Three-dimensional evaluation of postural stability in Parkinson’s disease with mobile technology

Abstract

BACKGROUND:

Postural instability is a hallmark of Parkinson’s disease. Objective metrics to characterize postural stability are necessary for the development of treatment algorithms to aid in the clinical setting.

OBJECTIVE:

The aim of this project was to validate a mobile device platform and resultant three-dimensional balance metric that characterizes postural stability.

METHODS:

A mobile Application was developed, in which biomechanical data from inertial sensors within a mobile device were processed to characterize movement of center of mass in the medial-lateral, anterior-posterior and trunk rotation directions. Twenty-seven individuals with Parkinson’s disease and 27 age-matched controls completed various balance tasks. A postural stability metric quantifying the amplitude (peak-to-peak) of sway acceleration in each movement direction was compared between groups. The peak-to-peak value in each direction for each individual with Parkinson’s disease across all trials was expressed as a normalized value of the control data to identify individuals with severe postural instability, termed Cleveland Clinic-Postural Stability Index.

RESULTS:

In all conditions, the balance metric for peak-to-peak was significantly greater in Parkinson’s disease compared to controls (p < 0.01 for all tests).

CONCLUSIONS:

The balance metric, in conjunction with mobile device sensors, provides a rapid and systematic metric for quantifying postural stability in Parkinson’s disease.

1. Introduction

The underlying mechanism responsible for postural instability in Parkinson’s disease (PD) is not well understood which makes effective treatment challenging (Schoneburg, Mancini, Horak, & Nutt, 2013). In individuals with PD, the severity of postural instability is variable throughout the course of the disease. Postural instability results in falling, (Bloem, van Vugt, & Beckley, 2001) thus escalating costs associated with medical treatment, hospitalizations, and nursing home care (Muslimovic et al., 2008).

Current clinical assessments of postural stability are necessarily relatively rapid to administer and score. However, clinical evaluations, including the Motor Section (III) of the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS-III) and the Berg Balance Scale (Berg, Wood-Dauphinee, Williams, & Maki, 1992; Landers et al., 2008) often lack sensitivity and precision of characterizing specific aspects of sensorimotor impairments that may be critical to understand and aid in the management of postural dysfunction through physical therapy, medication and surgery (Downs, Marquez, & Chiarelli, 2013).

A biomechanically based approach to balance assessment that employs posturography is the NeuroCom Sensory Organization Test (SOT) (NeuroCom International, Clackamas, OR). The SOT manipulates sensory information utilized in effective postural control (i.e. visual, vestibular, and somatosensory) to determine the relative contributions of each input in the maintenance of balance under various conditions (Nashner, 2001). The SOT provides an Equilibrium Score, based on the individual’s maximum anterior-posterior (AP) sway during each balance trial (Nashner & Peters, 1990). However, it does not account for important biomechanical variables of postural stability, such as movement in the medial-lateral (ML) direction as well as rotation about the trunk (TR). Components of balance maintenance that occur in the ML and AP directions (Blaszczyk, Orawiec, Duda-Klodowska, & Opala, 2007; Frenklach, Louie, Koop, & Bronte-Stewart, 2009) and trunk control (Adkin, Bloem, & Allum, 2005) are affected by PD, contributing to postural instability and recurrent falls. Thus, the relatively superficial outcome measure may explain discrepancies across study findings for diagnostic utility, estimating fall-risk, and assessing treatment effects (Nallegowda et al., 2004; Vervoort et al., 2013; Visser, Carpenter, van der Kooij, & Bloem, 2008).

An alternative to using traditional biomechanical approaches is the utilization of inertial sensors consisting of accelerometers and/or gyroscopes. Wearable sensors used to instrument individuals with PD during clinical tests quantify trunk movements in ML, AP, and TR directions (Mancini et al., 2012; O’Sullivan, Blake, Cunningham, Boyle, & Finucane, 2009; Whitney et al., 2011) and have been able to significantly identify differences in postural stability between balance test conditions and populations (Mancini et al., 2012; Ozinga & Alberts, 2014; Ozinga, Machado, Miller Koop, Rosenfeldt, & Alberts, 2015; Whitney et al., 2011). Use of inertial sensors may provide a practical alternative to clinical assessments, force plates, and optical motion capture systems.

