An automated algorithm for the detection of atrial fibrillation in the presence of paced rhythms

Approximately 5% of diagnostic ECGs acquired in the hospital setting are from patients with pacemakers. A significant percentage of these patients are in atrial fibrillation or flutter (AF), and have increased risk of stroke. Many automated diagnostic algorithms abort analysis when paced rhythms are identified. The Philips DXL algorithm can detect AF in the presence of paced rhythms and provides interpretations for both rhythms. The algorithm uses QRST subtraction with frequency domain analysis of the residual. A decision tree classifier uses features from the power spectrum, as well as irregularity of non-paced beats. On a training database of 355 paced ECGs with 265 in AF, the algorithm had AF detection sensitivity of 76%, PPV of 82%, and specificity of 73%. On a testing set of 1,057 paced ECGs with 194 AF cases, the algorithm had sensitivity of 71%, PPV of 83%, and specificity of 97%. Automated detection of AF in the presence of pacing is a clinically valuable tool to assist cardiologists in ECG diagnosis, and can be done with relatively high accuracy.