Preventice uses real-world data to validate wearable ECG monitoring
May 21, 2109
At this month’s Heart Rhythm conference, Texas-based Preventice presented clinical data to validate wearable technology using machine learning and artificial intelligence for detecting atrial fibrillation.
Held in San Francisco, the conference was the Heart Rhythm Society's 40th annual scientific sessions.
The data validated its BodyGuardian remote monitoring system with the BeatLogic deep learning platform. This technology leverages machine learning and AI for detection of atrial fibrillation (AF) and was validated using clinician adjudicated data.
The BodyGuardian is designed to create a constant connection to monitor cardiovascular data in patients outside the clinic while they go about their daily activities. The data were presented by Hamid Ghanbari from the University of Michigan in Ann Arbor, and Ben Teplitzky and Mike McRoberts from the Preventice data science team.
"One of the exciting advances in the diagnosis of AF is the use of machine learning techniques and deep learning technology because it can allow physicians to manage the massive amount of data that is collected," said Ghanbari, who treats patients who have arrhythmias. "Sensor technologies are creating so much data it's not feasible for physicians to be able to manage and review all of them. With accurate artificial intelligence to identify AF episodes, physicians can focus more on how their patients are feeling and the treatment approach they should take in each case. Artificial intelligence is freeing up the human potential with remote monitoring technologies."
Results from the study demonstrate how the BeatLogic deep learning platform is used to detect accurately the beginning and end of arrhythmias ensuring accurate burden calculations and increasing clinical value. The platform leverages multiple deep neural networks to detect AF episodes at rates that meet or exceed the best-reported values within the literature. Perfect detection performance was achieved for AF episodes lasting more than one minute.
The study evaluated the AF detection performance of the BeatLogic platform using real-world clinician adjudicated data. The platform consists of multiple deep neural networks, which were trained using data from 10,946 BodyGuardian Heart patients. Performance was measured using real-world data from 512 patients that was annotated and then adjudicated by three board certified electrophysiologists. Specific results showed:
- AF duration sensitivity (Se) and a positive predictive value (PPV) were 95.9 and 99.2 per cent, respectively.
- Episode detection Se and PPV were 96.7 per cent.
- Episode detection Se and PPV increased to 100 per cent for AF episodes with duration more than one minute.
"Our focus in advancing Preventice remote monitoring technologies is on how to make these technologies most meaningful to the physician and the patient," said Jon Otterstatter, chief executive officer at Preventice. "Given the large amount of data available through remote monitoring, we're investing heavily in machine learning, designing and validating algorithms. The precise detection of cardiovascular events, like AF, is critical to reaching an accurate diagnosis and treatment."
Ghanbari has received consultation fees from Preventice for advising on remote monitoring technology and services.
BodyGuardian Heart is a small, lightweight, wireless monitor in the BodyGuardian family of monitors. It records physiological data such as heart rate, ECG, respiratory rate and related activity. Through Bluetooth, the smartphone can also capture additional physiological measurements such as blood-oxygen, glucose levels, blood pressure and weight, anytime, anywhere. The system creates a virtual connection between patients and their care teams, allowing physicians to monitor vital signs outside the clinical setting.
The BodyGuardian remote monitoring system includes integration of the BodyGuardian family of monitors and additional third-party sensors, the BodyGuardian Connect smartphone patient application and the PatientCare platform. The system uses machine learning to recognise AF remotely and integrate data into the electronic health record.
Patients wear cardiac monitors, which feed real-time data into the cloud-based health platform that physicians can access. Growing clinical use resulting from increased incidence of cardiac disease and a rising aging population forces a greater reliance on algorithms to provide high-quality and timely reporting. These factors are amplified in the case of mobile cardiac telemetry, where ECG is streamed directly to data processing centres, annotated, and may be used to alert clinicians quickly of potentially critical cardiac events.