Validating a New Machine-Learned Accelerometer Algorithm Using Doubly Labeled Water

The purpose of this study is to validate previously developed physical function-clustered specific machine-learned accelerometer algorithms to estimate total daily energy expenditure (TDEE) in individuals with general movement and functional limitations.

Current algorithms for examining accelerometer data were developed primarily using data from individuals without movement limitations or impairments. As such, the current available analytic algorithms are inadequate for use with individuals with limitations and impairments to estimate total daily energy expenditure (TDEE). The creation of a new algorithm that can accurately assess TDEE in individuals with movement limitations will be beneficial for future research examining physical activity interventions targeted to these individuals. This study will serve to validate a new algorithm that was developed specifically to analyze accelerometer data for individuals with movement limitations, and gauge the accuracy of the new algorithm's ability to accurately assess TDEE against one of the gold standards of TDEE measurement, the doubly labelled water technique. Approximately 125 adults, 50 from Colorado, and 75 from Wisconsin, will participate in this study. Participants will complete three study visits. During the first visit, physical function will be assessed during a series of tests, and a dual-energy X-ray absorptiometry (DXA) scan will be performed to obtain information on body composition. During the second visit, the participant will complete a resting metabolic rate (RMR) examination, will consume a dose of doubly labeled water, and will provide urine and saliva samples. At the end of the second visit, participants will be given a set of accelerometers to wear for 8-10 days, and will be asked to complete a wear log for documentation. After 8-10 days have passed, during the final visit, participants will provide additional urine samples and return the accelerometers. The hypothesis being tested is that physical function-clustered specific machine-learned accelerometer algorithms will produce more accurate and precise estimations of TDEE during free-living compared with healthy population derived accelerometer algorithms applied to diverse populations.