Time-Dependent ECG-AI Prediction of Fatal Coronary Heart Disease

Background Sudden cardiac death (SCD) affects >4 million people globally, and ∽300,000 yearly in the US. Fatal coronary heart disease (FCHD) is used as a proxy to SCD when coronary disease is present and no other causes of death can be identified. Electrocardiographic (ECG) artificial intelligence (AI) models (ECG-AI) show promise in predicting adverse coronary events yet their application to FCHD is limited. Objectives This research aimed to develop accurate ECG-AI models to predict risk for FCHD within the general population using waveform 12- and single-lead ECG data as well as assess time-dependent risk. Methods Standard 10-second 12-lead ECGs sampled at 250Hz, demographic and clinical data from University of Tennessee Health Science Center (UTHSC) were used to develop and validate models. Eight models were developed and tested: two classification models with convolutional neural networks (CNN) using 12- and single-lead ECGs as inputs (12-ECG-AI and 1-ECG-AI, respectively) and six time- dependent cox proportional hazard regression (CPHR) models using demographics, clinical data and ECG-AI outputs. The dataset was split into 80% for model derivation, with five-fold cross-validation, and 20% holdout test set. Models were evaluated using the AUC and C-Index. Correlation of predicted risks from the 12-lead (12-ECG-AI) and single-lead (1-ECG-AI) CNN models was assessed. Results A total of 50,132 patients were included in this study (29,093 controls and 21,039 cases) with a total of 167,662 ECGs with mean age of 62.50±14.80years, 53.4% males and 48.5% African-Americans. The 12- and 1-ECG-AI models resulted AUCs=0.77 and 0.76, respectively on the holdout data. The best performing model was C12-ECG-AI-Cox (demographics+clinical+ECG) with no time restriction AUC=0.85(0.84-0.86) and C-Index= 0.78(0.77-0.79). 2-year FCHD risk prediction reached AUC=0.91(0.90-0.92). The 12-/1-ECG-AI models’ predictions were highly correlated (R2 = 0.72). Conclusion 2-year risk for FCHD can be predicted with moderate accuracy from ECG data alone. When combined with other data, a very high accuracy was obtained. High correlation between single-lead and 12-lead ECG models infer opportunities for screening larger patient populations for FCHD risk. ![Figure][1] ### Competing Interest Statement The authors have declared no competing interest. ### Funding Statement This project was funded by an internal grant awarded by the Wake Forest Biomedical Informatics Center ### Author Declarations I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained. Yes The details of the IRB/oversight body that provided approval or exemption for the research described are given below: The study was approved by the IRB of both University of Tennessee Health Science Center, Memphis, TN and Atrium Health Wake Forest Baptist, Winston-Salem, NC. I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals. Yes I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance). Yes I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable. Yes Data used was obtained from the University of Tennessee Health Science Center under a Data User Agreement. Final AI models will be shared on GitHub after manuscript publication. [1]: pending:yes