Recently, significant correlation in measures of center of mass (COM) acceleration were shown between data collected with the Apple iPad’s (Apple, Inc. Cupertino, CA, USA) built-in accelerometer and gyroscope and three-dimensional motion analysis in a cohort of healthy young adults, healthy older adults, and individuals with PD (Alberts et al., 2015; Ozinga & Alberts, 2014; Ozinga et al., 2015) during a variety of sensory conditions. These studies demonstrate that the data derived from the internal sensor package within a mobile device such as the iPad or iPhone provide biomechanical data of sufficient resolution to accurately characterize postural stability. The primary aim of this study was to develop and validate a three-dimensional measure of postural instability using data collected with a readily available mobile device application that could be used by clinicians to identify postural instability in individuals with PD in situations that mimic common life conditions. To that end, sway acceleration range (P2P) in multiple movement directions (i.e. ML, AP, and TR) during a variety of sensory conditions was investigated to determine if it could differentiate individuals with PD from age-matched healthy controls. In addition, the capability of using this metric to rank the balance performance of individuals with PD in comparison to age-matched control subjects was explored, in order to identify individuals with PD with directional and overall postural instability on a per-trial basis.

2. Methods

2.1. Participants

Fifty-four individuals (27 with PD and 27 healthy, age-matched controls (within three years) served as participants within the Cleveland Clinic Biomechanics Laboratory. Individuals with PD were recruited from the Cleveland Clinic and met the following criteria: diagnosis of PD, age between 30–80, and no known musculoskeletal injury that could impair postural stability. All participants provided consent to procedures approved by the institutional review board.

2.2. Data collection

For the Parkinson’s group, severity of PD was evaluated by the same trained rater using the MDS-UPDRS-III and the Hoehn and Yahr Scale during the test session.

During balance testing, linear and angular accelerations from the iPad’s built-in accelerometer and gyroscope, respectively, were recorded at a 100-Hz sampling frequency with a 3.5-Hz cut-off, 4th order, low-pass Butterworth filter and stored all within the Cleveland Clinic-Balance App written for the iPad. Participants completed a series of postural stability tests while the mobile device was affixed to their waist (level with the sacrum) to approximate whole body COM. The balance tests consisted of 60-second trials under six conditions: 1) double-leg stance, eyes open, firm surface; 2) double-leg stance, eyes closed, firm surface; 3) tandem stance, eyes open, firm surface; 4) double-leg stance, eyes open, foam surface; 5) double-leg stance, eyes closed, foam surface; and 6) tandem stance, eyes open, foam surface.

2.3. Data analysis

The Cleveland Clinic-Postural Stability Index (CC-PSI) metric was computed using peak-to-peak (P2P) sway and a standardized z-score representing the amount of postural sway as a percentile relative to the age-matched control population, in order to: (1) develop a unitless outcome metric; (2) characterize postural instability relative to a healthy control population; and (3) quantify the magnitude of postural instability in each movement direction (ML, AP, and TR). To compute P2P sway, the first and last 1% of the filtered linear and angular acceleration were truncated across movement directions, and the difference between the minimum and maximum values of the remaining intervals was computed. The P2P data were normalized within each unit of measurement and movement direction (ML and AP [m/sec²] and TR [deg/sec²]) by computing the natural-log of all P2P values (lnP2P) to assure the spread of values was normally distributed across balance conditions and to account for the different measurement units. After normalization, a standard normal distribution was calculated for the control population. For the PD group, standard normal distributions were calculated using the mean and standard deviation of the control group across sway directions and balance conditions, to indicate how many standard deviations (STD) a given value was above or below the mean of an age-matched control distribution. The standardized scores were calculated using the z-transformation equation:

$${\text{Z}}_{\text{pd}}=\left({\text{X}}_{\text{pd}}-\overline{x}\right)/s$$ (1)

where s is the STD of the lnP2P of the control sample, $\overline{x}$ is the mean of the lnP2P of the control sample, X_pd is the lnP2P value within a given sway direction for each individual with PD, and Z_pd is the standardized score within a given sway direction to be computed for the individual with PD. This transformation was used in each sway direction, yielding three Z_pd values (z-scores) for each trial. A positive z-score indicated balance performance was worse than the mean of the control group for the specific sway direction. The z-scores were computed to produce a percentile score based on the transformed variable, indicating the rank of each subject’s performance in terms of percentages with respect to the healthy, age-matched control group. To compute percentile scores, the three z-scores were added to get a single value (comp Z_pd; subscript ‘pd’ indicating individuals with PD, ‘c’ indicating control data). Since the three z-scores were normally distributed independent random variables, their sum was normal, but not standardized in relation to the age-matched control distribution; therefore, a standardized CC-PSI was calculated using the z-transformation equation:

