Comparison of a portable balance board for measures of persistence in postural sway.

Abstract

Measuring postural sway is important for determining functional ability or risk of falling. Gathering postural sway measures outside of controlled environments is desirable for reaching populations with limited mobility. Previous studies have confirmed the accuracy of the magnitude of postural sway using the Nintendo Wii Balance Board (WBB). However, it is unclear if the WBB can accurately measure persistence of postural sway, i.e., the pattern of center-of-pressure fluctuations over time. The purpose of this study was to compare measures of persistence of postural sway (through detrended fluctuation analysis) using WBB and a force platform (FP). Seventeen healthy individuals performed three standing conditions: eyes open, eyes closed, and one-leg standing. The WBB (30 Hz) was placed on top on the FP (600 Hz) to collect data simultaneously, then the FP data were downsampled to 100 Hz and 30 Hz. The agreement between WBB and FP for measures of postural sway were influenced by the sampling rate and postural sway direction. Intraclass correlation coefficient was excellent (range: 0.953 – 0.998) for long-term scaling regions in the anterior-posterior direction, but lower (range: 0.352 – 0.877) and inconsistent for medial-lateral direction and short-term scaling regions. The three comparison groups (WBB at 30 Hz, FP at 30 Hz, and FP at 100 Hz) showed dissimilar abilities in detecting differences in persistence of postural sway. In summary, the WBB is accurate for quantifying persistence of postural sway measurements in long-term scaling regions in the AP direction, but has limitations for short-term scaling regions and the ML direction.

Introduction

Standing balance measurements can be invaluable in determining functional ability and risk of falling (Roman-Liu, 2018). Variability during postural sway has been shown to characterize healthy and pathological systems (Goldberger et al., 2002; Quatman-Yates et al., 2015; van Emmerik and van Wegen, 2002; Zhou et al., 2017). Specifically, persistence from center of pressure displacement (i.e., how variability of postural control fluctuates over time) is a useful description of postural control in healthy populations (Rand et al. 2015; Rhea et al., 2011) and in the investigation of pathologies such as Parkinson’s disease and developmental disorders (Deffeyes et al., 2009; Minamisawa et al., 2009; Stergiou and Decker, 2011).

Obtaining measurements of postural control outside of a laboratory setting may be helpful to increase data accessibility, but portable laboratory-grade force platforms may be cost prohibitive. Previous studies have shown that a cost-effective portable device such as the Nintendo Wii Balance Board (WBB) obtains similar measures of postural sway compared to a force platform in healthy and pathological populations (Clark et al., 2010; Holmes et al., 2013; Young et al., 2011). Supporting studies examined traditional measures of postural control including center-of-pressure displacement as quantified by root-mean-square or velocity-based measures. However, it is currently unknown whether the WBB can accurately measure persistence in postural sway.

The purpose of this study is to compare the accuracy of measures in persistence of postural sway (quantified by detrended fluctuation analysis [DFA], Peng et al., 1995) of center-of-pressure displacement obtained between a Wii Balance Board and a force platform. We hypothesized that the WBB and the force plate will show high agreement with intraclass correlations greater than 0.9. Lastly, we hypothesized that two devices will show similar ability to detect differences in DFA measures between various standing conditions.

Methods

Participants

Seventeen healthy volunteers were recruited (3 females; mean age 24.5±4.5 years; height 1.75±0.08m; body weight 78.7±10.0kg). Exclusion criteria included self-reported pregnancy, gait or balance abnormality, and below 19 or above 35 years old. The local Institutional Review Board approved the study and all participants provided informed consent.

Equipment

A Wii Balance Board (Nintendo; Kyoto, Japan) was placed centrally upon a flush mounted laboratory grade force platform (AMTI; Watertown, MA, USA). This allowed for simultaneous data collection which has been shown to negate subject-to-subject variability while acquiring highly similar results (Huurnink et al., 2013). Both devices were zeroed and the WBB was calibrated each time by placing a known weight (20.4 kg) on the board and utilizing software by Cooper et al. (2014). The WBB has a plastic composite build with four load sensors on each corner and measures vertical forces. Freely available software was used to interface the WBB via Bluetooth to a laptop and extrapolate data (Cooper et al., 2014). As a previous study demonstrated measures of persistence varies with sampling rate (Rhea, et al., 2011; Rhea et al., 2015), the FP was sampled at 600 Hz and downsampled to 100 Hz and 30 Hz for comparison to the WBB which was sampled at 30 Hz. The frequency 30 Hz was chosen due to similar frequencies (10 – 50 Hz) used in prior studies (Haworth and Stergiou 2018; Huisinga 2012; Kent 2012; Rand 2015; Rand and Mukherjee 2018; Schneipp et al., 2013). A railing was placed around the force platforms for the safety of the participants.

