Artificial Intelligence can now predict long-term risks of heart attack and cardiac death

A new study in Cardiovascular Research finds that machine learning, the patterns and inferences computers use to learn to perform tasks, can predict the long-term risk of heart attack and cardiac death. Indeed, machine learning appears to be better at predicting heart attacks and cardiac deaths than the standard clinical risk assessment used by cardiologists.

Researchers here studied subjects from the imaging arm of a prospective, randomized research trial, who underwent coronary artery calcium scoring with available cardiac CT scans and long-term follow-up. Participants here were asymptomatic, middle-aged subjects, with cardiovascular risk factors, but no known coronary artery disease.

Researchers used machine learning to assess the risk of myocardial infarction and cardiac death in the subjects, and then compared the predictions with the actual experiences of the subjects over fifteen years. Subjects here answered a questionnaire to identify cardiovascular risk factors and to describe their diets, exercise and marital status.

The final study consisted of 1,912 subjects, fifteen years after they were first studied. 76 subjects presented an event of myocardial infarction and/or cardiac death during this follow-up time. The subjects’ predicted machine learning scores aligned accurately with the actual distribution of observed events. The atherosclerotic cardiovascular disease risk score, the standard clinical risk assessment used by cardiologists, overestimated the risk of events in the higher risk categories. Machine learning did not. In unadjusted analysis, high predicted machine learning risk was significantly associated with a higher risk of a cardiac event.

“Our study showed that machine learning integration of clinical risk factors and imaging measures can accurately personalize the patient’s risk of suffering an adverse event such as heart attack or cardiac death,” said While machine learning models are sometimes regarded as “black boxes”, we have also tried to demystify machine learning; in this manuscript, we describe individual predictions for two patients as examples. When applied after the scan, such individualized predictions can help guide recommendations for the patient, to decrease their risk of suffering an adverse cardiac event. ”

###

Direct questions about the study to:

Damini Dey, PhD

Biomedical Imaging Research Institute

Cedars-Sinai Medical Center

116 N Robertson Blvd.

Los Angeles, CA 90048


[email protected]

To request a copy of the study, please contact:

Emily Tobin


[email protected]

Sharing on social media? Find Oxford Journals online at

@OxfordJournals

Please acknowledge Cardiovascular Research as a source in any articles.

DOI: 10.1093/cvr/cvz321

Funding: This work was funded by National Institute of Health/National Heart, Lung, and Blood Institute grant 1R01HL133616 and in part by Bundesministerium für Bildung und Forschung 335 (01EX1012B, Spitzencluster Medical Valley), and a grant from the Miriam and Sheldon G. Adelson Medical Research Foundation.