Develop and Validate Machine-Learning Algorithm to Detect Atrial Fibrillation With Wearable Devices
Develop and Validate Machine-Learning Algorithm to Detect Atrial Fibrillation With Wearable Devices
Atrial Fibrillation (AF) is an abnormal heart rhythm. Because AF is often asymptomatic, it often remains undiagnosed in the early stages. Anticoagulant therapy greatly reduces the risks of stroke in patients diagnosed with AF. However, diagnosis of AF requires long-term ambulatory monitoring procedures that are burdensome and/or expensive.
Smart devices (such as Apple or Fitbit) use light sensors (called "photoplethysmography" or PPG) and motion sensors (called "accelerometers") to continuously record biometric data, including heart rhythm. Smart devices are already widely adopted.
This study seeks to validate an investigational machine-learning software (also called "algorithms") for the long-term monitoring and detection of abnormal cardiac rhythms using biometric data collected from consumer smart devices.
The research team aims to enroll 500 subjects who are being followed after a stroke event of uncertain cause at the Emory Stroke Center. Subjects will undergo standard long-term cardiac monitoring (ECG), using FDA-approved wearable devices fitted with skin electrodes or implantable continuous recorders, and backed by FDA-approved software for abnormal rhythm detection.
Patients will wear a study-provided consumer wrist device at home, for the 30 days of ECG monitoring, 23 hours a day. At the end of the 30 days, the device data will be uploaded to a secure cloud server and will be analyzed offline using proprietary software (called "algorithms") and artificial intelligence strategies. Detection of AF events using the investigational algorithms will be compared to the results from the standard monitoring to assess their reliability. Attention will be paid to recorded motion artifacts that can affect the quality and reliability of recorded signals.
The ultimate aim is to establish that smart devices can potentially be used for monitoring purposes when used with specialized algorithms. Smart devices could offer an affordable alternative to standard-of-care cardiac monitoring.
An estimated 1.6% - 6% of the population over age 65 have undiagnosed and often asymptomatic AF. Oral anticoagulant therapy (OAC) reduces the risks of ischemic stroke by 64% and all-cause mortality by 26% for those diagnosed with AF. Hence, not proactively diagnosing and treating AF will be too great an opportunity to miss. Opportunistic AF screening is endorsed as a cost-effective way of diagnosing AF at primary care facilities and/or pharmacies using various techniques. However, the benefits, costs, and potential harms of more powerful systematic AF screening remain a matter of debate. Continuous AF monitoring is also needed to characterize AF occurrence in terms of its burden and temporal relation to symptoms. On the other hand, technologies for continuous monitoring of AF need excellent acceptability by patients. Well-established ambulatory techniques (e.g., Holter) are not suited because of their poor wearability and short monitoring duration. Techniques of implantable loop recorders have advanced significantly to support AF monitoring. However, only some patients can experience the benefits of these techniques because of their associated high costs and invasiveness. Cutaneous ECG patches are clinically used for AF monitoring, but they last for 2 to 4 weeks and are limited to a selected patient population with approved reimbursement. Consumer-facing solutions exist to provide spot-check ECG with an accuracy on par with that of clinical ECG devices, but they are not continuous and are infeasible for patients with compromised fine motor functions.
In contrast to these techniques, PPG is much better positioned for passive AF monitoring because of its strong physiological premise and the practical consideration that PPG sensors are ubiquitously available in more than 71% of consumer wearable devices. However, because PPG is ubiquitously available on mainstream wearables with companion software capable of generating AF alerts, laypeople can readily use PPG to monitor themselves and take actions without clinician guidance. An untoward consequence of this approach is the potential inappropriate utilization of healthcare resources when following up on false AF detections by potentially millions of users. Unfortunately, algorithms described in 24 published papers have not yet achieved adequate precision that can effectively combat such a risk. For example, many studies reported an accuracy of > 95% but a 5% of error is still too high for a technology that will be used by millions of people to continuously monitor AF in free-living settings.
Inclusion Criteria:
Exclusion Criteria:
xhu40@emory.edu404-712-8520
corey.williams2@emory.edu404-251-4060