219 P wave signal analysis is able to recognize with a good accuracy patients with or without previous atrial fibrillation

Purpose Paroxysmal atrial fibrillation (PAF) is a common cardiac arrhythmia but difficult to diagnose: frequently asymptomatic and when symptomatic too sporadic to be captured by electrocardiogram (ECG) or even long duration Holter monitoring. However this diagnosis carries a major importance as it could induce key changes in diagnosis and therapy (anticoagulation). Our aim was to evaluate the accuracy of the P signal wave analysis to discriminate patients with and without previous documented AF. Methods we included patients with previously documented AF and individuals without any suspicion of this arrhythmia. A digital ECG was recorded and analyzed blindly with a home-made signal P wave algorithm. This algorithm has been described in details in peer review journals (IEEE, IFMBE). Briefly it consists of three steps: 1) isolation of every P-wave by segmentation of each ECG beat; 2) measurement of parameters extracted from the P-wave signal such as length, pattern, energy and entropy in different time-frequency cells; 3) classification in two classes by a support vector machine algorithm. Results we included 50 patients, 30 (17 M, mean age 55.9 ± 17 years) with documented PAF and 20 (18 M, mean age 35 ± 14) without history of AF. Blinded P wave signal analysis correctly classified 22 patients within the 30 in the AF group and 16 among the 20 in the control group (sensitivity 73%, specificity 80%, positive predictive value 85% and negative predictive value 67%). Conclusion Signal P wave analysis selects with a good accuracy, comparable or even better than the one reported with imaging in the diagnosis of pulmonary embolism, patients with previous AF. These results, if confirmed in multisite trials could give arguments to modify the treatment of patients with palpitations or embolism of unknown origin.