A test built around what genes can't tell you

Standard cardiovascular risk assessments have relied on the same basic toolkit for decades: age, blood pressure, cholesterol, smoking status, and family history, sometimes supplemented in recent years by genetic risk scores. That approach has a built-in limitation researchers have long acknowledged but struggled to solve — genetic risk scores capture where a person started, biologically speaking, but they stay fixed for life, unable to reflect how someone's actual health has changed based on diet, exercise, stress, illness, or the countless other factors that shape cardiovascular risk over the course of a lifetime.

A research team at the Li Ka Shing Faculty of Medicine of the University of Hong Kong, known as HKUMed, has built a tool designed specifically to close that gap. Called CardiOmicScore, the system uses a single blood draw combined with artificial intelligence analysis to estimate a person's future risk across six major cardiovascular diseases — and, according to findings published in Nature Communications, can flag warning signs up to 15 years before any of those conditions become clinically apparent.

Why "multiomics" matters more than a single biomarker

The core innovation behind CardiOmicScore is its deliberate combination of multiple layers of biological data rather than relying on any single measurement. The approach, known as multiomics, brings together genomics, which examines a person's underlying genetic information; proteomics, which focuses on the proteins carrying out essential functions throughout the body; and metabolomics, the study of small molecules called metabolites that the body produces as it processes food, generates energy, and responds to disease or stress.

Professor Zhang Qingpeng, associate professor in HKUMed's Department of Pharmacology and Pharmacy and a lead researcher on the project, explained the underlying logic behind combining these different data layers: "Genes determine where we start — they define our baseline health risk. However, proteins and metabolites reflect our current physical health." That distinction is central to what makes this tool different from earlier genetic risk scores. Proteins and metabolites function, in effect, as real-time recorders of what's actually happening inside a person's body right now — capturing subtle shifts in immune function, metabolic activity, and vascular health that a fixed genetic score, established once and never updated, simply cannot reflect.

The scale of data behind the model

Building a model capable of this kind of prediction required an enormous underlying dataset. The HKUMed team drew on large-scale population data from the UK Biobank, one of the world's most extensively studied long-term health databases, analyzing 2,920 circulating proteins and 168 distinct metabolites measured directly from participants' blood samples. Using deep learning techniques, the researchers integrated all of that molecular information into a single predictive framework capable of generating disease-specific risk scores.

That framework produces two distinct sub-scores that feed into the overall prediction: a proteomic risk score, referred to as ProScore, and a metabolomic risk score, called MetScore. Combining both protein-based and metabolite-based signals, rather than relying on just one category of molecular data, is what allows CardiOmicScore to account for different biological pathways that can each independently signal elevated cardiovascular risk — some diseases may show clearer warning signs through protein changes, while others might reveal themselves more clearly through metabolic shifts, and a model incorporating both stands a better chance of catching whichever signal actually appears first in a given individual.

Six diseases, one blood draw

CardiOmicScore is designed to estimate risk across six specific cardiovascular conditions: coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease, and venous thromboembolism. That's a deliberately broad scope rather than a tool narrowly focused on just one condition — cardiovascular disease encompasses a range of distinct but often interconnected disorders, and a single risk score covering all six gives clinicians a considerably more complete picture from one blood sample than would be possible screening for each condition separately using conventional methods.

The stakes motivating this breadth of coverage are considerable. Cardiovascular diseases collectively remain the leading cause of death worldwide, responsible for approximately 19.8 million deaths in 2022 alone, according to figures cited in HKUMed's own announcement of the research. A tool capable of flagging elevated risk across this entire disease category up to 15 years before clinical symptoms emerge — rather than after a heart attack, stroke, or other acute event has already occurred — represents a meaningful shift in when intervention could realistically begin.

What earlier warning actually enables

The practical value of a 15-year early warning window comes down to timing relative to intervention. Most established approaches for reducing cardiovascular risk — changes to diet and exercise, blood pressure and cholesterol management, smoking cessation, and various preventive medications — work considerably better when applied well before disease processes have progressed to the point of causing measurable organ damage or triggering an acute cardiac event. A risk score capable of identifying elevated risk over a decade before symptoms would otherwise appear gives patients and physicians a genuinely different kind of runway to work with, compared to risk assessments that primarily reflect a person's current, already-elevated risk profile.

That said, it's worth being clear about what this study demonstrates and what it doesn't. The published research establishes that CardiOmicScore can predict future disease risk with meaningful accuracy using retrospective data from the UK Biobank population — it's a validated predictive model built and tested against a large existing dataset, not yet a tool that has been deployed and studied in real-world clinical settings tracking whether early identification through this specific test actually changes patient outcomes when acted upon. That distinction between predictive accuracy in a research dataset and demonstrated clinical benefit in actual practice is an important one, and it's typically the next major hurdle any risk-prediction tool must clear before becoming a standard part of routine healthcare.

How this fits alongside existing risk assessment tools

CardiOmicScore isn't necessarily positioned to replace the conventional risk factors doctors already rely on — age, blood pressure, cholesterol levels, and smoking history remain genuinely informative and inexpensive to measure. Rather, the tool's value proposition centers on adding a considerably more detailed, current, and dynamic layer of information on top of those established measures, particularly for patients whose conventional risk factors might look unremarkable on the surface while underlying molecular changes are already signaling elevated future risk that traditional metrics simply aren't sensitive enough to detect.

That framing matters for how this kind of research typically progresses from publication toward actual clinical use. A tool this data-intensive — requiring analysis of nearly 3,000 individual proteins and over 150 metabolites per blood sample — faces real practical questions around cost, accessibility, and the infrastructure needed to process this volume of molecular data at the scale routine clinical screening would require. Whether CardiOmicScore or a similar multiomics approach eventually becomes standard practice in cardiovascular risk assessment will likely depend as much on solving those practical deployment questions as on the underlying predictive science, which this Nature Communications publication suggests is already genuinely promising.

*This article was researched using publicly available reporting from Nature Communications, the University of Hong Kong's LKS Faculty of Medicine, ScienceDaily, Technology Networks, SciTechDaily, Knowridge, and MIMS coverage of the peer-reviewed study led by Professor Zhang Qingpeng and colleagues. It is intended for informational purposes and is not medical advice.*