The Duke Clinically Linked Electrocardiogram and Arrhythmia Repository

Artificial intelligence-electrocardiogram (ECG) algorithms are becoming increasingly powerful tools in the clinical world, allowing for earlier and subclinical detection of ubiquitous and harmful pathologies. However, the strength of these algorithms is directly informed by the strength of the data sets on which they are trained, and there is a growing need for large-scale, diverse, publicly available ECG repositories. The Duke Clinically Linked ECG and Arrhythmia Repository is a large-scale ECG database containing over 5 million clinically acquired ECGs linked to longitudinal clinical data. Its strengths-including demographic diversity, large sample size, linked clinical databases, breadth of represented conditions, access to linked procedure codes, various options for data export, and ease in creating targeted subcohorts-position it as a significant tool in the artificial intelligence-ECG research landscape. This Methods paper outlines the demographics and structure of this database and explores promising research applications of this resource.