By admin 
Cardiovascular diseases continue to cast a long shadow over global health, remaining the undisputed leading cause of death worldwide. The sheer scale of this challenge is staggering, with approximately 19.8 million lives tragically lost to CVDs in 2022 alone. This makes CVDs a more significant threat than all cancers combined, placing immense strain on healthcare systems and economies across the globe. The economic burden is equally formidable, encompassing direct medical costs, lost productivity, and premature mortality, collectively amounting to trillions of dollars annually. Despite decades of research and public health campaigns, the prevalence and impact of these conditions underscore an urgent need for more effective strategies, particularly in the realm of early detection and prevention.
Traditionally, doctors assess an individual’s cardiovascular risk using a set of well-established clinical factors. These include age, blood pressure measurements, cholesterol levels (both LDL and HDL), body mass index (BMI), smoking history, family history of heart disease, and the presence of conditions like diabetes. While these conventional indicators are undoubtedly valuable and form the cornerstone of current risk stratification models, they inherently possess limitations. They often reflect risk factors that are already well-established or changes that have progressed to a point where they are physiologically observable. Crucially, these standard measurements may not be sensitive enough to reveal the very earliest biological changes—the subtle molecular shifts—that are silently taking place inside the body long before a disease becomes overtly apparent or symptoms begin to manifest. As a result, many individuals who are silently accumulating risk may not be identified as high-risk until their biological clocks have advanced considerably, narrowing the window of opportunity for truly effective and timely prevention through lifestyle interventions or early pharmacological treatment. This often leads to a reactive approach to healthcare, where interventions are initiated only after significant disease progression has occurred.
Adding another layer to risk assessment, genetic risk tests have emerged as a powerful tool to estimate a person’s inherited likelihood of developing various diseases. Polygenic risk scores (PRS), for instance, represent a sophisticated method that combines the cumulative effects of thousands, or even millions, of common genetic variants across an individual’s genome into a single, composite measure of inherited risk. PRS can identify individuals born with a higher predisposition to certain conditions, including various CVDs. However, while incredibly informative about an individual’s innate genetic susceptibility, a person’s genetic makeup is largely fixed at birth. This inherent immutability means that genetic scores, while predictive of baseline risk, cannot fully capture or reflect the dynamic, ongoing changes caused by a multitude of environmental and lifestyle factors. These include dietary habits, levels of physical activity, the natural process of aging, the onset of intercurrent illnesses, exposures to environmental toxins, or other myriad influences that constantly interact with our genes to shape our health trajectory. Therefore, there has been a pressing need for a diagnostic tool that offers a more current, dynamic, and comprehensive picture of an individual’s physiological state, reflecting real-time biological responses to their environment and lifestyle choices. CardiOmicScore was specifically designed to bridge this gap, aiming to provide precisely this more immediate and nuanced insight into what is actively transpiring within the body’s complex biological systems.
To construct this innovative AI tool, the HKUMed team employed advanced deep learning methodologies, a subset of artificial intelligence particularly adept at processing complex patterns within vast datasets. Their approach was distinguished by its reliance on multiomics, a cutting-edge field that integrates several layers of biological information to provide a holistic view of biological systems. Multiomics transcends the limitations of studying individual biological components in isolation by bringing together diverse data types, including genomics, metabolomics, and proteomics.
Genomics, the study of an organism’s entire genetic material, provides insights into an individual’s inherited predispositions and the blueprint of life. Proteomics, on the other hand, focuses on the large-scale study of proteins – the workhorses of the cell – which carry out virtually all essential functions in the body, from structural support to enzymatic reactions and cell signaling. Changes in protein expression or modification can be direct indicators of disease processes. Metabolomics delves into the study of small molecules called metabolites. These are the end products of cellular processes, produced as the body processes food, generates energy, responds to disease, or interacts with its environment. The metabolome offers a real-time snapshot of physiological status, reflecting both genetic predispositions and environmental influences. By combining these ‘omic’ layers, the researchers aimed to build a comprehensive biological profile that captures both inherited risks and the dynamic, real-time physiological responses of the body to its internal and external environment.
The foundational data for training CardiOmicScore came from the UK Biobank, a world-renowned biomedical database containing in-depth genetic and health information from half a million UK participants. This large-scale, longitudinal population dataset is invaluable for AI research due to its comprehensive and high-quality measurements. The HKUMed model meticulously examined an astounding 2,920 circulating proteins and 168 metabolites, all measured in blood samples collected from participants. Together, this vast array of molecules provides an incredibly detailed and dynamic snapshot of a person’s current biological state. They can reflect subtle, often imperceptible, changes in fundamental biological processes such as immune activity, metabolic regulation, and vascular health, long before any noticeable symptoms develop or conventional clinical markers begin to flag an issue. This granular molecular insight is what empowers CardiOmicScore to identify individuals at risk far earlier than previously possible.
