Bayesian Covariate-Dependent Circadian Modeling of Rest-Activity Rhythms

Area of Science:

• Biostatistics
• Chronobiology
• Wearable Technology

Background:

• Rest-activity rhythms are crucial for health.
• Analyzing circadian data from wearables presents challenges.
• Existing models may lack flexibility in incorporating individual factors.

Purpose of the Study:

• To develop a flexible Bayesian model for analyzing wearable-derived circadian activity data.
• To integrate covariate effects on amplitude and phase of circadian rhythms.
• To promote model sparsity and identify significant predictors using an l1-ball projection prior.

Main Methods:

• Proposed a Bayesian covariate-dependent anti-logistic circadian model.
• Integrated covariates into amplitude and phase parameter modeling.
• Employed an l1-ball projection prior for model sparsity.
• Validated the model using simulated and real-world actigraphy data from epilepsy patients.

Main Results:

• The model effectively analyzes circadian activity data from wearable devices.
• Demonstrated the ability to uncover complex relationships influencing rest-activity rhythms.
• Identified significant demographic, psychological, and medical factors affecting circadian patterns.
• Showcased enhanced flexibility and interpretability in cohort-level analysis.

Conclusions:

• The proposed Bayesian model offers a powerful tool for analyzing wearable-based circadian data.
• Provides valuable insights into factors affecting rest-activity rhythms, particularly in clinical populations like epilepsy.
• Facilitates personalized clinical assessments and targeted healthcare interventions based on individual circadian profiles.