A single chest x-ray predicted 10-year major adverse cardiac events as effectively as a traditional risk-prediction score.

Recent advancements in artificial intelligence have enabled the use of “deep learning,” which involves processing complex patterns in pictures, text, and other data, to predict patient outcomes based on plain-film chest x-rays — likely by identifying coronary calcium deposits that go unrecognized by the human eye (Radiol Cardiothorac Imaging 2021; 3:e200486). Harvard investigators developed a deeplearning computer model using routine chest x-rays from ≈40,000 participants in a longitudinal cancer screening trial and evaluated its ability to predict 10-year risk for major adverse cardiac events (MACEs) in a separate group of 11,000 patients. Twenty percent of patients had both routine chest x-rays and sufficient electronic medical record data to estimate risk with a traditional cardiovascular disease risk score (ASCVD calculator).

Statin eligibility (i.e., 10-year risk for MACE ≥7.5%) was predicted similarly by the chest x-ray–based model and the traditional risk score (37% of patients were statin-eligible under each model). Moreover, adjustment for traditional risk score indicated that the chest x-ray–based model added substantial predictive ability, such that the 10-year incidence of MACE was 88% higher in statin-eligible patients than in statin-ineligible patients, independent of traditional risk score.

COMMENT
Although deep-learning models might not be actionable immediately by individual clinicians, healthcare systems might consider using existing chest x-rays to identify patients at elevated risk for MACE when other clinical factors are not available electronically. Such information then could be routed to physicians and patients for discussion or risk reduction. The authors do not recommend ordering screening chest x-rays to assess cardiovascular disease risk. — Daniel D. Dressler, MD, MSc, MHM, FACP

Dr. Dressler is Professor of Medicine at Emory University School of Medicine, Atlanta.

Weiss J et al. Deep learning to estimate cardiovascular risk from chest radiographs: A risk prediction study. Ann Intern Med 2024 Mar 26; [e-pub]. (https://doi.org/10.7326/M23-1898)

GKB-NON-2024-00277