Professor Zhang Qingpeng, Associate Professor in the Department of Pharmacology and Pharmacy at HKUMed, and a leading mind behind this research, elucidated the core philosophy: "Genes determine where we start – they define our baseline health risk, the hand we’re dealt. However, proteins and metabolites reflect our current physical health, the dynamic interplay between our genetic predispositions and our lifestyle choices and environmental exposures. Our AI tool is designed to decode these incredibly complex molecular signals, which are far too intricate for the human eye or conventional statistical methods to fully grasp. By doing so, we enable doctors and patients to identify risks much earlier than ever before, which can potentially change the entire trajectory of disease through timely lifestyle modifications, targeted preventive therapies, and proactive health management." He emphasized that this represents a fundamental shift from a reactive "sick care" model to a proactive "health care" paradigm.
The results of the study were compelling, unequivocally demonstrating that CardiOmicScore could successfully translate complex molecular measurements into highly personalized and actionable estimates of cardiovascular risk. The system performed substantially better than conventional polygenic risk scores when tested against the same cohort, indicating its superior ability to capture dynamic risk. Furthermore, its predictive accuracy improved even further when researchers integrated basic clinical information such as age and gender, highlighting the synergy between multiomic data and established demographic factors in refining risk prediction. This integration underscores that CardiOmicScore is not intended to replace existing clinical assessment but rather to augment and enhance it, providing a deeper layer of biological insight.
The model was specifically designed to assess the risk of six critical cardiovascular diseases, each carrying significant health burdens:
- Coronary Artery Disease (CAD): The most common type of heart disease, caused by plaque buildup in the arteries supplying blood to the heart, leading to heart attacks.
- Stroke: Occurs when blood flow to a part of the brain is interrupted, either by a clot (ischemic stroke) or a ruptured blood vessel (hemorrhagic stroke), leading to brain damage.
- Heart Failure: A chronic, progressive condition in which the heart muscle is unable to pump enough blood to meet the body’s needs.
- Atrial Fibrillation (AFib): An irregular and often rapid heart rate that can lead to blood clots in the heart, significantly increasing the risk of stroke and other heart-related complications.
- Peripheral Artery Disease (PAD): A circulatory condition in which narrowed blood vessels reduce blood flow to the limbs, most commonly the legs, causing pain and increasing the risk of heart attack and stroke.
- Venous Thromboembolism (VTE): A dangerous condition involving blood clots that form in a vein, which can be life-threatening if they travel to the lungs (pulmonary embolism) or cause deep vein thrombosis in the legs.
The ability of CardiOmicScore to identify elevated cardiovascular risk among high-risk individuals up to 15 years before symptoms actually emerged is perhaps its most astonishing and impactful finding. This extended lead time is unprecedented in cardiovascular risk assessment and offers a transformative opportunity for early intervention.
This research embodies a broader, fundamental shift occurring within precision medicine. For years, genomic approaches provided a relatively fixed estimate of inherited risk, offering a valuable but static view of an individual’s predispositions. However, the advent of multiomics tools like CardiOmicScore heralds a new era, offering a far more dynamic and nuanced assessment by continuously tracking the ever-changing tapestry of biological signals within the body. This allows for a truly personalized and adaptive approach to health management, where interventions can be tailored not just to genetic risk, but to real-time physiological status.
In the foreseeable future, the vision is that a small, routine blood sample could potentially be utilized to generate a detailed, multi-faceted risk profile covering several cardiovascular diseases simultaneously. This comprehensive information, processed by advanced AI, could empower both patients and their healthcare providers with an unprecedented window of opportunity – more time to respond with targeted lifestyle changes (such as diet and exercise modifications), closer monitoring protocols, or even early pharmacological interventions where appropriate. This proactive stance holds the promise of not just delaying, but potentially preventing, the onset of debilitating and often fatal cardiovascular events.
Professor Zhang further elaborated on the profound implications of their work, stating, "Our ultimate aim is to leverage cutting-edge technology like AI and multiomics to identify and prevent diseases before they ever fully develop. This isn’t just about managing illness; it’s about fundamentally reshaping the healthcare landscape. By shifting health management from a reactive treatment model, where we often intervene once damage is done, to one of proactive prediction and early intervention, we aim to create a lasting and significant impact for both public health at large and for the individualized care of each patient." This paradigm shift envisions a future where personalized prevention becomes the norm, significantly reducing the global burden of cardiovascular disease, enhancing quality of life, and extending healthy lifespans. Future research will undoubtedly focus on validating CardiOmicScore in diverse global populations, refining its predictive capabilities, and developing clear clinical pathways for its integration into routine medical practice, paving the way for a healthier future for all.
The pioneering study was collaboratively led by Professor Zhang Qingpeng, an Associate Professor jointly appointed in the Department of Pharmacology and Pharmacy at HKUMed and the HKU Musketeers Foundation Institute of Data Science (IDS). The primary authorship for this significant publication was attributed to Luo Yan, a talented researcher also affiliated with the HKU IDS, underscoring the interdisciplinary expertise in both medicine and advanced data science that underpinned this breakthrough.
