The technology, which led to a shift in strategy more than half of the time, could save $719 per symptomatic patient over 10 years.
MONTREAL, Canada—For symptomatic patients who’ve undergone coronary CT angiography (CCTA), added knowledge gleaned from artificial intelligence-enabled coronary plaque analysis (AI-CPA) changes the management strategy more than half the time, according to new data from the DECIDE registry.
Patients with diabetes, hypertension, hyperlipidemia, obstructive CAD, and CT-derived fractional flow reserve values below 0.8 were more apt to have their treatment adjusted—mostly escalated—following receipt of the AI-CPA information at 90 days.
“The DECIDE registry therefore identifies the potential for AI plaque analysis to guide the personalization of preventative therapies beyond just stenosis severity,” said lead investigator Cian P. McCarthy, MBBCh (Massachusetts General Hospital, Boston), who presented the findings here at the 2025 Society of Cardiovascular Computed Tomography (SCCT) meeting. “These results provide supportive analysis for future trials to see if AI plaque analysis can improve cardiovascular risk [assessment] and outcomes for our patients with coronary atherosclerosis.”
Similarly, co-investigator Sarah Rinehart, MD (Charleston Area Medical Center, WV), told TCTMD that new strategies are needed to combat the fact that “physicians are massively undertreating cholesterol.”
While preventive therapies are known to improve outcomes, she continued, “we know that the traditional risk factors—those equations we use to try to tell us what dose of statin—are not effective ways to really change our physician behavior or sometimes the patient’s willingness to take the medicines. So, if you go by a personalized methodology with disease and you quantitate it, we’re hoping that will change behavior.”
The DECIDE registry is an ongoing study that plans to enroll 20,000 US patients across 40 sites to compare changes in medical management among those who receive AI-CPA analysis on top of CCTA compared with CCTA alone. The researchers have developed a novel plaque staging classification system that includes mild, moderate, severe, or extensive.
Changes Across the Plaque Spectrum
For the study, researchers included 972 symptomatic patients (mean age 63.1 years; 50.2% female) with plaque detected on CCTA. More than two-thirds (68.1%) presented with typical angina, 20.1% had diabetes, 55.9% had hypertension, and 56.8% had hyperlipidemia.
All patients underwent AI-coronary plaque analysis (Heartflow) at the same time as their index CCTA, but the results were blinded until 90 days. When the results were revealed, 51.3% of patients experienced a change in clinical management (primary endpoint), including treatment escalation in 36%, new labs ordered in 13.9%, invasive coronary angiography, or stress tests in 0.7%, treatment de-escalation in 0.6%, and referral to a specialist in 0.5%. The median time to change in management was 20 days.
Changes in management grew more common as plaque burden and stage increased (P < 0.0001 for both).
What’s “impressive” here is that physicians already know patients with severe plaque are high risk, Rinehart said, “yet it’s the artificial intelligence coronary plaque analysis that actually drove the clinical change.”
Among the 204 patients with itemized medication data available who had their treatment changed, most of these changes involved either an addition of a new lipid-lowering agent (44.1%) or an intensification in the dose of the agent they were already prescribed (23.5%).
‘Tip of the Iceberg’
With several AI-based plaque analysis tools available or in development, their “clinical value independent of stenosis, I think, deserves further study,” he said. Once reliability is proven and the technology shows “long-term prognostic value, generates relevant, actionable steps, and is cost effective, this probably may change the way we treat patients and prevent coronary adverse events.”
These data “demonstrate that the caregivers involved in the study have already accepted this concept,” Rubinshtein added.
“This is just the tip of the iceberg,” Rinehart said. “For people who are skeptical who say, ‘I need more outcomes,’ I think the thing they have to realize is this is the merging of two worlds. It’s merging the lipid world with the CT world and . . . we’ll have outcomes eventually for this. But they have to remember, this is just increasing the utility of ordering the right meds for the patients with proven outcomes.”
Additionally, she highlighted, data from the study should help patient discussions. “We’re giving you how much plaque they have, pictures of their arteries, the LDL target, and it simplifies in one or two sentences showing the patient the pictures, what they have as their disease, what their goals are, and why they need the medicines. And we know that improves adherence and understanding long-term,” said Rinehart. “So, I think it benefits clinicians in action, but patients in compliance.”
Cost Savings, Too
Not only might this technology improve adherence, but it also could potentially save money. In a poster presentation at the meeting, Suzanne Baron, MD (Massachusetts General Hospital and BAIM Institute for Clinical Research, Boston), showed the possible cost-savings associated with the AI-CPA.
Her team conducted an analysis of Medicare fee-for-service costs using 3.5 years of data from 2,827 patients enrolled in the FISH&CHIPS study who had stable chest pain and underwent CCTA. Their study was based on a retrospective look at AI-CPA that categorized patients by plaque stages in order to guide lipid-lowering therapy.
The researchers assumed a $950 cost of plaque analysis and annual cost of medical management ranging from $15 to $4,779 depending on plaque stage, as well as costs for ambulance transport ($7,500), PCI ($22,705), and inpatient medical management ($18,456).
Their model found that adding AI-CPA to CCTA lowered all-cause mortality by 0.3%, 0.5%, and 1.1% over 3.5, 5, and 10 years, respectively, with numbers needed to treat to prevent one death of 112, 81, and 45. The savings offered by AI-CPA at each time point were calculated to be $263, $373, and $719.
“So even though we’re dealing with potentially higher costs from the medication standpoint, over the long-term standpoint, not only are we modeling to improve clinical outcome, but also it will lead to cost savings over time,” Baron told TCTMD.
Her team plans to expand this research to look at other outcomes like stroke and cardiovascular hospitalization.
While Baron has not yet had access to AI-CPA in her clinical practice, she is eager to adopt it. “This is something that would be very impactful from a preventative standpoint,” said Baron.
Importantly, the technology produces a cohesive picture of what the patients’ coronary arteries look like, she added, “which I think is actually not only good for interventionalists, because that’s what we’re used to seeing, but also for patients.”