Photo: Seksan Mongkhonkhamsao/Getty Images
Pharmacy researchers in Hong Kong have developed an AI-driven tool that predicts cardiovascular risks beyond genetics, monitoring near-real-time biological changes before diseases become apparent.
A research team at the Li Ka Shing Faculty of Medicine of the University of Hong Kong (HKUMed) created the CardiOmicScore, using deep learning to integrate multiomics data, including genomics, metabolomics, and proteomics. They utilised population data from the UK Biobank, analysing approximately 2,920 circulating proteins and 168 metabolites from blood samples.
In a media release, the team explained that these molecular data are "real-time recorders" of the body that are sensitive to subtle changes in the immune system, metabolism, and vascular health.
FINDINGS
In a study whose findings were published in Nature Communications, CardiOmicScore was evaluated for its accuracy in predicting six common cardiovascular diseases (CVDs) – coronary artery disease, stroke, heart failure, atrial fibrillation, peripheral artery disease, and venous thromboembolism.
Combining molecular signals with other clinical information, such as age and gender, the new model made better predictions than the popular Polygenic Risk Scores, the researchers found. It could also provide early warning signals up to 15 years before clinical onset, they added.
"Current risk tools, such as the Framingham Risk Score and Polygenic Risk Score, estimate cardiovascular risk using clinical factors or genetic variants, capturing only part of the disease process. CardiOmicScore integrates metabolomic and proteomic profiles using deep learning to capture molecular changes closer to disease onset," explained study lead and HKUMed associate professor Zhang Qingpeng to Mobihealth News.
"The proteomics-based score achieved C-index values of about 0.69-0.82 and the metabolomics-based score about 0.64-0.74, compared with roughly 0.52-0.60 for Polygenic Risk Scores. When added to conventional predictors, these multiomics scores improved discrimination by ΔC-index of approximately 0.005-0.102 across six cardiovascular outcomes," he explained further.
The study also found that certain proteins linked to cardiac stress and inflammation contributed to the model's predictive accuracy. These include NPPB and GDF15, vascular remodelling markers like MMP12, and metabolites reflecting cardiometabolic and renal pathways, including creatinine and linoleic acid.
WHY IT MATTERS
As current approaches remain limited, there is an "urgent" need for tools that can capture a person's biological state and provide accurate, early warning for CVDs, said the HKUMed researchers.
According to the team, assessing clinical indicators alone, including age and blood pressure, will not capture subtle and early biological changes indicative of CVDs. Polygenic Risk Scores, which represent a step forward from this conventional approach, still cannot reflect the immediate health impacts of lifestyle or environmental changes.
The team claims that their latest study marks a "shift in precision medicine from static, gene-centric paradigm towards a more dynamic, multiomics-based approach."
"In the future, a small-volume blood sample may be sufficient to generate a comprehensive cardiovascular risk profile for multiple diseases," the HKUMed media release noted.
Prof Zhang told this publication that large-scale external validation and early clinical pilot studies "could plausibly take place over the next few years." He said their present model can be retrained to make accurate predictions using a smaller set of informative biomarkers from a standard blood sample.
How would future clinical adoption look? A/Prof Zhang said: "A patient’s [blood] sample could be analysed using proteomic or metabolomic panels, with results integrated into a clinical decision-support system to generate the CardiOmicScore alongside conventional risk factors."
Such a comprehensive molecular risk profiling, the study lead mentioned, may "become more feasible" in the future as costs decline and further validation accumulates.
THE LARGER TREND
A Singaporean startup has also developed a different approach to cardiovascular disease risk calculation. Health BETA combines genetic and lifestyle factors to predict coronary artery disease risk, powered by an algorithm trained on an Asian-focused database.
An ongoing research in Singapore, called the Project RESET Parallel Cohort, is studying data from a diverse, multi-ethnic cohort of Asians using multiomics and big data to discover molecular, metabolic, and immunological markers of subclinical cardiovascular diseases.
The CardiOmicScore project in Hong Kong is also seeking Asian cohorts to further evaluate their CVD risk prediction model.
Meanwhile, in South Korea, a research team also utilised deep learning to train a model to diagnose coronary artery disease and predict major adverse cardiac events in emergency cases.
