Key Points
- In a prospective U.S. pivotal trial, CLAiR, a non-invasive, AI-powered retinal scan, identified elevated ASCVD risk with 91.1% sensitivity and 86.2% specificity.
- The study included 940 enrolled participants across 10 eye care and primary care sites, with 874 in the primary analysis.
- Imageability reached 94.1% with NW400 and 99.4% with NW500, supporting point-of-care feasibility.
Cardiovascular prevention often breaks down before treatment ever begins. Many patients at elevated risk never make it to formal risk assessment, leaving a large gap between those who qualify for preventive therapy and those who receive it. At ACC.26, investigators presented a different entry point: the retinal image already captured during a routine eye exam.
At the 2026 American College of Cardiology Scientific Sessions, Michael V. McConnell, MD, MSEE, presented “Prospective Multi-center Clinical Trial of Artificial-intelligence Analysis of Retinal Images for Identifying Elevated Atherosclerotic Cardiovascular Risk,” (NCT06808334) a pivotal, prospective, multi-center U.S. study. The trial evaluated CLAiR, an investigational AI system designed to classify whether a patient’s 10-year ASCVD risk is ≥7.5%.
Investigators enrolled 940 participants age 40-75 without prior ASCVD across 10 U.S. eye care and primary care sites. Patients taking lipid-lowering therapy, those with advanced eye disease, and those who were pregnant were excluded. After a standard retinal photograph was obtained, CLAiR analyzed the image immediately at the point of care. During the same visit, clinicians collected blood pressure, lipid values, and other clinical data needed to calculate the reference 10-year ASCVD risk score from the Pooled Cohort Equation. The pre-specified primary endpoints were sensitivity and specificity against that standard.
The primary analysis included 874 participants, or 93% of those enrolled, with both analyzable retinal image pairs and complete medical data. Mean age was 57.5±9.7 years; 50.7% were female, 19.3% were Black, and 25.9% were Hispanic. Overall, 26.3% had an ASCVD risk score ≥7.5%. CLAiR achieved a sensitivity of 91.1% (407/447; 95% CI 87.4%-94.4%) and a specificity of 86.2% (1077/1250; 95% CI 83.5%-88.6%), exceeding the pre-specified thresholds of 70% and 80%, respectively. Positive predictive value was 70.2% (407/580), negative predictive value was 96.4% (1077/1117), and AUC was 0.956.
The signal held up across settings. Investigators reported similar performance in eye care and primary care clinics and across both camera models, with imageability of 94.1% for NW400 and 99.4% for NW500. Dr. McConnell captured the practical aim of the study directly: “This approach would not replace the standard cardiovascular risk evaluation, but it’s a potential way to bring greater awareness, especially for people who should be on preventive care, but who have not yet had a thorough evaluation.”
There were clear limits. CLAiR remains investigational in the U.S., is not designed for people who are pregnant or who have advanced eye disease, and the presenters noted that patients will still need reliable pathways from an elevated screening result to clinical follow-up. Even so, the study left ACC.26 with a memorable possibility: one of the next front doors to cardiovascular prevention may open not in the cardiology clinic, but in the quiet retinal image captured during a routine eye exam.