that exists today, while PREVENT folds in longer-term and multisystem risk, so she would be careful about withholding therapy on the basis of a score of zero in a patient whose PREVENT risk is already 5% or higher. She also pointed out that a score of zero at borderline risk is not permanent, that lifestyle measures still apply if treatment is deferred, and that a repeat scan over the next three to five years is worth considering [2]. Stacey Rosen of the Katz Institute for Women’s Health, immediate past president of the American Heart Association, framed the findings as support for guideline-directed, individualized assessment across the whole risk factor spectrum rather than a verdict on the scan itself [2].
This clinic already applies a stricter rule than the guideline. Deferring a statin on the strength of a calcium score of zero requires that apolipoprotein B, a direct count of atherogenic particles, also sit below target, and the rule does not apply to established cardiovascular disease, LDL-C of 190 mg/dL or higher, familial hypercholesterolemia, or diabetes with additional risk factors. That logic is set out in /clinical/04-risk-stratification.html, and the scan’s place among the other advanced tests is in /clinical/06-advanced-tools.html. How the risk estimate itself is produced is described in /tools/prevent-calculator.html. Nothing in this analysis argues for loosening that combination requirement. If anything, the whole-cohort reclassification numbers, where the scan moved 39 future patients down and only 18 up, argue for keeping a second check on any decision to defer.
5.0 Open questions
The most consequential limitation is structural. MESA data were used to derive the PREVENT equations, so PREVENT should perform unusually well in this cohort, and the authors treat their estimate of the calcium score’s added value as conservative [1]. A different cohort might show the scan adding more.
The scans themselves are old. They were acquired on a research protocol between 2000 and 2002, and the analysis did not test whether clinically obtained scans, read in ordinary practice, perform the same way [1]. Nobody younger than 45 or older than 79 was included. The authors also did not calculate C statistics within each risk category, which is exactly the comparison a clinician deciding whether to order a scan for one borderline patient would want [1].
The signal in women and in Chinese participants is unresolved. The change in the C statistic was minimal in women and not statistically significant in Chinese participants, and the study does not explain why [1].
One question the study does not address at all is the 2026 guideline’s separate class 1 recommendation to act on coronary calcium noticed incidentally on a CT done for another reason [1]. Those scans are not risk-stratification studies, and their performance alongside PREVENT has not been measured here.
The analysis was funded by the National Heart, Lung, and Blood Institute and the American Heart Association [1].
This summary is provided for clinical decision support only and does not replace individualized clinical judgment. Findings are summarized from the cited primary sources; consult those sources before changing practice.
Version History
| Version | Date | Description |
|---|---|---|
| 1.0.0 | 2026-09-20 | Initial release |
References
- Huang X et al. Predictive Utility of Coronary Artery Calcium Score Added to the PREVENT Atherosclerotic Cardiovascular Disease Equations. JAMA. 2026-08-26. doi:10.1001/jama.2026.13233. Available at: https://doi.org/10.1001/jama.2026.13233
- CAC Found to Add Little to PREVENT-ASCVD Risk Prediction. 2026-09-09. https://www.medscape.com/viewarticle/cac-score-found-add-little-prevent-ascvd-risk-prediction-2026a1000xh4?src=rss.