With the help of artificial intelligence (AI), arterial inflammation measured with coronary computed tomography angiography (CCTA) can predict fatal and nonfatal events in patients with nonobstructive coronary artery disease (CAD), according to a study that suggests this approach would change treatment about half the time.
In patients with nonobstructive CAD, CCTA measurement of inflammation on the basis of the Fat Attenuation Index (FAI) “predicts fatal and nonfatal cardiac events independently from clinical risk scores and routine CCTA interpretation,” reported Charalambos Antoniades, MD, PhD, professor of cardiology, Radcliffe Department of Medicine, Oxford, England.
This analysis was based on data from ORFAN, an ongoing study that expects to eventually collect data from 250,000 CCTA. There were multiple goals. The first was to evaluate whether there is a need and a role of CCTA to risk stratify patients without obstructive CAD. A second objective was to evaluate if the FAI inflammation score can quantify residual risk in these patients.
Based on the answers to these questions, the investigators then proceeded to determine if an AI risk model that combines data from the FAI score and risk factors is widely generalizable and, in addition, whether it reclassifies patients in a way meaningful to management.
CCTA-based inflammation is promising
The answers to all these questions were yes, according to data presented by Dr. Antoniades in a late-breaker at the American Heart Association scientific sessions.
So far, ORPHAN, which has multiple participating sites in the United Kingdom, Europe, United States, South America, Asia, and Australia, have data on more than 100,000 CCTAs. Approximately 40,000 have been processed. Of these, 82% have had nonobstructive CAD and the remaining obstructive disease.
In long-term follow-up, the numbers of major adverse cardiovascular events (MACE) and cardiac deaths were compared in these two groups. In absolute terms, the nonobstructive CAD group had about twice as many MACE (2,587 vs. 1,450) and cardiac deaths (1,118 vs. 636).
The rate of these events was much lower in the nonobstructive group , which had four times more patients than the obstructive group, but Dr. Antoniades said these data demonstrate substantial rates of events in the nonobstructive group as well as an unmet need to identify and treat risk associated with nonobstructive CAD.
When determining if coronary inflammation as measured with CCTA could be a means identifying risk independent of other factors, the FAI scores were evaluated by quartile in a nested cohort of 3,666 consecutive patients. FAI, which has been validated, is calculated with spatial changes in CCTA-measured perivascular fat composition after standardization for anatomy and other variables.
The discrimination for risk with FAI was impressive. When evaluated across all patients (obstructive or nonobstructive CAD), those in the highest FAI quartile had a hazard ratio (HR) for MACE that was more than six times higher (HR 6.76; P P
“The prediction was independent of all other risk factors,” Dr. Antoniades reported.
Predictive value greater in nonobstructive CAD
When evaluated in nonobstructive disease, the predictive value of FAI was even greater. In obstructive CAD patients, the increased risk of MACE for the fourth relative to the first quartile was increased threefold (HR 3.15; P P P P
When a risk model based on AI that incorporated FAI plus other cardiovascular risk factors was applied retrospectively to the ORPHAN data, the predicted and actual event graph lines were nearly superimposable over a follow-up to 10 years at risk levels ranging from low to very high.
When this inflammation-based AI model was evaluated against standard risk prediction in patients with nonobstructive CAD, 30% of patients were reclassified to a higher risk category and 10% to a lower risk category.
When the AI-risk calculations were provided to clinicians at four hospitals over a recent 1-year period, it resulted “in changes of management in approximately half of patients,” Dr. Antoniades said.
Overall, Dr. Antoniades said these data provide evidence that coronary inflammation is an important driver of residual risk in patients who have nonobstructive CAD on CCTA, and he believes that the AI-enhanced interpretation of the FAI-based inflammatory burden has the potential to become an important management tool.
“AI-risk assessment may transform risk stratification and management of patients undergoing routine CCTA,” Dr. Antoniades said.
Imaging has potential for expanded risk assessment
The AHA-invited discussant, Viviany R. Taqueti, MD, director of the cardiac stress laboratory at Brigham and Women’s Hospital, Boston, agreed with the promise of evaluating inflammatory infiltrate in the coronary arteries as well as looking at fat in other tissues, such as skeletal muscle, to better risk stratify patients, but she cautioned about the limitations of conclusions based on observational data.
“A registry is not a randomized trial,” she said.
Characterizing AI as a “black box” in terms of understanding methodology, she also recommended further studies to validate the relative contribution of AI to inflammation alone in risk stratification.
Still, she believes that the “explosive growth” in imaging has created new opportunities for more precisely evaluating cardiovascular risk. She said these might be particularly helpful in the context of the “changing landscape” in CAD driven by less smoking, more obesity, and increased statin use. Overall, she endorsed the basic questions Dr. Antoniades is exploring.
“This is an incredibly intriguing idea that deserves continuing research,” she said.
Dr. Antoniades reported financial relationships with Amarin, AstraZeneca, Caristo Diagnostics, Covance, Mitsubishi Tanabe, MedImmune, Novo Nordisk, Sanofi, and Silence Therapeutics. Dr. Taqueti reported no potential conflicts of interest.
This article originally appeared on MDedge.com, part of the Medscape Professional Network.
Source link : https://www.medscape.com/s/viewarticle/998817?src=rss
Publish date : 2023-11-27 22:16:02
Copyright for syndicated content belongs to the linked Source.