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

Area of Science:

• Biomedical Engineering
• Digital Health Technology
• Cardiovascular Monitoring

Background:

• Wrist-worn photoplethysmography (PPG) sensors are crucial for continuous heart rhythm monitoring.
• Wrist PPG signals are susceptible to motion artifacts, impacting accuracy.
• Machine learning (ML) algorithms can enhance pulse rate (PR) tracking but face regulatory hurdles for clinical use.

Purpose of the Study:

• To evaluate the accuracy of a digital health technology using wrist-worn PPG sensors and an ML algorithm for continuous PR measurement.
• To assess the device's performance across various conditions, including motion and diverse demographic groups.

Main Methods:

• Three independent clinical trials enrolled volunteers monitored concurrently with the investigational device and FDA-cleared electrocardiography (ECG).
• Protocols simulated real-life activities to capture performance under both no-motion and motion conditions.
• Accuracy metrics included root-mean-square (ARMS), bias, mean absolute error (MAE), mean absolute percentage error (MAPE), and concordance correlation coefficients (CCC).

Main Results:

• The device achieved an accuracy of 1.67 bpm under no-motion and 4.39 bpm under motion, meeting clinical acceptance thresholds.
• Bias and limits of agreement (LoA) were within acceptable ranges for both conditions (No-motion: -0.09 bpm, -3.36 to 3.17 bpm; Motion: 0.51 bpm, -8.05 to 9.06 bpm).
• High concordance correlation coefficients (⍴ and CCC >0.98) indicated excellent agreement with ECG reference, demonstrating generalizability across subgroups.

Conclusions:

• The ML-based continuous PR estimation demonstrated clinical-grade accuracy and generalizability.
• The technology effectively mitigates confounding factors from physical motion, health conditions, and demographics.
• This validates the device for populations benefiting from continuous PR monitoring, supporting its clinical utility for retrospective review and trend analysis.