Multicenter Evaluation of Machine-Learning Continuous Pulse Rate Algorithm on Wrist-Worn Device

Abstract

Listen

Translate article

Plain Language SummaryMonitoring heart rate continuously and unobtrusively is vital for managing various health conditions and for advancing clinical research. Traditional methods for tracking heart rate, like electrocardiography monitors, are often bulky and inconvenient for continuous use. Optical sensors embedded in wristbands conveniently measure PR, a proxy of heart rate. However, the accuracy of these devices for clinical care or research purposes has been poorly established to date, with limitations including a lack of data on accuracy in motion conditions and across demographic factors such as skin tone, as well as a lack of established methodological and accuracy standards from medical device regulators. This validation study aimed to test the accuracy of a new wrist-worn medical device that analyzes optical sensor data with a ML algorithm to continuously monitor PR. The study involved 157 participants who were monitored with the wristband and an electrocardiography monitor simultaneously during different activities, such as resting, walking, and activities of daily living. To make sure that the ML algorithm was equally accurate in different patient populations, data were collected from a range of participants (three independent sites, varying demographic and clinical traits), and accuracy was evaluated using multiple performance indicators and subgroup analyses. Results demonstrated clinical-grade accuracy and generalizability of the device, paving the way for device usage by populations most likely to benefit from continuous PR monitoring. This study significantly advances efforts to transparently report on the methodologies and performance of digital health technologies for PR monitoring.