Performance Evaluation of Device-Based Algorithms to Estimate Step Counts in Free-Living Adults Compared with Direct...

Area of Science:

• Physical activity measurement
• Wearable sensor technology
• Health research methodology

Background:

• Step counts are a key metric for physical activity and health research.
• Limited evidence exists on the accuracy of step counting in free-living adults.

Purpose of the Study:

• To evaluate the accuracy and precision of six step counting methods.
• Comparison of research-grade devices against video-recorded direct observation.

Main Methods:

• Twenty adults wore thigh-worn activPAL and wrist-worn ActiGraph GT3X+ devices for seven days.
• Five algorithms (ActiLife, Oak, Step Detection Threshold, Verisense, stepcount) were applied to ActiGraph data.
• Participants underwent video recording during specific sessions for comparison with device data.

Main Results:

• The wrist algorithm 'stepcount' had the lowest mean absolute percent error (17.1%).
• The activPAL and 'stepcount' algorithm were statistically equivalent to direct observation (15% level).
• Accuracy was highest during walking/running and decreased with varied activities like biking or pushing a stroller.

Conclusions:

• Step count accuracy varies significantly among different algorithms and devices.
• Thigh-worn activPAL and wrist-worn 'stepcount' algorithm demonstrated the highest accuracy.
• These findings are crucial for selecting appropriate devices in physical activity research.