Research shows CardiOmicScore predicts cardiovascular disease risk up to 15 years — Evidence Review
Published in Nature Communications, by researchers from Department of Pharmacology and Pharmacy at the LKS Faculty of Medicine of the University of Hong Kong
Researched by
— the AI search engine for science
Table of Contents
Researchers at the University of Hong Kong have developed an AI-driven blood test, CardiOmicScore, which predicts the risk of six major cardiovascular diseases up to 15 years in advance. Related studies broadly support the use of AI and multiomics approaches for improved cardiovascular risk prediction.
- Multiple studies have demonstrated that machine learning and AI models, especially those integrating genetic, proteomic, and metabolomic data, offer enhanced predictive accuracy for cardiovascular diseases compared to traditional risk scores, aligning with the new findings 1 2 4.
- Existing research supports the identification of novel biomarkers and the use of advanced algorithms—such as XGBoost, SVM, and ensemble models—for early detection and risk stratification, which is conceptually similar to CardiOmicScore’s approach 1 3 4.
- Systematic reviews highlight the promise of AI in cardiovascular risk prediction but also note challenges such as lack of external validation and potential biases, suggesting that while the new tool is an advance, further validation is needed for clinical adoption 4 5.
Study Overview and Key Findings
Cardiovascular diseases (CVDs) remain the leading cause of death globally, with many cases developing insidiously over years before symptoms manifest. Early detection is crucial for preventive intervention, but traditional clinical risk assessments often miss subtle, preclinical changes. This study introduces CardiOmicScore, an AI-based tool that leverages multiomics data—genomics, proteomics, and metabolomics—from a single blood test to predict the risk of six major CVDs up to 15 years before clinical onset. Unlike fixed genetic risk scores, CardiOmicScore reflects real-time molecular signals, potentially enabling more dynamic and actionable risk stratification.
| Property | Value |
|---|---|
| Study Year | 2026 |
| Organization | Department of Pharmacology and Pharmacy at the LKS Faculty of Medicine of the University of Hong Kong |
| Journal Name | Nature Communications |
| Authors | Yan Luo, Nan Zhang, Jiannan Yang, Mengyao Cui, Kelvin K. F. Tsoi, Gregory Y. H. Lip, Tong Liu, Qingpeng Zhang |
| Population | Individuals at risk for cardiovascular diseases |
| Sample Size | n=2,920 proteins and 168 metabolites |
| Outcome | Risk of six major cardiovascular diseases |
| Results | CardiOmicScore predicts CVD risk up to 15 years in advance |
Literature Review: Related Studies
To situate this research within the current scientific landscape, we searched the Consensus database of over 200 million research papers. The following search queries were used:
- AI blood test cardiovascular disease prediction
- CardiOmicScore stroke heart failure risk
- long-term cardiovascular disease risk assessment
| Topic | Key Findings |
|---|---|
| How effective are AI and machine learning models at predicting CVD? | - AI/ML models, especially those using multi-omics data, consistently outperform traditional risk scores, achieving high predictive accuracy for CVD 1 2 3 4. - Systematic reviews confirm the potential of AI algorithms but highlight the need for external validation and attention to bias in model development 2 4 5. |
| What are the advantages and limitations of current CVD risk tools? | - Traditional models (e.g., SCORE2, WHO charts) are widely used but may miss early or subtle risk in some populations, and can be less dynamic than multiomics-based approaches 8 9 11. - AI models promise individualized prediction but face challenges with generalizability, transparency, and integration into clinical workflows 4 5. |
| Which biomarkers and data types improve early CVD prediction? | - Integration of transcriptomic, proteomic, and metabolomic biomarkers enhances early detection and risk stratification for CVD compared to clinical factors alone 1 4. - Studies find that real-time molecular signals reflect dynamic health changes, supporting tools like CardiOmicScore for pre-symptomatic risk assessment 1 4. |
| How do risk prediction models perform across populations and diseases? | - Most AI prediction models are still under development and require validation across diverse populations 5 8 9. - Existing models vary in performance for different CVD subtypes; novel algorithms often outperform established scores for specific outcomes like ischemic stroke in heart failure 7 8. |
How effective are AI and machine learning models at predicting CVD?
Research consistently finds that AI and machine learning (ML) approaches yield improved accuracy in predicting cardiovascular diseases compared to conventional clinical risk scores. The new CardiOmicScore tool aligns with this trend, leveraging multiomic data and deep learning to forecast risk far earlier than traditional models, a capability echoed in related studies.
