In a study presented at the European Society of Cardiology meeting, researchers presented a health app that can diagnose atrial fibrillation using the accelerometer and gyroscope built into most smartphones.
Diagnosing atrial fibrillation (AF), particularly silent AF, has gotten a lot of attention from the digital health community. Modern innovations like the iRhythm single-lead ECG patch and AliveCor Kardia smartphone ECG have raised the possibility of low-cost, highly accurate screening of large populations.
In this study, a group of Finnish researchers presented a health app that can diagnose atrial fibrillation using only the smartphone gyroscope and accelerometer. Users of the the health app simply lay down and place their smartphone on their chest. According to the study author Tero Kovisto, vice-director of the Technology Research Centre at the University of Turku in Finland,
We use the accelerometer and gyroscope of the smartphone to acquire a heart signal from the patient. A measurement recording is taken, and the acquired data is pre-processed by signal processing methods. Multiple features such as autocorrelation and spectral entropy are then extracted from the pre-processed data. Finally, a machine learning algorithm (KSVM) is used to determine if the patient suffers from atrial fibrillation.
In their preliminary study, they tested 16 patients with chronic AF and 20 patients without AF. In that population, they reported a sensitivity and specificity of greater than 95%.
While interesting, these results, reported in an oral presentation, should be interpreted with some healthy skepticism though. We’ve previously talked about the PULSESMART study from the University of Massachusetts, University of Connecticut, and other centers. In that study, a health app uses the smartphone camera & flash to detect the pulse, analyzing variation to determine if a patient has AF. While that health app does well in a completely normal rhythm vs. chronic AF study, specificity in particular declined significantly when atrial or ventricular ectopy (extra beats) were introduced. And in the population that we’d actually want to screen for AF, older adults, that’s incredibly common. The use of machine learning rather than a fixed algorithm based on a reference population could help address that problem over time as more and more data is gathered.
The AliveCor Kardia smartphone ECG showed promising results in a community-based screening program with about 1,000 patients. The iRhythm Zio patch is now being tested in large studies involving thousands of patients. It will be interesting to see how this health app performs in larger studies with more diverse populations, both in terms of demographics and rhythm.