$$\text{CC-PSI}=\left({\text{comp\_Z}}_{\text{pd}}-{\overline{x}}_c\right)/s_c$$ (2)

where s_c is the STD of comp_Z_C of the control sample, ${\overline{x}}_c$ is the mean of comp_Z_C of the control sample, comp_Z_pd is the sum of the three Z_pd values of the PD group, and CC-PSI metric is the standardized score for each balance condition for each individual with PD (i.e. CC-PSI₁ indicates the score for condition 1, CC-PSI₂ for condition 2, etc.).

2.4. Statistical analysis

Differences between the PD population and age-matched controls across sway directions and balance conditions were determined using a one sample t-test. Welch correction was applied to account for possible unequal variances. Two sample t-tests (Wilcoxon Rank Sum) were used to determine differences in demographic variables across the groups. Spearman Rank correlation analyses were used to determine the relationship between the PIGD subscores (the sum of MDS-UPDRS-III items 19–21 [gait, freezing of gait, and postural stability]) and the CC-PSI scores for each condition. Differences were assumed significant when P < 0.05. All statistical analyses were performed with SPSS software Version 22.0 (SPSS Inc., Chicago, IL, USA).

3. Results

Gender and age were not significantly different between the control and PD groups (p > 0.05). The PD group’s mean MDS-UPDRS-III score, on medication was 37 (±11.0) and the mean PIGD score was 4 (± 2.3) (Table 1).

Table 1.

Demographics and clinical rating scores with individuals with PD

Subject	Gender	Age	Disease Duration (years)	MDS UPDRS-III	PIGD	BMI (kg/m2)
1	F	69	4.7	30	7	32.3
2	F	69	3.7	32	4	23.6
3	M	57	6.5	32	6	24.7
4	M	70	4.4	40	7	31.1
5	F	61	5.1	22	2	31.0
6	F	55	6.1	22	2	21.6
7	M	68	11.2	37	4	27.1
8	F	62	4.1	45	5	18.5
9	F	51	1.0	23	2	18.3
10	M	76	10.0	63	9	28.3
11	M	72	10.5	35	9	27.7
12	M	59	4.0	36	4	24.1
13	M	76	2.9	54	6	27.8
14	F	41	4.0	23	2	32.1
15	M	50	1.5	37	4	22.3
16	F	66	5.3	38	7	29.0
17	F	63	1.9	38	3	24.0
18	M	60	2.8	41	1	33.5
19	M	66	1.7	27	2	32.0
20	M	53	4.4	43	2	37.0
21	M	79	8.2	48	6	24.9
22	F	55	2.6	48	5	31.0
23	M	60	13.2	60	7	27.3
24	F	59	6.4	25	2	38.6
25	M	72	0.6	28	2	29.7
26	M	72	11.0	40	5	28.0
27	F	57	5.3	35	3	27.9
Mean		62.91	5.30	37.11	4.37	27.90
STD		9.28	3.39	11.00	2.32	4.95

Abbreviations: MDS UPDRS-III, Movement Disorder Society Unified Parkinson’s Disease Rating Scale Part III; PIGD, Postural Instability and Gait Disorder subscore, the sum of Items 19–21 (gait, freezing of gait, and postural stability) in the MDS-UPDRS-III; BMI, Body Mass Index.

Postural instability for the PD group measured by the P2P metric was significantly greater compared to controls across all conditions, 1–6, and in all directions separately (ML, AP, and TR; p < 0.05) and combined (CC-PSI; p < 0.01) (Table 2). Inspection of the mean and confidence interval in each direction revealed that conditions 5 and 6 were most challenging (mean value and lower end of CI are the farthest from zero) in the ML and TR directions, whereas conditions 3 and 6 were most challenging in the AP direction (Table 2). The greatest differences between the PD and control groups were seen in condition 5 in which participants stood in a double-leg stance on foam with their eyes closed and in condition 6 where subjects stood in tandem on foam with their eyes open. The smallest differences were found in conditions 1 and 2, where participants stood in a double-leg stance on a firm surface with their eyes open and closed, respectively. Correlation analysis between PIGD subscores and the CC-PSI showed that across all conditions, except condition 3 in which participants stood in tandem on a firm surface with their eyes open, there was a significant positive relationship (Table 2).