- Ensemble AI/ML models have achieved predictive accuracies as high as 96–98% using omics and clinical data 1 3.
- Meta-analyses show that SVM, boosting, and custom-built AI algorithms outperform traditional models, with AUC values often exceeding 0.9 2.
- Systematic reviews highlight the promise of AI but caution that many models lack external validation and may be at risk of bias 4 5.
- CardiOmicScore’s use of deep learning on multiomics data represents a continuation and expansion of this successful application of advanced algorithms 1 4.
What are the advantages and limitations of current CVD risk tools?
Traditional risk prediction tools—such as the SCORE2 and WHO risk charts—have been instrumental in guiding prevention strategies but are limited by their reliance on static, population-level risk factors. AI-powered models, like CardiOmicScore, offer more personalized and dynamic risk assessment, but their adoption is hindered by issues of bias, transparency, and lack of external validation.
- Standard tools can miss early or subtle risk, particularly in diverse or underrepresented populations 8 9 11.
- AI models improve prediction accuracy but often lack transparency and independent validation, raising concerns about their readiness for clinical use 4 5.
- Most AI-based models are still in developmental stages and are at high risk of bias due to methodological limitations 5.
- The integration of AI into clinical practice is challenged by the need for replicability, explainability, and workflow compatibility 4 5.
Which biomarkers and data types improve early CVD prediction?
Incorporating advanced molecular biomarkers—such as transcriptomic, proteomic, and metabolomic profiles—significantly enhances early detection and individualized risk prediction for CVD. The CardiOmicScore’s multiomics approach leverages these data types to capture real-time biological changes, a strategy supported by recent literature.
- Studies have identified panels of transcriptomic biomarkers and circulating proteins that can predict CVD with high accuracy 1 4.
- Real-time molecular signals, reflecting ongoing physiological changes, provide a more dynamic and actionable risk profile than genetic or static clinical factors 1 4.
- AI tools that combine omics data with clinical features outperform those relying on a single data type 1 3 4.
- This approach enables earlier intervention and more precise risk stratification, as demonstrated by improved prediction windows in recent studies 1 4.
How do risk prediction models perform across populations and diseases?
The performance of risk prediction models, including AI-based tools, varies across different populations and CVD subtypes. Most models have been developed and validated in European and North American cohorts, with limited data on generalizability to other regions or minority groups.
- AI models are predominantly developed in Western populations, with little external validation in other regions 5 8 9.
- Novel risk scores have been shown to outperform established tools (e.g., CHADS-VASc) for specific outcomes like stroke risk in heart failure 7.
- Traditional risk charts have demonstrated considerable variation in risk estimates across global regions, highlighting the need for calibration to local epidemiology 8 9.
- The lack of diverse validation raises concerns about the applicability of new AI tools like CardiOmicScore in broader clinical settings 5 8 9.
Future Research Questions
While the new CardiOmicScore tool represents a significant advance in early CVD risk prediction, several areas require further research. Key questions involve validation in diverse populations, integration into clinical practice, long-term outcomes, and the ethical implications of AI-driven health prediction.
| Research Question | Relevance |
|---|---|
| How does CardiOmicScore perform in external validation across diverse populations? | Most AI models have not been validated outside their development cohorts, limiting generalizability and clinical utility. Examining performance in varied populations is critical for equitable risk prediction and broad adoption 5 8 9. |
| What are the clinical outcomes when CardiOmicScore is used to guide preventive interventions? | It remains to be seen whether early risk identification with CardiOmicScore leads to improved patient outcomes, reduced disease incidence, or cost-effectiveness in real-world settings 4. |
| How can AI-based risk prediction tools be integrated into existing clinical workflows? | Adoption in routine care will require seamless integration, clinician acceptance, and clear protocols for acting on risk predictions, which are current barriers for AI-based tools 4 5. |
| What ethical, legal, and privacy challenges arise from AI-driven multiomics risk prediction? | Use of sensitive omics and health data raises questions about data privacy, consent, and potential misuse, which must be addressed for responsible implementation 4. |
| Can multiomics-based AI tools like CardiOmicScore be adapted for risk prediction in other chronic diseases? | Given the success in CVD, exploring applications for cancer, metabolic, or neurodegenerative diseases could further extend the benefits of multiomics AI models 4. |
This article synthesizes current evidence and expert consensus on AI-driven, multiomics-based cardiovascular risk prediction, situating the new CardiOmicScore tool within the broader scientific landscape while outlining key directions for future research and clinical translation.