Introduction: In suspected chronic coronary artery disease (CAD), the probability of pretest probability is estimated according to the modified Diamond and Forrester criteria, which are based on age, gender, and symptoms. In addition, the presence of cardiovascular risk factors is further accounted for in clinical routine. However, even when combined, these criteria suffer from low accuracy, making further diagnostic testing, e.g. via coronary computed tomography angiography prior to coronary angiography (CA) necessary. Artificial intelligence (AI)-enhanced evaluations of broadly available ECGs have shown excellent results in the detection of multiple cardiovascular diseases. In the present analysis, we aimed to evaluate the ability of neural networks to predict obstructive CAD and long-term mortality in patients prior to conventional coronary angiography in addition to established cardiovascular risk factors.

Methods: Our analysis is based on the cohort of the ECAD registry, including patients undergoing conventional coronary angiography at the West German Heart and Vascular Center between 2004 and 2019. Details of the population have been described previously. Patients with a digitally available resting ECG within 90 days prior to coronary angiography, containing structured data on 648 characteristics for each ECG, were included. The overall cohort was divided in a learning (60%), a validation (20%) and a test cohort (20%). Obstructive CAD was defined by percutaneous intervention during coronary angiography, as by discretion of the interventional cardiologist. The predictive ability of neural networks using traditional cardiovascular risk factors (age, sex, systolic blood pressure, LDL-cholesterol, diabetes, smoking status and positive family history of premature CAD) and the ECG data for detection of obstructive CAD and all-cause mortality were evaluated. Calculations were run 100 times and compared by their mean area under the receiver operating curve (AUC) in the validation cohort.

Results: Data from 7076 coronary angiographies were included. Prevalence of obstructive CAD was 29.3%. For prediction of obstructive CAD, neural networks based on ECG characteristics outperformed traditional risk factors [AUC for ECG: 0.623 (0.619, 0.626), AUC for risk factors: 0.590 (0.585, 0.594)]. Combining ECG characteristics with traditional risk factors led to best prediction model [AUC: 0.656 (0.652, 0.660)]. All-cause mortality rate during the median follow up of 2.4 (0.8, 6.3) years was 16.1% (1137 patients). For prediction of all-cause mortality, neural networks based on structured ECG data provided better prediction as compared to risk factors [ECG alone: 0.707 (0.701, 0.711) , RF alone: 0.581 (0.571, 0.589), ECG+RF: 0.711 (0.704, 0.716)].

Conclusion: Neural networks based on structured data from resting ECGs improve the prediction of obstructive CAD and long-term survival in patients undergoing conventional coronary angiography. These results direct towards the utilization of AI-enhanced ECG-evaluation for clinical decision-making.