Validity and Reliability of a Smartphone-Based Gait Assessment in Measuring Temporal Gait Parameters: Challenges and Recommendations

Abstract

Smartphone-embedded inertia sensors are widely available nowadays. We have developed a smartphone application that could assess temporal gait characteristics using the built-in inertia measurement unit with the aim of enabling mass screening for gait abnormality. This study aimed to examine the test-retest reliability and concurrent validity of the smartphone-based gait assessment in assessing temporal gait parameters in level-ground walking. Twenty-six healthy young adults (mean age: 20.8 ± 0.7) were recruited. Participants walked at their comfortable pace on a 10 m pathway repetitively in two walking sessions. Gait data were simultaneously collected by the smartphone application and a VICON system during the walk. Gait events of heel strike and toes off were detected from the sensors signal by a peak detection algorithm. Further gait parameters were calculated and compared between the two systems. Pearson Product-Moment Correlation was used to evaluate the concurrent validity of both systems. Test-retest reliability was examined by the intraclass correlation coefficients (ICCs) between measurements from two sessions scheduled one to four weeks apart. The validity of smartphone-based gait assessment was moderate to excellent for parameters involving only heel strike detection (r = 0.628-0.977), poor to moderate for parameters involving detection of both heel strike and toes off (r = 0.098-0.704), and poor for the proportion of gait phases within a gait cycle. Reliability was good to fair for heel strike-related parameters (ICC = 0.845-0.388), good to moderate for heel strike and toes-off-related parameters (ICC = 0.827-0.582), and moderate to fair for proportional parameters. Validity was adversely affected when toe off was involved in the calculation, when there was an insufficient number of effective steps taken, or when calculating sub-phases with short duration. The use of smartphone-based gait assessment is recommended in calculating step time and stride time, and we suggest collecting no less than 100 steps per leg during clinical application for better validity and reliability.

Keywords: accelerometer; gait; sensor; smartphone; temporal parameters; wearable.

Figure 1

Smartphone placement and orientation with the running custom application interface. While clicking the ‘START’ button, the application will continuously collect the data of acceleration and gyroscope until the ‘STOP’ button is clicked. The application will send a signal to the VICON motion capture system for synchronization upon clicking the ‘SEND MARKER’ button.

Figure 2

An example of the data processing of the smartphone data for gait events. (A) Gait events including the heel contact (heel strike) and toes off were defined based on the reference signal. (B) Left and right were detected based on the reference gyroscope signal. (C) Compare the results of smartphone with the gold standard of the VICON motion capture system. AP, anterio-posterior. SHC, smartphone heel contacts. STO, smartphone toes off. SRHC, smartphone right heel contacts. SLHC, smartphone left heel contacts. SRTO, smartphone right toes off, SLTO, smartphone left toes off, VRHC, VICON right heel contact, VLHC, VICON left heel contact, VRTO, VICON right toes off. VLTO, VICON left toes off.

Figure 3

Gait events and gait parameters identification and definition. The comma superscript represents the gait event of the next gait cycle.

Figure 4

Flow diagram of study subjects.

Figure 5

Bland–Altman plots of different gait parameters between VICON and smartphone. (A) Step Time of Both Legs; (B) Left Step Time; (C) Right Step Time; (D) Stride Time; (E) Duration of Right Stance Phase in a Gait Cycle; (F) Duration of Left Stance Phase in a Gait Cycle; (G) Duration of Double Support I in a Gait Cycle; (H) Duration of Single Support in a Gait Cycle; (I) Duration of Double Support II in a Gait Cycle; (J) Duration of Swing Phase in a Gait Cycle; (K) Proportion of Right Stance Phase in a Gait Cycle; (L) Proportion of Left Stance Phase in a Gait Cycle; (M) Proportion of Double Support I in a Gait Cycle; (N) Proportion of Single Support in a Gait Cycle; (O) Proportion of Double Support II in a Gait Cycle; (P) Proportion of Swing Phase in a Gait Cycle. Note: Differences between VICON and smartphone (smartphone—VICON) are plotted against means between VICON and smartphone for different gait parameters. Dashed lines represent the means of difference between VICON and smartphone while solid lines represent the limits of agreement (mean ± 1.96 SD). Error bars represent 95% confidence intervals of the means of difference between VICON and smartphone.