A method to estimate free-living active and sedentary behavior from an accelerometer.

Area of Science:

• Wearable sensor technology
• Machine learning in health
• Physical activity and sedentary behavior research

Background:

• Validating physical activity (PA) and sedentary behavior (SB) estimation from wearable monitors in free-living conditions is crucial.
• Existing methods often lack accuracy when applied outside controlled laboratory settings.

Purpose of the Study:

• To develop and validate two novel machine-learning algorithms, Sojourn-1 Axis (soj-1x) and Sojourn-3 Axis (soj-3x), for estimating PA and SB.
• To assess the performance of these new algorithms against a previously laboratory-calibrated neural network (lab-nnet) in a free-living environment.

Main Methods:

• Participants underwent direct observation for 10 hours across three separate occasions in their natural environment.
• PA and SB estimates from soj-1x, soj-3x, and lab-nnet were compared against direct observation data.

Main Results:

• Soj-1x and soj-3x demonstrated significantly improved accuracy in estimating MET-hours and time spent in different intensity categories (sedentary, light, moderate-to-vigorous PA [MVPA]) compared to lab-nnet.
• Both soj-1x and soj-3x provided accurate estimates of guideline-recommended minutes and breaks from sedentary time.
• Soj-3x showed superiority over soj-1x in distinguishing sedentary behavior from light-intensity activity.

Conclusions:

• The soj-1x and soj-3x algorithms offer enhanced accuracy and precision for estimating free-living MET-hours, SB, light-intensity activity, and MVPA.
• These novel methods represent a significant advancement in validating wearable sensor data for PA and SB monitoring in real-world settings.
• Soj-3x provides a more refined differentiation between sedentary and light-intensity behaviors than soj-1x.