A better way to predict a patient’s risk of coronary artery disease

LA JOLLA, CA—Coronary artery disease (CAD) is the leading cause of death in the United States. Although effective preventative treatments exist, these measures are often underutilized, in part because people don’t know they’re at risk of CAD until it’s too late.

Now, scientists at the Scripps Research Translational Institute have developed a machine learning model that more accurately estimates a patient’s risk of CAD compared to the standard clinical practice, which is based primarily on age. The findings, published in Nature Medicine on April 16, 2025, leveraged data that spanned 10 years. Their new model is personalized and integrates factors including genetics, lifestyle and medical history, enabling clinicians to provide patients with advice and preventative treatment tailored to their individual needs.

“I think more precise and personalized risk prediction could motivate patients to engage in early prevention,” says senior author Ali Torkamani, PhD, professor and director of Genomics and Genome Informatics at the Scripps Research Translational Institute. “Our model first predicts the risk that a person will develop CAD, and then it provides information to allow personalized intervention.”

CAD is caused by a buildup of plaque in the heart’s arteries, which then blocks blood supply. Many people don’t realize they have CAD until they experience a heart attack, but the disease is cumulative—the earlier treatment begins, the more effective it will be.

“We tend to observe bad outcomes for CAD in patients who are in their mid-50s and older, but the disease actually begins to develop much earlier, sometimes even while people are teenagers. There’s a lot of room for us to take action,” says first author Shang-Fu ‘Shaun’ Chen, a former doctoral student in Torkamani’s group.

To enable more accurate CAD risk prediction, Torkamani’s team set out to create a model that incorporates factors beyond age, including genetic predisposition, lifestyle and medical history. They used data from the UK Biobank to train a machine learning model to recognize factors associated with CAD. Then, they tested the model on longitudinal data from a different cohort of individuals within the Biobank to see whether it could predict their risk of developing CAD over the course of 10 years based on their baseline data.

The model started with around 2,000 predictive features that could factor into CAD risk, but the team eventually whittled this list down to 53 risk factors. These included physical measurements, blood biomarkers, family medical history, mental illness, sleep duration and the presence of specific gene variants.

The new model outperformed the standard clinical model and enabled the prediction of two times as many CAD events. After 10 years of follow-up, 62.9% of individuals that the model categorized as being at highest risk had developed CAD, compared to only 0.3% of individuals in the lowest risk group.

“Compared to traditional clinical tools, the new model improved risk classification for approximately one in four individuals — helping to better identify those truly at risk while avoiding unnecessary concern for those who are not,” said Chen.

The model’s accuracy was partially due to its better ability to predict CAD in groups of people who are usually categorized as “low risk,” such as younger individuals and women.

“Our model can pick up individuals who would be considered at low risk of CAD due to their age, but who are actually high risk due to their underlying genetics,” says Torkamani.

Though the model’s accuracy depended on the inclusion of multiple factors, the team showed that genetic predisposition was by far the strongest predictor of CAD risk. This included not only genetic predisposition for CAD itself, but also for related conditions such as high blood pressure, high cholesterol and diabetes.

“The higher your genetic risk for one of those traits—high cholesterol levels or high blood pressure levels or high diabetes risk—the greater benefit you get from intervening on that particular aspect through medication or lifestyle changes,” says Torkamani.

When the team validated the model using the National Institutes of Health All of Us dataset—which includes more diverse populations than the UK Biobank—they showed it was able to predict CAD risk equally well for people with European, African and Hispanic ancestries.

Now, the researchers are planning a long-term clinical trial to test whether informing patients of their risk of CAD can help prevent the disease.

“We think the most important thing is for patients to be aware of their individual risks so that they can receive the appropriate treatments and make lifestyle changes,” says Chen.

In addition to Torkamani and Chen, authors of the study, “Meta-Prediction of Coronary Artery Disease Risk,” include Hossein J. Sadaei, Ahmed Khattab, Corneliu Henegar, Nathan E. Wineinger and Evan D. Muse of Scripps Research; Sang Eun Lee of University of Ulsan College of Medicine; Jun-Bean Park of Seoul National University Hospital; and Jei-Fu Chen of Memorial Sloan Kettering Cancer Center.

This work was supported by funding from the National Institutes of Health (R01HG010881 and UM1TR004407) and the Scripps Research HPC team, including JC Ducom and Lisa Dong.