Comparing Step Counting Algorithms for High-Resolution Wrist Accelerometry Data in NHANES 2011-2014.

Area of Science:

• Wearable technology and digital health
• Epidemiology and public health
• Biostatistics and data analysis

Background:

• High-resolution accelerometry from wrist-worn devices is increasingly used in free-living studies.
• Step counting algorithms are crucial for analyzing this data but their performance varies.
• Understanding algorithm performance is vital for accurate interpretation in translational research.

Purpose of the Study:

• To compare the performance of five different step counting algorithms.
• To analyze data from free-living, high-resolution wrist accelerometry.
• To assess the implications for translational research and mortality prediction.

Main Methods:

• Applied four open-source and one proprietary algorithm to National Health and Nutrition Examination Survey (NHANES) data (2011-2014).
• Utilized publicly available, high-resolution wrist accelerometry data.
• Compared mean daily step counts using correlation, predictive performance, and mortality hazard ratios.

Main Results:

• Step estimates were highly correlated (median 0.91), with comparable mortality prediction (median concordance 0.72).
• Mean daily step counts varied significantly across algorithms (2453 to 12,169 steps).
• Mortality risk reduction estimates for a 500-step increase differed by up to 300% (HR 0.88-0.96).

Conclusions:

• Step counting algorithms yield correlated estimates and similar mortality prediction, outperforming traditional predictors.
• Significant differences exist in step count distributions and estimated mortality risk reduction among algorithms.
• Algorithm choice critically impacts the interpretation of physical activity and mortality risk in population studies.