We appreciate the opportunity to respond to the thoughtful letter by Van Domelen et al. (5) in relation to our report that empirically developed a new algorithm to classify acceler-ometer wear-nonwear time (1). The intent of our work was to further optimize the algorithm developed by Troiano et al. (4) that has served many groups well, including ours (2), in recent years. We were motivated by the observation that the existing algorithm sometimes classified the accelerometer as not being worn (nonwear), when in fact the participant was observed to be wearing the device. Our study, using 24 h of observation during a stay in a metabolic chamber as an unambiguous indicator of monitor wear, showed that the original algorithm could be improved in this respect by reducing the activity count threshold to end the nonwear period (from 100 to 0 counts per minute), by increasing the time window of allowable zeros from 60 to 90 min, and by eliminating mis-classification of wear-nonwear time between 11:00 p.m. and midnight. We retained and enhanced the component of the original algorithm that excludes artifactual monitor movement. The proposed algorithm, which is modifiable, is now available in R package "PhysicalActiviry" that can be downloaded at http://www.r-project.org/ (3).
展开▼
