Nonparametric time series summary statistics for high-frequency accelerometry data from individuals with advanced...

Area of Science:

• Health Sciences
• Biostatistics
• Gerontology

Background:

• Accelerometry is crucial for measuring activity and circadian rhythms in health sciences, particularly for individuals with advanced dementia.
• High-frequency accelerometry generates rich datasets, presenting challenges in statistical analysis and optimal parameter selection.

Purpose of the Study:

• To propose and validate novel statistical summary statistics for high-frequency accelerometry data.
• To optimize the implementation and calculation of existing and new metrics for activity and circadian rhythm analysis.

Main Methods:

• Utilized existing metrics: Interdaily Stability (IS), Intradaily Variability (IV), and Detrended Fluctuation Analysis (DFA) scaling exponent.
• Introduced a novel nonparametric estimator, Proportion of Variance (PoV), using spectral density estimation.
• Validated methods on data from 26 individuals with advanced dementia and 14 without, analyzing optimal subsampling rates for IV and proposing daytime/nighttime DFA analysis.

Main Results:

• Recommended an optimal subsampling rate of approximately 5 minutes for calculating IV in the studied dataset.
• Demonstrated the utility of applying the DFA scaling exponent separately to daytime and nighttime data for enhanced individual effect separation.
• Showcased that IS, IV, DFA, and PoV effectively capture distinct features of time-series activity data.

Conclusions:

• The proposed statistical methods and optimizations enhance the analysis of high-frequency accelerometry data for activity and circadian rhythm research.
• These advancements provide more nuanced insights into individual activity patterns, particularly valuable in populations like those with advanced dementia.
• The study offers a framework for optimal data processing and feature extraction from wearable sensor data.