Smartwatch as a detector for AF?
A smartwatch photoplethysmography coupled with a mobile application using a machine learning algorithm can passively and accurately detect atrial fibrillation (AF) in patients undergoing cardioversion, albeit with less accuracy in ambulatory individuals with self-reported AF, a proof-of-concept study shows.
“AF is often asymptomatic and thus can remain undetected until a thromboembolic event occurs,” according to the researchers. “A readily accessible means to continuously monitor for AF could prevent large numbers of strokes and death.”
The multinational observational study involved three cohorts: a remote cohort of 9,750 participants (mean age 42 years, 63 percent male, and 347 with AF), an external validation cohort of 51 patients undergoing cardioversion, and an exploratory cohort of 1,617 ambulatory participants (of which 64 self-reported having AF). [JAMA Cardiol 2018;doi:10.1001/jamacardio.2018.0136]
The machine learning algorithm employing deep neural network was trained using over 139 million heart rate measurements collected from the remote cohort.
In the external validation cohort, the deep neural network demonstrated a C statistic of 0.97 (p<0.001)—indicating a good fit in the regression model for AF detection against the reference standard diagnosis using 12-lead ECG. The sensitivity for detecting ECG-diagnosed AF was 98.0 percent, with a specificity of 90.2 percent. The accuracy did not change after normalizing all tracings by heart rate.
In an exploratory analysis using ambulatory cohort, individuals with self-reported AF were correctly classified with modest accuracy—the C statistic for AF detection dropped to 0.72 (95 percent confidence interval, 0.64–0.78), with a sensitivity of 67.7 percent and a specificity of 67.6 percent. Nonetheless, AF prediction by neural network remained significantly associated with persistent AF (adjusted odds ratio, 1.98; p=0.02), after adjusting for age, sex, diabetes, hypertension, coronary artery disease, heart failure, and ethnicity.
“[Although the] exploratory analysis demonstrating that our deep neural network substantially outperforms standard techniques [based on AF risk factors] to detect self-reported persistent AF from ambulatory data … this proof-of-concept experiment likely demonstrates the challenges of accurately detecting ambulatory arrhythmia among constantly mobile individuals in natural environments,” said the researchers.
What’s ahead for wearable AF monitor?
“The ideal instrument for AF detection would be noninvasive and provide real-time, accurate AF detection in a passive fashion—specifically, not requiring the user to remember to perform some action and not limited to any one snapshot in time,” the researchers stated. “Smartwatches are well positioned to accomplish these goals in a cost-efficient and resource-efficient fashion.”
Noting that continuously wearing smartwatches might lead to increased new AF diagnoses and costs associated with increased demand for care, the researchers believed that “the potential reduction in stroke could ultimately provide cost savings.”
However, as the study focused on individuals with known AF, the researchers acknowledged that they “did not demonstrate an ability to identify new diagnosis of the disease.”
“Despite the excellent test characteristics observed among sedentary patients undergoing cardioversion, the modest performance in the ambulatory scenario, a context more representative of the ultimate application of this technology, suggests that these data should be primarily interpreted as a proof of concept,” they concluded.
“As a field, we need to think creatively but prudently about finding optimal use cases for these technologies—assistive vs independent diagnostician, screening vs disease management, retail deployment vs prescription only, patient self-management vs clinician feedback loops ... the list goes on,” wrote Dr Mintu Turakhia from the Veterans Affairs Palo Alto Health Care System in Palo Alto, California, US, in an editorial. [JAMA Cardiol 2018;doi:10.1001/jamacardio.2018.0207]