Table 2.

Differences (indicated using the mean, p-value and 95% CI) from the mean of the control population using a one-sample t-test across sway directions and balance conditions and Spearman rank-order correlations between PIGD and the CC-PSI from the PD population across all conditions (last two columns)

Condition	ML			AP			TR			CC-PSI			PIGD vs. CC-PSI
	mean	P	Cl	mean	P	Cl	mean	P	Cl	mean	P	Cl	Spearman ρ	P
1	0.400	0.009	(0.109, 0.690)	0.615	0.030	(0.066, 1.165)	0.339	0.040	(0.037, 1.078)	0.516	0.008	(0.384, 2.323)	0.564	0.001*
2	0.907	0.000	(0.494, 1.320)	1.344	0.000	(0.835, 1.854)	0.727	0.000	(0.445, 0.987)	1.095	0.000	(1.945, 3.991)	0.655	0.000*
3	0.532	0.000	(0.269, 0.796)	1.828	0.000	(1.042, 2.279)	1.495	0.000	(1.026, 1.948)	1.485	0.000	(2.560, 4.799)	0.360	0.082
4	0.980	0.000	(0.625, 1.334)	1.407	0.000	(0.889, 1.926)	1.128	0.000	(0.713, 1.544)	1.289	0.000	(2.399, 4.632)	0.470	0.007*
5	1.762	0.001	(0.825, 2.688)	1.704	0.001	(0.822, 2.586)	1.641	0.000	(1.047, 2.234)	1.890	0.000	(2.762, 7.442)	0.585	0.001*
6	1.042	0.000	(0.787, 1.216)	1.808	0.000	(1.427, 2.166)	1.831	0.000	(1.409, 2.252)	1.621	0.000	(3.696, 5.561)	0.616	0.001*

Bold face indicates the PD population mean is significantly different from the control population mean at P < 0.05. Asterisks indicate a significant positive relationship between PIGD subscores and the CC-PSI scores at P < 0.05. Abbreviations: CI, confidence interval; ML, medial-lateral; AP, anterior-posterior; TR, trunk rotation; CC-PSI, Cleveland Clinic-Postural Stability Index; PIGD, Postural Instability and Gait Disorder subscore.

Individual CC-PSI scores were expressed as a function of the entire cohort of PD participants to identify the number of individuals with PD ranging from superior postural stability to severe postural instability (Fig. 1). The number of individuals with PD with severe postural instability (greater than 1.96 STDs or 95% from the control data mean value) increased as the balance tasks increased in difficulty. In particular, the most challenging balance conditions on a firm surface (condition 3; 41% of individuals with PD) and on a foam surface (conditions 5 and 6; 37% of individuals with PD) had the largest number of individuals with PD with severe postural instability.

Fig. 1.

CC-PSI scores for the entire cohort of PD subjects for all balance conditions. Circles represent each individual with PD that had greater postural stability relative to controls; asterisks represent each individual with PD that was within 1.96 STDs of control group (but less postural stability than the control group); crosses represent each individual with PD that was greater than 1.96 STDs of control group (above dotted line), indicating postural instability was greater than 95% of the control data mean value.

Inspection of the individual directions, using z-scores of P2P acceleration across sway directions within each condition, revealed that in all conditions there were more individuals with PD identified with severe postural instability in at least one direction than what was seen in the CC-PSI for that condition (Fig. 2A). Specifically, a larger group of individuals with PD with severe postural instability, 52%, was identified in condition 6 in the TR direction compared to the CC-PSI metric, which showed a maximum percent of 37%. On average, the direction with the largest number of individuals with PD identified with severe postural instability across all conditions was AP with 38% followed by TR: 27%, and ML: 14%.

Fig. 2.

A) Number of trials in each condition and each direction (ML, AP, and TR along with the CC-PSI) in which postural instability was greater than 95% of the control data mean value. B) Histogram plots for frequency distribution of z-scores representing P2P acceleration during the tandem stance on a firm surface with eyes open (condition 3) for age-matched control subjects are shown. The horizontal axis in each plot represents the z-scores with 0 corresponding to the mean of the control group, and the gray bars represent the trials from the control data. Dotted, dashed, and solid black vertical lines represent ± 1, ±2, and ±3 STD away from the mean, respectively. Diamonds represent standardized scores of individuals with PD with z-scores less than or equal to 1.96 STD away from the mean of the control sample. Circles represent standardized z-scores of individuals with PD with z-scores greater than 1.96 STD and the corresponding number of individuals with PD (N) out of the entire population (N = 27) who fell within this range are displayed above each graph. The asterisks indicate the average z-scores of P2P acceleration within the PD population across sway directions. Z-scores greater than zero correspond to postural instability greater than the mean value of the control group.

