Predicting circadian phase in community-dwelling later-life adults using actigraphy data

Area of Science:

• Chronobiology
• Gerontology
• Wearable Technology

Background:

• Accurate circadian phase estimation has applications in chronotherapeutics, fatigue reduction, and scheduling.
• Algorithms using wearable data can predict laboratory-based circadian phase measurements.
• Older adults may exhibit altered circadian timekeeping, necessitating real-world validation.

Purpose of the Study:

• To validate and extend algorithms for circadian phase prediction in later-life adults (58-86 years).
• To assess model performance using actigraphy data in a home-based setting.
• To compare activity-based models with light-based and sleep-metric predictions.

Main Methods:

• Four mathematical models were used to predict dim light melatonin onset (DLMO).
• Actigraphy data from 58- to 86-year-old adults in a home setting were analyzed.
• Model predictions were compared against ground truth DLMO and validated against light-based models and actigraphy-derived sleep metrics.

Main Results:

• All four models predicted DLMO with mean absolute errors of approximately 1.4 hours or below using actigraphy data.
• Activity-based model simulations performed as well as or better than light-based predictions.
• Higher-order and nonphotic activity-based models showed superior performance compared to actigraphy-derived sleep metrics.

Conclusions:

• Circadian rhythms can be reasonably estimated in later-life adults using mathematical modeling of wearable device data.
• Activity-based models offer a promising approach for real-world circadian phase estimation in older populations.
• This study validates previous findings in a novel cohort and highlights the potential of wearable technology for personalized health applications.