Temporal trends in myocardial ischemia risk estimated from 12-lead electrocardiograms using deep learning in individuals with suspected cancer during health checkups.

Abstract

BACKGROUND: Previous studies have suggested the potential effect of psychological stress related to cancer diagnosis on cardiovascular mortality. This study aimed to investigate the temporal trends of cardiovascular risk before and after cancer diagnosis using a deep learning model applied to 12-lead electrocardiograms (ECGs). METHODS: We developed a deep learning model using a publicly available large-scale dataset to quantify myocardial ischemia risk from 12-lead ECGs. We collected ECG records from individuals diagnosed with cancer at a university hospital who also underwent an ECG as part of a health checkup within 90 days prior to cancer diagnosis. The deep learning model was then applied to the ECGs of individuals with cancer, and the temporal trend of cardiovascular risk was examined. RESULTS: The deep learning model demonstrated high predictive performance, with an area under the receiver operating characteristic curve of 0.930 (95% confidence interval = 0.920–0.941). The model was then applied to 523 ECG records of 89 individuals with cancer. The estimated probability of ECG-indicated myocardial ischemia increased until cancer diagnosis, peaked shortly after diagnosis, and then declined. CONCLUSIONS: These findings support the immediate effect of psychological stress related to cancer diagnosis on increased cardiovascular risks.