CONFIRM2 Registry Highlights AI-Quantified Atherosclerosis as Key Predictor of Cardiovascular Events

The CONFIRM2 study, published in 2026, provides new evidence on how whole-heart atherosclerosis volume measured through artificial intelligence-enhanced coronary computed tomography angiography (CCTA) correlates with major adverse cardiovascular events (MACE) across varying levels of clinical likelihood for obstructive coronary artery disease (CAD). Led by Alexander van Rosendael along with co-authors Rine Nakanishi, Jeroen J. Bax, Gianluca Pontone, Saima Mushtaq, Ronny R. Buechel, Christoph Gräni, Gudrun Feuchtner, Pietro G. Lacaita, Amit R. Patel, Cristiane C. Singulane, Andrew D. Choi, Mouaz Al-Mallah, Daniele Andreini, Ronald P. Karlsberg, Geoffrey Cho, Carlos E. Rochitte, Mirvat Alasnag, Ashraf Hamdan, Filippo Cademartiri, and Ibrahim Danad, the research draws from a large international cohort to refine risk stratification beyond traditional clinical models.

Available at https://www.sciencedirect.com/science/article/pii/S2666667726002874, the paper underscores the value of quantitative plaque assessment in patients undergoing clinically indicated CCTA. CONFIRM2 builds on the original CONFIRM registry by incorporating AI-driven quantitative coronary CT (AI-QCT) to measure total plaque volume, noncalcified plaque, calcified plaque, and percentage atheroma volume across the entire coronary tree.

Background on Coronary Artery Disease and Imaging Advances

Coronary artery disease remains a leading cause of morbidity and mortality worldwide. Traditional risk assessment relies on clinical factors such as age, sex, chest pain characteristics, and risk factor-weighted clinical likelihood (RF-CL) models endorsed by the European Society of Cardiology. These models categorize patients into very-low, low, moderate, or higher risk for obstructive CAD, guiding decisions on whether to pursue noninvasive testing like CCTA.

CCTA has evolved from a tool primarily for detecting luminal stenosis to one capable of characterizing atherosclerotic plaque composition and burden. Noncalcified plaque, in particular, has been linked to higher vulnerability for rupture and acute events compared with stable calcified deposits. The integration of AI allows precise volumetric quantification that was previously labor-intensive and variable across readers.

CONFIRM2 addresses gaps in prior studies by evaluating these quantitative measures stratified by clinical likelihood categories, offering insights into whether plaque burden adds prognostic information uniformly across risk spectra or primarily in certain subgroups.

Study Design and Patient Cohort in CONFIRM2

CONFIRM2 is a multicenter, international observational registry enrolling patients referred for clinically indicated CCTA. The analysis focused on symptomatic individuals without prior CAD history in many cases, with follow-up extending a median of approximately 4.3 years. One reported cohort included 3,551 patients with a mean age of 58.8 years, roughly half male.

Patients were classified according to the ESC RF-CL model into very-low risk (36%), low risk (43%), and moderate risk (22%). This distribution reflects real-world referral patterns where many individuals present with atypical symptoms or intermediate pretest probability.

AI-QCT software processed CCTA images to derive whole-heart metrics including total plaque volume (TPV), noncalcified plaque (NCP) volume, calcified plaque volume, percentage atheroma volume (PAV), and diameter stenosis severity. The primary endpoint was MACE, encompassing cardiac death, myocardial infarction, late revascularization, and related events. A secondary endpoint combined all-cause death and myocardial infarction.

Key Findings on Atherosclerosis Volume and MACE Risk

During follow-up, MACE occurred in approximately 4.7% of the cohort. Multivariable analysis identified two independent predictors: percent diameter stenosis (hazard ratio 1.25 per 10% increase) and noncalcified plaque volume (hazard ratio 1.07 per 50 mm³ increase). Calcified plaque volume did not retain independent predictive value after adjustment.

