By Michael H. Crawford, MD, Editor
SYNOPSIS: A deep learning-derived algorithm for measuring coronary artery calcium scores in non-ECG-gated, non-contrast chest CT scans ordered for non-cardiac reasons was predictive of death and adverse atherosclerotic cardiovascular events. This may provide an opportunity for earlier prevention interventions.
SOURCE: Peng AW, Dudum R, Jain SS, et al. Association of coronary artery calcium detected by routine ungated CT imaging with cardiovascular outcomes. J Am Coll Cardiol 2023;82:1192-1202.
Despite its value as a predictor of atherosclerotic cardiovascular disease (ASCVD) events and mortality, ECG-gated CT scans for determining coronary artery calcium scores is underused for a variety of reasons. However, millions of non-gated chest CT scans are performed for non-cardiac reasons every year. Prior studies have shown the prognostic value of reporting observed coronary calcium on these CT scans.1 Peng et al went much further by employing an artificial intelligence deep learning algorithm to quantify incidental coronary artery calcium (CAC) on non-gated chest CTs.
The authors validated the algorithm using patients who had undergone both gated and non-gated CT scans, which resulted in a more accurate Agatston score from the non-gated studies. The aim of their study was to assess whether adding this CAC score from non-gated CT scans to traditional CV risk factors would improve the prediction of mortality and adverse CV events. Among 8,040 adult patients who had undergone non-contrast, non-ECG-gated chest CT scans between 2014 and 2019, 5,678 met inclusion criteria. Researchers excluded those with missing information, existing ASCVD, metastatic cancer, and death on the date of the CT scan. Investigators gleaned clinical information for calculating the 10-year ASCVD risk and follow-up information from the medical record. The primary outcome was death. The two secondary outcomes were the composite of death, myocardial infarction (MI) and stroke, and the composite endpoint plus any arterial revascularization. The study population was 51% women, 18% Asian, and 13% Hispanic (mean age = 61 years).
About half the patients recorded a CAC score of higher than 0 (52%), 33% recorded a score of 100 or higher, and 15% scored between 0 and 100. Patients with a score of 100 or higher had an increased adjusted risk of death (HR, 1.51; 95% CI, 1.28-1.79), death/MI/stroke (HR, 1.57; 95% CI, 1.33-1.84), and death /MI/stroke/revascularization (HR, 1.68; 95% CI, 1.45-1.98) compared to those with a CAC score of 0. Those with a score of 100 or higher recorded an ASCVD clinical 10-year risk score of 24%, yet only 26% were on statins. The authors concluded an incidental CAC score of 100 or higher was associated with an increased risk of death and adverse ASCVD events beyond that predicted by traditional risk factors. This information from routine chest CT scans may facilitate earlier ASCVD prevention interventions.
COMMENTARY
Since the publication of studies showing that the incidental finding of coronary calcium in non-gated chest CT scans was of prognostic value, my institution and many others have started noting this finding in their CT scan reports. However, few are reporting a quantitative analysis of the Agatston score, largely because it is time consuming. Thus, to have an artificial intelligence-derived algorithm at our disposal to quantitate the CAC is an important advance.
CV event rates increase as the CAC score goes up, so the therapeutic response can be customized to the patient. For example, those with a score of 100 or higher are twice as likely to experience an ASCVD event vs. those with a score of 0, even when adjusted for traditional risk factors. These patients could benefit from an appropriate intervention; however, in the Peng et al study, only one-quarter of patients were taking a statin.
One strength of this paper was the fact the authors studied a diverse group of patients, with CT scans ordered for many reasons. Prior non-quantitative studies focused on specific patient populations, such as those with very high LDL cholesterol levels.2 Another strength of the Peng study was that the algorithm was tested and refined using ECG-gated studies.
There were several limitations to the Peng et al study. It was performed at one institution, but the authors countered they conducted the work in different settings using various CT scanners. The adjustments for traditional risk factors were weakened by considerable missing data for the clinical risk calculation. These chest CT scans probably do not represent the entire pool of primary prevention patients because the scans were ordered for clinical reasons. Some of the mortality observed was undoubtedly not ASCVD-related. Finally, despite the fact that only about two-thirds of the population was white, there were few Black patients enrolled (less than 4%). Despite these weaknesses, I believe the Peng et al study is compelling and represents a positive use for artificial intelligence.
REFERENCES
1. Detrano R, Guerci AD, Carr JJ, et al. Coronary calcium as a predictor of coronary events in four racial or ethnic groups. N Engl J Med 2008;358:1336-1345.
2. Mortensen MB, Caínzos-Achirica M, Steffensen FH, et al. Association of coronary plaque with low-density lipoprotein cholesterol levels and rates of cardiovascular disease events among symptomatic adults. JAMA Netw Open 2022;5:e2148139.
A deep learning-derived algorithm for measuring coronary artery calcium scores in non-ECG-gated, non-contrast chest CT scans ordered for non-cardiac reasons was predictive of death and adverse atherosclerotic cardiovascular events. This may provide an opportunity for earlier prevention interventions.
You have reached your article limit for the month. Subscribe now to access this article plus other member-only content.
- Award-winning Medical Content
- Latest Advances & Development in Medicine
- Unbiased Content