Atrial fibrillation is an ECG rhythm with a significant mortality due to stroke. Its detection is essential so that suitable treatment can be initiated. Although the detection of atrial fibrillation is not easy in the general population, the detection of patients who are not currently in fibrillation, but who are likely to develop this rhythm, is particularly difficult. There is, however, some evidence that the characteristics of some ECG rhythms are associated with the development of atrial fibrillation.
The objective of this study was to detect those patients most likely to develop atrial fibrillation from a data set made available through PhysioNet, and also in those patients to detect the ECG rhythm strip closest to the initiation of fibrillation. From the learning set of one hundred series of RR intervals of 50 patients, 25 without atrial fibrillation (normal) and 25 who subsequently developed atrial fibrillation, an algorithm was developed to detect the presence of atrial ectopic beats, separating them from ventricular ectopic beats and noise. This was achieved by identifying short intervals with no or small compensatory pauses. In the learning set, 37/50 abnormal and 34/50 normal patients were identified, giving a potential screening accuracy of 71%. As a prediction test to detect the series closest to atrial fibrillation, 19/25 were correctly identified using the highest atrial counts closest to the onset of atrial fibrillation. When the algorithm was assessed on the test data, a total of 29/50 were correctly assigned to the normal and fibrillation groups (reference 20010430.191646 entrant 11), and a 39/50 score obtained in detecting the series closest to the development of atrial fibrillation (reference 20010430.194799 entrant 11).
Supported by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under NIH grant number R01EB030362.