A RIEMANN MANIFOLD MODEL FRAMEWORK FOR LONGITUDINAL CHANGES IN PHYSICAL ACTIVITY PATTERNS.

Area of Science:

• Biostatistics
• Wearable Technology
• Public Health

Background:

• Physical activity (PA) is crucial for health outcomes.
• Wearable activity trackers generate rich data, but traditional analysis loses temporal patterns.
• Novel methods are needed to fully utilize PA data from trackers.

Purpose of the Study:

• To propose a novel functional data analysis approach using Riemann manifolds for modeling PA and its longitudinal changes.
• To analyze PA patterns and their relationship with health outcomes and interventions using real-world data.

Main Methods:

• Modeled daily PA as 1D Riemann manifolds and longitudinal changes as manifold deformations.
• Utilized functional principal component analysis (FPCA) to characterize PA variability.
• Applied the approach to data from two clinical trials (Reach for Health and Metabolism, Exercise and Nutrition at UCSD).

Main Results:

• Identified unique modes of PA changes, including overall enhancement, boosted morning activity, and shifts in active hours.
• Demonstrated the approach's ability to reveal cohort-specific PA patterns.
• Showcased the link between PA changes and weight loss in one trial.

Conclusions:

• The proposed Riemann manifold approach offers a powerful new way to study longitudinal PA changes.
• This method provides deeper insights into PA dynamics than traditional aggregation methods.
• Findings can inform the design of more effective health interventions and public health guidelines.