Procedures

Participants conducted standing trials in randomized order for three different conditions: quiet standing on both legs with eyes open (i.e., Eyes Open), quiet standing on both legs with eyes closed (i.e., Eyes Closed) and standing on one leg with eyes open (i.e., One-leg). Each trial lasted three minutes and participants were asked to stand comfortably while maintaining eye contact on an ‘X’ placed at each participant’s preferred eye level. During the One-leg trial, participants were asked to stand on their right leg (regardless of dominance) in a comfortable manner. Outlines of each participant’s feet were traced to ensure consistent foot placement between trials. Trials were marked invalid and repeated if the indicated standing position was violated.

Data Processing and Analysis

Center of pressure (CoP) displacement data for the WBB was extracted from freely available software (Cooper et al., 2014). A custom-written Matlab (MathWorks, Natick, RI, USA) script was created to analyze and compare data between the WBB and the FP. The FP data was downsampled from 600 Hz to 100 Hz and 30 Hz. Data between WBB and FP were time-synchronized using a pre-built MATLAB function (AlignSignals), which utilizes cross-correlation, allowing the lagging or leading signal to be trimmed accordingly. The data were not filtered in any other way. The CoP data were then plotted and compared against each other for accuracy (Figure 1).

Figure 1:

(A) Representative example of center of pressure in AP direction (CoP) plotted across time for force platform (FP) data at 30 Hz (dashed line), 100 Hz (solid line), and WBB (dotted line). The WBB (dotted line) is highly correlated and overlaps significantly with the FP signal, thus a zoomed in section is provided for clarity. (B) Example anterior-posterior detrended fluctuation analysis (DFA) plots. The slopes represent the DFA alpha values for the short-term and long-term regions

Root-mean-square (RMS) was used to quantify the magnitude of postural sway from the CoP displacement. To quantify the persistence of postural sway, detrended fluctuation analysis (DFA) was applied on the CoP displacement (Rand et al., 2015). Persistence of CoP signals have been shown to have multiple scaling regions (i.e. areas of significance at different time lengths). Using the ginput function in Matlab, an inflection point was chosen through a visual inspection of the log/log plot where the slope became less steep (Figure 1B). On average, the inflection point was at a time scale of 1.43s. The persistence of postural sway was assessed on each side of the inflection point for two scaling regions (Collins and De Luca, 1994; Kuznetsov et al., 2013; Rand and Mukherjee, 2018): a ‘short-term’ region (ranging from 0.16 to 0.82s) characterizing CoP fluctuations happening in less than a second, and a ‘long-term’ region (ranging from 2.15 to 11.85s), characterizing CoP fluctuations happening on a longer time scale over a few seconds. The algorithm for DFA was constructed in MATLAB using descriptions in prior studies (Peng, et al. 1995; Damouras, et al. 2010; Mirzayof and Ashkenazy, 2010).

Statistical analysis was done using SPSS software (IBM, Subscription Version). Two-way mixed absolute agreement intraclass correlation, also referred to as ICC(3,1) (Bartko, 1966; Koo and Li 2016) was calculated between the WBB and FP at both sampling rates for RMS and DFA measures. Normal distribution of the data was confirmed with the Shapiro-Wilk test. One-way repeated measures ANOVA was used to compare differences in measures of RMS and DFA across the three standing conditions (Eyes Open, Eyes Closed, and One-leg) for the WBB and FP data (sampled at 30 and 100 Hz). This test was used to examine the ability of each device to detect differences in postural sway measurements among the standing conditions. For statistical differences additional ANOVA analyses were conducted as post hoc test, and a Bonferroni correction was used. Level of statistical significance was set at p < 0.05.

