ESC: Smartphone Makes a Smart Afib Diagnosis

ROME -- The accelerometers and gyroscopes in a smartphone can do more for mobile health monitoring than measuring exercise or sleep, found researchers in an early study showing accuracy for diagnosing atrial fibrillation.

A smartphone app using recordings from the phone, placed directly on the chest with no hardware or add-on device, had 98.5% sensitivity and 95.2% specificity for the arrhythmia, Tero Koivisto, MD, of the University of Turku, Finland, and colleagues reported here at the European Society of Cardiology meeting.

The out-of-bag classification error in Random Forests, a measure of prediction error for machine learning models like the one used for the app's algorithm, was 4.95%.

The study included 16 patients with chronic atrial fibrillation from the Turku Heart Center and 20 recordings from healthy volunteers to validate the performance of the algorithm used by the app.

For its use, the patient lies down and places the smartphone on their chest. The data captured by the accelerometer and gyroscope of the phone is automatically processed and filtered by the app, then uploaded to a desktop (or, in the future, the cloud) where a machine learning algorithm determines presence of atrial fibrillation.

The strategy is device agnostic, able to be used on Apple or Android devices. The pilot study included both men and women, and Koivisto suggested gender shouldn't make a difference in performance.

Koivisto suggested screening with the app would be particularly helpful for older adults and those with prior stroke as well as in resource-limited settings. However, it cannot distinguish atrial fibrillation from other arrhythmias at this time, and people would need to have the screening diagnosis confirmed by an ECG in the office.

Dan Atar, MD, PhD, of the University of Oslo, commented that the atrial fibrillation guidelines released by the ESC at the conference emphasize the need to search for silent atrial fibrillation. "And here we have an excellent opportunity to increase this search function. I think it's a very valuable approach."

Gerhard Hindricks, MD, PhD, of Germany's Leipzig University Heart Center, who co-chaired a press conference at which the findings were discussed, called the strategy a "very smart application without any significant add-ons, which for the economical balance and the implementation of the workflow it's really beautiful."

Click here for MedPage Today's ESC 2016 video page, which includes comments from authors of the Hot Lines trials, and leading cardiologists from around the world providing daily commentary.

"What's the crux with all these detectors of atrial fibrillation is sensitivity and specificity," he cautioned. "You have to find what you look for."

Some prior devices or strategies for smartphone detection have a high incidence of false-positive results, "which completely undermines the trust in the technology," or missed cases, he noted, calling for further validation.

The Reveal implantable cardiac monitor showed sensitivity of 96% and specificity of 85.4% for atrial fibrillation detection compared with ECG in the XPECT trial.

In another study, the sensitivity and specificity, respectively, of the Cardiio Rhythm smartphone application was 92.9% and 97.8% and for the AliveCor heart monitor was 71.4% and 99.4%.

Addressing Koivisto, Hindricks concluded "How specific is your technology or your application of an existing technology for the detection of atrial fibrillation really deserves further in-depth and ECG-in-parallel controlled studies to explore that."

Comment