Mobile Phone Detection of Atrial Fibrillation With Mechanocardiography

atrial fibrillation ◼ mobile health B ecause of the frequent asymptomatic presentation of atrial fibrillation (AF), stroke is too often its first manifestation. 1 For effective stroke prevention, timely diagnosis of AF is crucial.Mobile devices are becoming ubiquitous, providing significant possibilities for screening applications.In mechanocardiography, mechanical cardiac activity is recorded with accelerometers and gyroscopes, standard components of modern smartphones.2 In our previous proof-of-concept study, smartphone mechanocardiography demonstrated 94% sensitivity and 100% specificity for detecting AF among 39 subjects.2 Here, we validate smartphone mechanocardiography detection of AF against visual interpretation of telemetry electrocardiographic recordings in hospitalized patients.For the present case-control study, 150 consecutive patients in AF and 150 ageand sex-matched patients in sinus rhythm (SR) were enrolled from the cardiology and internal medicine wards of Turku University Hospital, Finland, between April and September 2017.After informed consent was obtained, a 3-minute mechanocardiography recording was acquired from each subject with a Sony Xperia smartphone placed on the sternum, and a simultaneously obtained 5-lead telemetry electrocardiography (Philips IntelliVue MX40) recording was used as the comparison method to assess rhythm and the number of supraventricular and ventricular extrasystoles.Electrocardiographic rhythm classifications were confirmed by 2 independent cardiologists, and a third cardiologist made the final decision if interpretations diverged.In addition, physical measurements were recorded, and electronic patient records were searched for the subjects' clinical history and investigations conducted during the index hospitalization.The institutional ethics review board approved the study protocol.The mechanocardiography recordings were analyzed with an algorithm developed beforehand by investigators blinded to the underlying rhythm.The data were first preprocessed by applying a band-pass filter to remove signal noise and bias.The algorithm then examined each of the 6 data axes of the signal with 5-second autocorrelation windows to find evidence of constant beat-to-beat intervals.Finally, for classification as AF or SR, the share of signal segments with regularity was determined.A visual presentation of mechanocardiography data is shown in the Figure.The mean age of all subjects was 74.8 years (95% confidence interval [CI], 73.7-75.9),and 132 (44.0%) were female.The mechanocardiography algorithm correctly classified AF in 143 of 150 cases and SR in 144 of 150 controls.Altogether, 4 of the 6 cases in SR misclassified as AF had marked sinus arrhythmia, whereas no