Methods to estimate aspects of physical activity and sedentary behavior from high-frequency wrist accelerometer measurements

Abstract

This investigation developed models to estimate aspects of physical activity and sedentary behavior from three-axis high-frequency wrist-worn accelerometer data. The models were developed and tested on 20 participants (n = 10 males, n = 10 females, mean age = 24.1, mean body mass index = 23.9), who wore an ActiGraph GT3X+ accelerometer on their dominant wrist and an ActiGraph GT3X on the hip while performing a variety of scripted activities. Energy expenditure was concurrently measured by a portable indirect calorimetry system. Those calibration data were then used to develop and assess both machine-learning and simpler models with fewer unknown parameters (linear regression and decision trees) to estimate metabolic equivalent scores (METs) and to classify activity intensity, sedentary time, and locomotion time. The wrist models, applied to 15-s windows, estimated METs [random forest: root mean squared error (rSME) = 1.21 METs, hip: rMSE = 1.67 METs] and activity intensity (random forest: 75% correct, hip: 60% correct) better than a previously developed model that used counts per minute measured at the hip. In a separate set of comparisons, the simpler decision trees classified activity intensity (random forest: 75% correct, tree: 74% correct), sedentary time (random forest: 96% correct, decision tree: 97% correct), and locomotion time (random forest: 99% correct, decision tree: 96% correct) nearly as well or better than the machine-learning approaches. Preliminary investigation of the models' performance on two free-living people suggests that they may work well outside of controlled conditions.

Reporting accelerometer methods in physical activity intervention studies: a systematic review and recommendations for authors

“…This finding is informative, given that machine learning approaches have been recognised for their potential in improving the measurement of physical behaviours since the mid-2000s,30 31 with many recent validation studies showing improvement of energy expenditure prediction compared to traditional methods and the ability to identify PA intensities and types and for activity count and raw accelerometer data 17 20 32–36. However, machine learning is more complex than currently used data analysis techniques, and automated processes are not yet available for machine learning.…”

Objective Assessment of Physical Activity

“…Although complex, the algorithms run smoothly in R (package available https://cran.r-project.org/web/packages/TLBC/index.html) and do not consume much computer power or time. To date, no study has shown whether simpler approaches (29) can deliver similar accuracy levels in totally free living data.…”

“…NHANES, there are numerous large cohort studies with raw data from hip based accelerometers which could benefit from new data processing techniques. Further, it is not yet clear if the wrist location will provide equally accurate assessments as the hip in free living adults (12,29,33). Accelerometer non-wear time was defined as 90 minutes of consecutive zeros (7).…”

Performance of Activity Classification Algorithms in Free-Living Older Adults

“…However, the differences in recognition rates ranged from 2 to 11%, precluding any definitive conclusion as to which placement is superior for classifying activity type in free-living conditions. In 2011, the NHANES adopted the wrist as the standard monitor placement for monitoring physical activity in Americans (11); however, no studies have validated methods to analyze raw acceleration data from the wrist in free-living conditions, only in laboratory (9, 14, 19). Future studies will need to further examine if activity type is accurately classified from wrist accelerometer data in free-living settings.…”