Identification Of Actigraph Wgt3x-bt Device Non-wear In Infants

Abstract

Traditionally, device non-wear time is determined by examining periods of consecutive zero counts, however, zero counts may also indicate periods of non-movement or sleep. In infants, evaluating non-wear is challenging due to their sporadic nature of movement and sleep frequency. These unique behavior characteristics make a zero counts approach prone to misclassification of non-movement and sleep as non-wear. Thus, an infant-specific method to identify device non-wear time is necessary. PURPOSE: To compare a novel method for identifying device non-wear to consecutive zero counts in infants. METHODS: Fifteen infants (mean±SD; age, 8.7±1.7 wk; 5.1±0.8 kg, 56.2±2.1 cm) wore an ActiGraph wGT3X-BT on the hip and ankle. Criterion data (minutes of wear and non-wear) were collected during two, 2-hour periods of direct observation during which infants spent time in an infant bouncer including sleeping and waking time. A vector magnitude and the inclination angle of each individual axis were calculated from raw 30 Hz acceleration data before being averaged into 1-min epochs. Using the 1-min data, a 4-min rolling coefficient of variation (CV) of each axis was calculated for each minute. Three decision tree models were developed using data from the 1) hip, 2) ankle, and 3) hip and ankle combined. For the consecutive zero counts method, two or more minutes of consecutive zero counts were considered non-wear; this was examined for the hip, ankle, and hip and ankle combined (i.e. if one site indicated “wear” the combined label was “wear”). RESULTS: There were 3,506 total min of observation with 1,987 min of sleep and 1,519 min of waking time with zero criterion non-wear minutes during the observation period. The decision tree approach resulted in lower misclassification of wear as non-wear (5.1-6.0%; 178-212 min) compared to the zero counts method (43.8-51.7%; 1,534-1,813 min). Of the misclassified minutes for the decision tree, 5.3-8.8% (106-175 min) was sleep time compared to 66.8-77.3% (1,328-1,535 min) for the consecutive zero counts method. CONCLUSIONS: Overall, using movement variability (i.e. CV) and device position (i.e. inclination angle), device non-wear can be more robustly identified when worn during periods of non-movement and sleep compared to a consecutive zero counts approach. Supported by NIH P30DK072476-10.