Preventive Health

An AI Blood Test That Spots Heart Disease Up to 15 Years Early - What It Is (and Isn't) Yet

Hong Kong researchers built CardiOmicScore, an AI model that reads thousands of proteins and metabolites from one blood sample to flag six cardiovascular diseases up to 15 years before symptoms. What's promising, what's still research-stage, and what already works today.

· Jul 22, 2026
An AI Blood Test That Spots Heart Disease Up to 15 Years Early - What It Is (and Isn't) Yet
Illustration generated by AI
Table of contents
  1. What the test actually measures
  2. Why the "current state" part is the big deal
  3. The honest caveats
  4. What you can do now (that already works)
  5. Bottom line

Imagine a single blood test that could flag your risk of a heart attack or stroke up to 15 years before any symptoms appear. Researchers at the University of Hong Kong's medical school (HKUMed) have built exactly that as a research tool — an AI model called CardiOmicScore — and the results, published in Nature Communications, are genuinely promising. Here's what it does, and, just as important, what it doesn't do yet.

What the test actually measures

Instead of relying on your genes alone, CardiOmicScore reads the current state of your body. From a single blood sample it analyzes about 2,920 proteins and 168 metabolites, and uses deep learning to combine that "multiomics" picture into a risk estimate for six major cardiovascular diseases: coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease, and venous blood clots. In people at elevated risk, it detected warning signals as much as 15 years before clinical onset, and it outperformed conventional genetic risk scores in the study.

Why the "current state" part is the big deal

Traditional genetic risk scores fix your risk roughly at birth — your DNA doesn't change. Proteins and metabolites do. They shift with your diet, your weight, aging and illness. That means a test like this could, in principle, catch risk building — and, crucially, catch it receding as someone changes course. A 15-year warning is only useful if there's still time to act on it, and a moving signal is something you can actually influence.

The honest caveats

  • It's research, not a test at your clinic. The study was built on large-scale population data (the UK Biobank). That's a strong foundation, but it's not the same as a validated, approved test you can book tomorrow — expect further validation before anything like this reaches routine care.
  • Population data has limits. Results from one large cohort don't automatically transfer to every group; broader validation across diverse populations is the usual next step.
  • A risk score isn't a diagnosis — and it isn't destiny. It's a prompt to act, in partnership with a doctor.

What you can do now (that already works)

The exciting future of predictive testing doesn't change today's playbook — and the basics are proven:

Bottom line

CardiOmicScore is a compelling glimpse of where preventive cardiology is heading: earlier warnings, based on a body's changing state rather than a fixed genetic baseline. For now it's a research breakthrough, not a product — so treat the headline as motivation, not a substitute for the proven steps that already lower your risk today. As always, this is general information, not medical advice; talk to your doctor about your own heart health.

Sources: University of Hong Kong (HKUMed), Nature Communications, ScienceDaily. CardiOmicScore is a research model and not yet an approved clinical test.