Collecting, detecting, and handling non-wear intervals in longitudinal light exposure data

Area of Science:

• Chronobiology and Circadian Rhythms
• Wearable Technology and Biosensing
• Data Science and Signal Processing

Background:

• Wearable light loggers are essential for measuring light exposure in field studies.
• Participants frequently remove devices, creating non-wear intervals of unknown duration.
• Accurate identification of non-wear time is critical for reliable data pre-processing.

Purpose of the Study:

• To deploy and compare multiple strategies for detecting non-wear intervals during light exposure monitoring.
• To establish the ground truth for non-wear detection using a multi-modal approach.
• To evaluate the impact of non-wear detection methods on derived light exposure metrics.

Main Methods:

• Utilized a longitudinal study with 26 healthy participants wearing light loggers for one week.
• Collected non-wear event data via an event marker button, a black bag method, and an app-based Wear log.
• Employed algorithms to detect non-wear based on low illuminance and low activity clusters, comparing performance against self-reported ground truth.

Main Results:

• Self-reported non-wear time constituted 5.4% of total participation time.
• Event marker detection exceeded 85.4% accuracy for intervals >1 minute.
• Illuminance-based detection (F1=0.78) outperformed activity-based detection (F1=0.52) for non-wear intervals.
• Light exposure metrics showed minimal differences between full, self-report filtered, and algorithm-filtered datasets.

Conclusions:

• Systematic collection and detection of non-wear intervals are feasible and vital for robust data.
• While non-wear detection's criticality may vary with cohort compliance, validated methods ensure data integrity.
• The study demonstrates the effectiveness of multi-modal non-wear detection strategies in wearable light exposure research.