AI improves prediction of heart attack, cardiac death
Machine learning (ML) can predict a patient’s long-term risk of myocardial infarction (MI) or cardiac death, according to new findings published in Cardiovascular Research.
“With constant growth of artificial intelligence across various disciplines, especially in cardiology, robust cardiac risk assessment will benefit from quantification of automated imaging biomarkers, increasing the relevant information available for clinical decision making,” wrote lead author Frederic Commandeur, PhD, Cedars-Sinai Medical Center in Los Angeles, and colleagues.
ML techniques have already been used to predict all-cause mortality in patients who show signs of coronary artery disease, but can they predict future cardiovascular events after coronary artery calcium (CAC) scoring? That’s the exact question the authors hoped to answer with their research.
Commandeur et al. explored data from nearly 2,000 asymptomatic patients who participated in the prospective EISNER (Early Identification of Subclinical Atherosclerosis by Noninvasive Imaging Research) trial. All patients underwent CAC scoring using non-contrast CT scans, and the researchers had approximately 15 years of long-term follow-up information to review. Follow-up data was collected through mailed questionnaires, telephone interviews, clinical visits and official death registry data.
Overall, the extreme gradient boosting (XGBoost) approach to ML—"which has shown to outperform other algorithms for sets of features in a variety of tasks”—resulted in an area under the ROC score (AUC) of 0.82, which was higher than both atherosclerotic cardiovascular disease (ASCVD) risk scores (0.77) and CAC scores (0.77).
“In this prospective trial, ML demonstrated high performance in risk assessment for MI and cardiac death in asymptomatic subjects,” the authors wrote. “By objectively combining clinical data and quantitative CT measures, ML provided significantly superior risk prediction compared with the ASCVD risk score or CAC score. These promising results suggest that ML has a potential for clinical implementation to improve risk assessment.”
Both ASCVD risk scores and CAC scores did contribute to the strong performance of XGBoost. Age an systolic blood pressure were also found to play key roles in risk prediction.
What’s the next step for this research? Going forward, the team thinks there is an opportunity to implement ML methods as a day-to-day part of patient care.
“In the future, we foresee ML working in the background of standard coronary calcium scoring and electronic reporting software, gathering the variables automatically and allowing ‘on-the-fly’ risk score computation by integrating all relevant measures of clinical risk and imaging biomarkers from coronary calcium scoring scans,” Commandeur and colleagues concluded.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.