Han et al., investigates whether machine learning models which analyze daily atrial fibrillation (AF) burden on implantable cardiac devices can predict short-term risk of stroke better than more traditional clinical methods. The researchers analyzed cardiac device data from Veterans Health administration patients by applying machine learning models such as neural networks, random forests, and logistic regression. They then compared this data to the clinical tool CHA₂DS₂-VASc, to see which could better predict the risk of stroke over a period of 30 days. They tested both test and validation data. The CHA₂DS₂-VASc performed worse on both data sets, with an (area under the curve) AUC ≤ 0.5, while the random forest model performed best on test data with an AUC of 0.662 and the logistic regression model performed best on validation data with an AUC of 0.702. However, the best results came when combining the CHA₂DS₂-VASc and machine learning models, yielding a validation AUC of 0.696 and test AUC of 0.634, which were the highest averages on non-training models. These results demonstrate that AF burden patterns derived from implantable devices provide valuable data for predictive signals of short-term stroke. Machine learning is a powerful tool which provides incremental improvement in stroke prediction.