Save the Date!
ActiGraph Digital Data Summit 2021November 4 - 5 | Sign Up for Event Updates Now
Normalization And Extraction Of Interpretable Metrics From Raw Accelerometry Data
- Presented on June 17, 2013
Introduction Accelerometers provide objective measurements of human activity and have been used extensively in health studies. In many of these studies, analysis was done not based on the raw data, but on summarized metrics like “activity counts”, which are the result of proprietary pre-processing software. Such metrics do not have a clear interpretation and are not comparable between devices or even batches of the same device. Thus, there is a clear and urgent need to introduce data normalization and signal extraction approaches that are transparent and meaningful.
Purpose The goal of this study was to introduce a transparent and explicit normalization procedure of raw accelerometry data and associated visualization tools. We also propose a series of novel metrics for transforming the large amount of information gathered by accelerometers into simple, meaningful, consistent and reproducible measurements.
Methods The tri-axial raw accelerometry data were first processed and each interval of one second was assigned a label, either “active” or “sedentary”. Based on the raw data and the label, 7 metrics were proposed: Wake Time, Time Active, Time Active Mean, Time Active Variability, Cumulative Relative Time Active, Activity Intensity, Activity Intensity Mean, Activity Intensity Variability and Cumulated Relative Activity Intensity. The association between some of these metrics and the subjects’ demographic predictors were studied, using the data from the Baltimore Memory Study (34 subjects, each with 3-5 days of observation).
Results Self Reviewed Health, Quality of Life, Age, and Divorced were found to be significant predictors of both the mean and variability of Time Active and Activity Intensity. Gender was also a statistically significant predictor of those metrics except Activity Intensity Mean.
Conclusions The metrics proposed are valid, transparent and reproducible. They are generated following a clear and interpretable approach. The regression analysis demonstrates the validity of the metrics.
Supported by NIBIB R01EB012547, NINDS R01NS060910