AI blood test CardiOmicScore reads proteins for risk early

CardiOmicScore, an AI blood test from the University of Hong Kong, predicts major cardiovascular risks from one blood sample.

The research team built the model to analyze 2,920 proteins and 168 metabolites, using circulating biological signals to assess future cardiac danger. The test is designed to estimate risk for six cardiovascular diseases, including heart attack, stroke and heart failure.

The researchers said the model can identify risk up to 15 years before symptoms appear. The findings, published in Nature Communications, showed CardiOmicScore outperforming conventional genetic risk scores, according to the university researchers.

2,920 proteins in one sample

The central shift is from inherited risk to present biological condition. Genetic risk scores are fixed at birth, while proteins and metabolites can change as the body responds to diet, aging, illness and treatment.

That distinction matters because heart disease prevention depends on timing. A fixed genetic profile can flag predisposition, but a moving blood-based profile may show whether risk is building or easing as a patient changes course.

The approach also compresses several layers of cardiovascular screening into a single blood draw. The supplied research summary does not say the test replaces imaging, specialist review or standard clinical measures, but it suggests a possible earlier triage tool.

Genes versus shifting signals

CardiOmicScore sits in a wider push to use high-volume molecular data for preventive medicine. Instead of relying only on age, family history or inherited variants, the model reads thousands of markers that reflect the body’s current state.

For patients, the potential appeal is a warning that arrives before chest pain, neurological damage or heart failure symptoms force urgent care. For clinicians, the value would depend on whether the score changes decisions: earlier medication, closer monitoring or more aggressive lifestyle intervention.

The University of Hong Kong gains a research claim in one of medicine’s most commercially active fields: AI-enabled diagnostics. The article summary, however, does not provide cohort size, population mix, false-positive rates, regulatory status or pricing, all of which are essential before clinical use can be assessed.

Fifteen-year warnings need proof

If CardiOmicScore is validated across large and diverse populations, the macro effect would come through prevention rather than emergency treatment. Fewer severe cardiac events would reduce pressure on health systems, insurers and employers, especially where aging populations are raising chronic-disease costs.

For the University of Hong Kong, broad replication would strengthen the model’s standing and could attract clinical partnerships or further translational research. For diagnostics companies and hospital laboratories, the mechanism is clear: a single-sample proteomic and metabolomic screen could create demand for new testing platforms and data interpretation services.

If performance weakens outside the original study setting, the model’s role would likely narrow. It could remain useful as an adjunct for selected patients, while the wider sector would face a reminder that AI diagnostics need external validation, not just strong discovery results.

A third path depends on whether repeated testing can track improvement after intervention. If the score falls when risk factors are controlled, it becomes a feedback tool for prevention; if not, its value is mainly long-range risk classification.

The next questions are practical and clinical. Researchers, regulators and health systems will need evidence on who was tested, how accurate the score is across demographic groups, how often it should be repeated and what doctors should do when it flags a high-risk patient years before symptoms.