Results

For RMS measures, the averaged intraclass correlation coefficient between WBB and FP was 0.979 for both 30 Hz and 100 Hz (Table 1). For DFA measures, the agreement between the WBB and FP was dependent on the FP sampling rate, scaling region, and the direction of postural sway. The averaged intraclass correlation coefficients were 0.989 and 0.971 for long-term regions in the anterior-posterior (AP) direction at 30 Hz and 100 Hz respectively, but decreased for all other conditions, notably in all of the medial-lateral (ML) and short-term regions (Table 1).

Table 1:

Intraclass correlation using a two-way mixed absolute agreement methods, also known as ICC(3,1) (Koo and Li 2016) were conducted between the Wii Balance Board at 30 Hz and the Force Platform at 30 Hz, and 100 Hz for root-mean-square, and both short-term and long-term scaling regions of detrended fluctuation analysis in the anterior-posterior (AP) and medial-lateral (ML) directions.

N = 17	Intraclass Correlation Coefficient
	Root-Mean-Square		Detrended Fluctuation Analysis Long		Detrended Fluctuation Analysis Short
	Force Platform 30Hz	Force Platform 100Hz	Force Platform 30Hz	Force Platform 100Hz	Force Platform 30Hz	Force Platform 100Hz
Eyes Open	AP: 0.992 ML: 0.991	AP: 0.992 ML: 0.991	AP: 0.995 ML: 0.561	AP: 0.989 ML: 0.364	AP: 0.701 ML: 0.461	AP: 0.660 ML: 0.381
Eyes Closed	AP: 0.977 ML: 0.984	AP: 0.977 ML: 0.984	AP: 0.998 ML: 0.582	AP: 0.971 ML: 0.569	AP: 0.657 ML: 0.375	AP: 0.640 ML: 0.522
One Leg	AP: 0.981 ML: 0.951	AP: 0.981 ML: 0.951	AP: 0.975 ML: 0.352	AP: 0.953 ML: 0.398	AP: 0.517 ML: 0.603	AP: 0.877 ML: 0.468

One-way repeated measures ANOVA showed no statistically significant differences of RMS values at any sampling rate (p>0.132) in both directions. One-way repeated measures ANOVA showed there were significant differences of DFA measures between the WBB and the FP across sampling rates for the short-term region in both directions (AP and ML). Pairwise comparison demonstrated significant differences between both devices and all frequencies. Comparison of the long-term regions showed no statistical differences in the AP direction (p = 0.313) but differences in the ML direction (p =0.005). Comparison of the short-term regions showed differences in both the AP (p<0.001) and ML (p<0.001). Pairwise comparison demonstrated difference between the WBB at 30 Hz compared to the FP at 30 Hz and 100 Hz.

The WBB and FP had differences in their ability to detect significant differences in DFA measures between various standing conditions (Table 2, Figure 2). The FP at 30 Hz detected four significant differences when the FP at 100 Hz did not. The WBB at 30 Hz detected two significant differences when the FP at 100 Hz did not, but the WBB did not detect one difference that the FP at 100 Hz did.

Table 2:

Comparison of detrended fluctuation analysis alpha value mean differences between the Wii Balance Board (WBB) and force platform (FP) at 30Hz and 100Hz across three different standing conditions: eyes open, eyes closed, and one-leg. Significant values are bolded.

N = 17	Long Region DFA Measures Between Standing Conditions
	Anterior-Posterior DFA Mean Difference			Medial-Lateral DFA Mean Difference
	WBB 30 Hz	FP 30 Hz	FP 100 Hz	WBB 30 Hz	FP 30 Hz	FP 100 Hz
Eyes Open vs Eyes Closed	0.128 (p = 0.065)	0.112 (p = 0.102)	0.118 (p = 0.151)	0.073 (p = 0.201)	0.1–39 (p = 0.013)	0.101 (p = .131)
Eyes Open vs One Leg	0.128 (p 0.056)	0.125 (p = 0.035)	0.131 (p = 0.061)	0.106 (p = 0.028)	0.075 (p = 0.783)	0.062 (p = 1.00)
Eyes Closed vs One Leg	0.00007 (p = 1.00)	0.013 (p = 1.00)	0.013 (p = 1.00)	0.033 (p = 0.602)	0.065 (p = 0.609)	0.039 (p = 1.00)
	Short Region DFA Measures Between Standing Conditions
Eyes Open vs Eyes Closed	0.015 (p = 1.00)	0.017 (p = 1.00)	0.004 (p = 1.00)	0.026 (p = 0.157)	0.035 (p = 0.696)	0.004 (p = 1.00)
Eyes Open vs One Leg	0.040 (p 0.407)	0.133 (p = 0.003)	0.018 (p = 1.00)	0.048 (p = 0.132)	0.112 (p < 0.001)	0.062 (p = 0.049)
Eyes Closed vs One Leg	0.056 (p = 0.114)	0.116 (p = 0.004)	0.015 (p = 1.00)	0.075 (p = 0.025)	0.077 (p = 0.143)	0.058 (p = 0.159)

