The accuracy of a smartphone to measure laboratory and free-living physical activity

Accelerometers worn on the wrist or hip can be used to measure physical activity (PA) levels in free-living populations. Most modern smartphones also contain an inbuilt accelerometer but the capacity of this technology to accurately measure parameters of PA needs to be further established. 
PURPOSE 
The primary objective of this study was to assess the validity of a popular smartphone to count steps and estimate energy expenditure (EE) during laboratory-based PA. A second objective was to compare free-living daily step counts measurements from the smartphone with a waist-worn accelerometer commonly used in research studies. 

METHODS 
Healthy adults (n=20, 28 ± 5 yrs) took part in a single laboratory trial and a free-living trial (n=16, 42 ± 17 yrs). Participants wore the smartphone and accelerometer in a waist-mounted pouch continuously during both trials. Laboratory trials comprised 5 min bouts of treadmill walking and jogging. Step counts were manually counted (MC) and EE was measured using indirect calorimetry (IC). The estimates of PA parameters from the smartphone and accelerometer were compared to each other and to the gold standard measures (MC and IC) using the concordance correlation coefficient (CCC) with the thresholds: almost perfect >0.90; substantial >0.8 – 0.9; moderate 0.65 – 0.8; poor <0.65. Levels of agreement are expressed as mean bias with 95% limits of agreement (LOA). 

RESULTS 
Compared to MC (700 ± 98 steps), the smartphone (703 ± 97 steps; CCC 0.992; mean bias 3 steps, LOA –19 to 25 steps) and accelerometer (675 ± 133 steps; CCC 0.76; mean bias –25 steps, LOA –179 to 129 steps) provided accurate measurements of step count. Compared to IC (8 ± 3 kcal·min−1), the smartphone (6 ± 1 kcal·min−1) underestimated EE with poor agreement between methods (CCC = 0.48; mean bias –1.9 kcal·min−1, LOA –5.6 to 1.8 kcal·min−1). During free-living, the smartphone (7990 ± 4673 steps·day-1) substantially underestimated step count compared to the accelerometer (9085 ± 4647 steps·day-1; mean bias –1095 steps·day-1, LOA –4780 to 2591 steps·day-1). 

CONCLUSION 
The smartphone was able to provide accurate measurements of step count during a controlled laboratory walking trial but substantially underestimates PA in comparison to an accelerometer during a period of free living.