Modelling a response as a function of high-frequency count data: The association between physical activity and fat...

Area of Science:

• Biostatistics
• Physical Activity Epidemiology
• Wearable Technology in Health

Background:

• Accelerometers are crucial for quantifying physical activity in health sciences.
• Current methods use lossy summaries with implicit assumptions about physical activity effects.
• There is a need for more accurate and flexible analysis of high-frequency accelerometer time series.

Purpose of the Study:

• To propose a novel functional summary using histograms for accelerometer data.
• To develop a statistical method for near-lossless dimension reduction and improved interpretability.
• To investigate the relationship between physical activity and fat mass using this new approach.

Main Methods:

• Utilizing histograms as functional summaries for high-frequency accelerometer counts.
• Employing regression on additive multi-dimensional functional predictors.
• Handling time series of varying lengths and missing values with fewer assumptions.

Main Results:

• The histogram summary provides near-lossless dimension reduction and enhances comparability.
• The novel method offers a flexible way to model physical activity effects.
• Analysis of Avon Longitudinal Study of Parents and Children data reveals non-linear and non-constant effects of physical activity on fat mass.

Conclusions:

• The histogram-based functional summary is a powerful tool for accelerometer data analysis.
• This method improves upon existing techniques by reducing data loss and accommodating complex relationships.
• Physical activity's effect on fat mass is complex, varying by intensity and not strictly linear or constant.