Figure 2:

The Wii Balance Board shows identical ability to detect differences in RMS across three standing conditions (Eyes Open, Eyes Closed, and One-leg) when compared to the force plate at 30 Hz or 100 Hz. However, the DFA-derived measures differed between all three devices in their ability to detect various standing conditions. Brackets and ‘*’ denote significant difference between conditions (p<0.05).

Discussion

The main purpose of this study was to investigate the WBB’s ability to measure persistence in postural control. First, our results are consistent with previous literature with respect to the WBB’s ability to assess magnitude of sway across sampling rates (Clark, et al., 2010). We also found that i) the WBB provides excellent ability to measure persistence in postural sway (measured with DFA) for long-term regions in the AP direction compared to a FP at 30 Hz or 100 Hz, but has less accuracy in the short-term regions and ML directions; and ii) the ability to quantify differences between standing tasks when considering persistence in postural sway differs between a WBB at 30 Hz, a FP at 30 Hz and a FP at 100 Hz.

We found that the estimates of DFA using the WBB were comparable to that of the FP at both sampling rates (30 and 100 Hz) in the long-term regions of the AP direction, but differed greatly in short-term regions and in the ML direction. This may necessitate a visual inspection of the data, in particular where the crossover region occurs which was on average 1.42s, and long-term regions which occurred between 2.15 – 11.85s on average. Differences in alpha values of the scaling regions could be due to the temporal nature of DFA which may be altered using various processing techniques, specifically downsampling (Rhea et al., 2015).

There was no clear pattern of detecting differences between standing conditions (i.e. eyes open vs. eyes closed vs. one leg) when analyzing long-term and short-term scaling regions, different devices and sampling rate. This is an interesting result given that posture data is typically collected between 10 Hz and 50 Hz (Rand et al., 2015). In these cases, it should be noted that previous research suggests temporal structure may be altered when downsampling (Rhea et al., 2015).

When utilizing the WBB for postural sway measurements, several limitations should be considered. There is a general consensus that the Wii is not interchangeable, but rather a comparable alternative for certain circumstances that should be used with caution and consideration (Bonnechère et al. 2016; Clark et al., 2010; Holmes et al., 2013; Pagnacco et al., 2011; Ruff et al., 2015; Young et al., 2011; Young 2014). Specifically, there has been debate about the Wii’s lack of FDA approval, inter-device reliability due to manufacturing standards (i.e. systematic bias and standard error of measure), and overall quality of center of pressure measurements (Clark et al., 2010; Pagnacco et al., 2011; Reed-Jones, 2013; Young et al., 2011; Young 2014).

In summary, the WBB is an accurate tool to quantify measures of postural sway using traditional measures (e.g., RMS of CoP displacement), but measures of persistence (derived via DFA) have notable limitations. Persistence measures from WBB have excellent ICC values (range: 0.953 – 0.998) in the AP direction for long-term scaling regions, mostly moderate ICC values (range: 0.517 – 0.877) for AP direction for short-term scaling regions, and poor to moderate ICC values (range: 0.352 – 0.603) in the ML direction for both long-term and short-term regions (Koo and Li 2016). Given our data, we fail to fully support our hypothesis that the FP and WBB would be able to detect similar differences in standing conditions. Persistence data collected by the WBB is sufficient in the AP long-term scaling regions which can provide useful information; however, caution should be used if information from the short-term region or ML direction is desired.