•  
  •  
 

UTILIZING ACTIVPAL ACCELEROMETERS TO MONITOR WEIGHTED VEST WEAR COMPLIANCE IN THE INVEST RCT: ALGORITHM DEVELOPMENT

KH Alphin

Abstract

Kathryn H. Alphin. Wake Forest University, Winston-Salem, NC.

Background: Despite adverse consequences of obesity, dietary weight loss recommendation remains controversial for older adults due to concomitant reduction in bone mineral density and increased risk of osteoporotic fracture. The INVEST in bone health study (NCT04076618) was designed to determine if external weight replacement via weighted vest use can counter weight loss associated bone loss in older adults. As success of the weighted vest intervention depends on sufficient daily wear time, we sought to develop an accelerometer-based algorithm to objectively monitor vest wear time. Methods: A single user (KHA) wore a HyperVest Pro ® (Hyperwear, Austin TX) weighted vest for three consecutive days, describing periods of vest wear time and recording total minutes of wear. An ActivPAL (PAL Technologies, Glasgow, Scotland) 4 triaxial accelerometer was embedded in a front chest pocket of the weighted vest to capture triaxial acceleration at 20Hz over the same period. Data were downloaded and summed for each axis (X,Y,Z) over 15-second epochs, before being converted to a vector magnitude (VM). Four algorithms were implemented for classifying wear-time using PHP version 7.0. The first three (A1) were modeled on common scoring techniques in physical activity monitoring such that epochs falling below a predetermined threshold (100, 500, or 1000 VM units) were classified as non-wear. The second algorithm (A2) was modeled on wear time scoring in GGIR, which classifies a 15-minute epoch as non-wear if the range and standard deviation of acceleration on two of three axes fell below predefined thresholds. These are computed on a 60-minute window centered on the target 15 minutes. Results: Average daily wear minutes reported from the log were mean±SD (range): 443±68 (364-482) minutes. Average daily wear time minutes from each algorithm were: A1-100 VM: 444±69 (364-486) minutes, A1-500 VM: 376±76 (292-439) minutes, A1-1000 VM: 241±83 (179-335) minutes; A2: 443±60 (375-481) minutes. Conclusion: Preliminary results suggest that A1-100 VM and A2 capture vest wear time to within <1% error relative to detailed self-report. Next, we plan to add more wear time days to provide more data to help validate the algorithm. Finally, we plan to apply both algorithms to the INVEST dataset to compare objective, algorithm-based wear time to traditional, self-reported logs.

This document is currently not available here.

Share

COinS