These CT-derived measures significantly improved risk discrimination. Adding AI-QCT parameters raised the area under the curve from 0.63 (based on the RF-CL model alone) to 0.76. Similar gains occurred when compared against traditional cardiovascular risk factors or the atherosclerotic cardiovascular disease (ASCVD) risk score.

Importantly, the prognostic value held across clinical likelihood strata, suggesting that quantitative atherosclerosis assessment provides incremental information even in patients deemed very-low or low risk by clinical models alone. Higher whole-heart plaque volumes correlated with elevated event rates regardless of pretest probability category.

Photo by Lucas Santos on Unsplash

Implications for Risk Stratification in Clinical Practice

The results challenge reliance solely on clinical likelihood or stenosis severity. Noncalcified plaque burden emerged as a stronger driver of events than calcified components, aligning with biological understanding that lipid-rich, noncalcified lesions are more prone to instability.

Clinicians may consider integrating AI-QCT outputs into decision-making for preventive therapies, such as intensified lipid-lowering or antiplatelet regimens, particularly when NCP volumes are elevated. This approach could help identify higher-risk individuals within lower clinical likelihood groups who might otherwise receive less aggressive management.

The study also noted that calcified plaque, while contributing to overall burden, did not independently predict MACE once NCP and stenosis were accounted for, refining understanding of plaque phenotypes.

Role of Artificial Intelligence in Advancing Cardiovascular Research

AI-QCT represents a shift toward reproducible, scalable plaque quantification. Manual segmentation of coronary plaques is time-consuming and subject to interobserver variability; AI mitigates these limitations while enabling analysis of the entire coronary tree rather than focal segments.

CONFIRM2 demonstrates how such technology can be deployed in large, real-world registries spanning multiple countries and sites. This scalability supports broader adoption in both clinical trials and routine care, potentially standardizing risk assessment globally.

Researchers in medical imaging, biomedical engineering, and cardiology informatics may find opportunities to build on these methods, developing next-generation algorithms that incorporate additional features such as plaque composition texture or longitudinal changes on serial scans.

Opportunities for Academics and Early-Career Researchers

The findings open avenues for interdisciplinary work at the intersection of cardiology, radiology, and data science. PhD programs and postdoctoral fellowships focused on cardiovascular imaging or AI applications in medicine can leverage datasets like CONFIRM2 for secondary analyses or methodologic innovations.

Institutions with strong CCTA programs or access to large imaging repositories are well positioned to contribute to similar registries. Collaboration across borders, as seen in CONFIRM2’s 18 sites across 13 countries, highlights the value of international networks for powering robust prognostic studies.

Funding bodies and academic centers increasingly prioritize translational research that bridges imaging biomarkers with clinical outcomes, creating demand for investigators skilled in both quantitative imaging and biostatistics.

Future Directions and Broader Impact

Ongoing expansions of CONFIRM2 aim to reach up to 30,000 patients, enabling more granular subgroup analyses and validation across diverse populations. Integration with electronic health records and wearable data could further personalize predictions.

Longer-term goals include prospective trials testing whether AI-QCT-guided therapy reduces MACE rates compared with standard care. Sex-specific differences observed in related analyses—where plaque features conferred higher relative risk in women despite lower absolute burden—warrant dedicated investigation within the clinical likelihood framework.

Ultimately, these advances support a precision-medicine paradigm in cardiology, where quantitative imaging complements clinical judgment to optimize resource allocation and patient outcomes.

Photo by GuerrillaBuzz on Unsplash

Conclusion

The CONFIRM2 study advances understanding of whole-heart atherosclerosis volume as a robust, independent predictor of MACE that enhances risk assessment across the spectrum of clinical likelihood for obstructive CAD. By demonstrating the superiority of AI-quantified noncalcified plaque and stenosis measures over traditional models, the research paves the way for more refined, imaging-informed strategies in cardiovascular prevention and management. Academics and clinicians alike can draw on these insights to guide both practice and future investigative efforts. Further details are available in the full publication at the provided link.