Performance Evaluation of the Verily Numetric Watch Sleep Suite for Digital Sleep Assessment Against In-Lab...

Area of Science:

• Sleep Science
• Wearable Technology
• Biomedical Engineering

Background:

• Accurate sleep monitoring is crucial for health assessment.
• Wearable devices offer a convenient alternative to traditional sleep studies.
• Evaluating the performance of new wearable sleep trackers is essential.

Purpose of the Study:

• To assess the accuracy of 12 sleep measures from the Verily Numetric Watch (VNW).
• To compare VNW sleep data against polysomnography (PSG) in a diverse population.
• To identify potential biases and performance variations across different demographic groups.

Main Methods:

• Simultaneous one-night recordings using VNW and PSG in 41 participants.
• Epoch-by-epoch comparison of sleep/wake and sleep stage classifications.
• Analysis of continuous sleep measures using Bland-Altman plots.
• Assessment of count metrics and subgroup analyses by sex, age, BMI, and skin tone.

Main Results:

• High sensitivity (0.97) for sleep/wake classification; moderate specificity (0.66).
• VNW showed biases in total sleep time, wake after sleep onset, and sleep efficiency.
• Overall accuracy for sleep stage classification was 0.78, with notable variations.
• Proportional biases and heteroscedasticity were observed in most measures.

Conclusions:

• The Verily Numetric Watch demonstrates potential for classifying sleep versus wake and sleep stages.
• While generally consistent, VNW measures exhibit biases and variability, requiring careful interpretation.
• Further research with larger, diverse cohorts is needed to confirm findings and refine algorithms.