PURPOSE: Both hip/thigh and wrist-worn accelerometers are used to estimate moderate-vigorous intensity physical activity (MVPA). Few studies have directly compared these MVPA estimates in free-living environments across distinct activity domains. The purpose of this study was to compare AG wrist estimates of MVPA to AG-hip and the thigh-worn AP in five activity domains. METHODS: Fifteen adults (10F and 5M; age 18-36y) participated in two, 2-hr sessions that were categorized by activity domains; household (H, N=5), active leisure (AL, N=8), sedentary leisure (SL, N=6), work (W, N=7), and transportation/errands (TE, N=4). Participants were given general instructions (e.g., at least 45 min spent on household- related behaviors), but sessions took place in the participants’ natural environment. During the sessions, participants wore AG (non-dominant wrist, right hip) and AP on the right thigh. MVPA was estimated for AG-hip using machine learning (S3x), Freedson (F), Crouter (C2), and Sasaki (VM) methods. The AG-wrist data was processed using a random forest (RF). Pearson correlations and paired t-tests were used to compare MVPA estimates across methods. Linear mixed effects models were used to test if there was a significant difference in MVPA estimates between wrist and hip methods across activity domains. P-values <0.05 were considered statistically significant. RESULTS: Correlation between wrist RF and the hip methods ranged from R=0.63-0.66, while hip/thigh methods were all highly correlated (R=0.94-0.99). The AL domain had the highest time spent in MVPA (between method range: 50.7-60.6 min), while SL (range: 1.3-15.8 min), and W (range: 2.0-23.5 min) were the lowest. Wrist estimates of MVPA were significantly higher than the S3x (+10.8 min), F (+15.3 min), AP (+12.7 min), and VM (+12.1 min), all p<0.01. The estimates of MVPA from the RF were not significantly different than C2, p=0.11. Agreement between methods did not differ by activity domain (p>0.05). CONCLUSIONS: The wrist RF method was moderately correlated with hip/thigh measures and consistently produced higher estimates of MVPA compared to hip/thigh algorithms, across all activity types. Future research using direct observation as a criterion measures is needed. Supported by Bill and Linda Frost Fund.
Read full abstract