Validation of smartphone step count algorithm used in STARFISH smartphone application

Abstract

BACKGROUND:

Smartphone sensors are underutilised in rehabilitation.

OBJECTIVE:

To validate the step count algorithm used in the STARFISH smartphone application.

METHODS:

Twenty-two healthy adults (8 male, 14 female) walked on a treadmill for 5 minutes at 0.44, 0.67, 0.90 and 1.33 m $\cdot$ s ${}^{-1}$ . Each wore an activPAL ${}^{\text{TM}}$ and four Samsung Galaxy S3 ${}^{\text{TM}}$ smartphones, with the STARFISH application running, in: 1) a belt carrycase, 2) a trouser or skirt pocket), 3a) a handbag on shoulder for females or 3b) shirt pocket for males and 4) an upper arm strap.

Step counts of the STARFISH application and the activPAL ${}^{\text{TM}}$ were compared at corresponding speeds and Bland-Altman statistics used to assess level of agreement (LOA).

RESULTS:

The LOA between the STARFISH application and activPAL ${}^{\text{TM}}$ varied across the four speeds and positions, but improved as speed increased. The LOA ranged from 105–177% at 0.44 m $\cdot$ s ${}^{-1}$ ; 50–98% at 0.67 m $\cdot$ s ${}^{-1}$ ; 19–67% at 0.9 m $\cdot$ s ${}^{-1}$ and 8–53% at 1.33 m $\cdot$ s ${}^{-1}$ . The best LOAs were at 1.33 m $\cdot$ s ${}^{-1}$ in the shirt pocket (8%) and upper arm strap (12%) positions.

CONCLUSIONS:

Step counts measured by the STARFISH smartphone application are valid in most body positions especially at walking speeds of 0.9 m $\cdot$ s ${}^{-1}$ and above.