Figure 2B displays histograms for the frequency distribution of P2P acceleration (z-scores) while standing in tandem on a firm surface with eyes open (condition 3) for age-matched control subjects in each sway direction and CC-PSI₃. Inspection of each plot shows the ML direction is the most similar between controls and PD groups for this condition, with the PD mean value close to zero and no PD subjects with a z-score greater than 95% of the control population. However, AP and TR directions show a larger number of individuals with PD with postural instability greater than 95% of the control scores (N = 14 and N = 12, respectively) and consequently mean values much greater than zero. The graph of the CC-PSI₃ scores is a combination of all directions and therefore captures the postural instability identified in the ML, AP, and TR directions, with 11 individuals with PD demonstrating postural instability greater than 95% of control scores, which is slightly less severe due to the contribution from the ML direction.

4. Discussion

The aim of this project was to develop and validate a metric that can be used to objectively quantify postural stability across neurological and aging populations incorporating ML, AP and TR directions in a variety of balance conditions using a readily available device with a built-in accelerometer and gyroscope. The primary results of this project indicate: (1) individuals with PD showed significantly greater postural instability relative to the control group across all balance conditions shown by the CC-PSI metric and (2) within the PD group, postural instability increased across sway directions as sensorimotor integration became more challenging (i.e. conditions 5 and 6).

When analyzing CC-PSI scores for P2P for individuals with PD compared to age-matched controls (Fig. 1), it was evident that as test conditions became more difficult, the CC-PSI scores became increasingly sensitive in discriminating between populations. Condition 1 (double-leg stance on a firm surface with eyes open) may not be demanding enough to detect postural instability in individuals with PD compared to age-matched controls, as this condition had the largest number of PD participants with lower CC-PSI scores (indicating greater postural stability) compared to the controls.

As conditions became more difficult (i.e. decreased the reliability of sensory inputs or increased motor control requirements), a greater number of individuals with PD exhibited greater postural instability compared to controls, reflecting an inflexibility to adapt to altered sensory conditions. Neurologically healthy older adults may exhibit sufficient motor control to maintain postural stability with less sway under conditions in which somatosensory or visual feedback is incongruent, thus successfully re-weighting sensory information to maintain stability. This was especially evident in conditions 4–6, in which somatosensory input was dampened with use of a compliant surface.

Results from this study suggest that the CC-PSI metric is able to discriminate between individuals with PD and controls across movement directions and in overall performance. Also, the CC-PSI metric displayed significant concurrent validity relating to PIGD subscores of the MDS-UPDRS-III based on clinical ratings of gait, freezing of gait, and postural stability (Table 2). However, CC-PSI scores were not significantly related with the overall MDS-UPDRS-III Motor Score, suggesting that postural instability in individuals with mild to moderate PD is not necessarily correlated with overall motor symptom (tremor, bradykinesia, and rigidity) severity. Our findings support use of a biomechanical balance assessment that will be particularly useful to track postural stability over time with repeated-measures at the patient’s home or in the office. These features will allow for greater monitoring of disease progression and assessment of treatment effects.

Utilizing z-scores of P2P acceleration across sway directions revealed during the most difficult balance tasks, conditions 3 (tandem stance on a firm surface with eyes open), 5 (double-leg stance on a foam surface with eyes closed) and 6 (tandem stance on a foam surface with eyes open), individuals with PD had significantly greater sway under ankle and hip control. For example, when faced with inaccurate somatosensory information and unavailable vision while in the double-leg stance (condition 5), individuals with PD undergo a hip load/unload strategy in the ML direction and an ankle strategy in the AP direction, and when faced with narrowing the base of support with and without accurate somatosensory information (conditions 3 and 6, respectively) individuals with PD undergo a hip load/unload strategy in the AP direction and an ankle strategy in the ML direction (Winter et al., 1996). By inspection of the confidence intervals across balance conditions (Table 2), condition 5 was most challenging for individuals with PD in the ML direction, and conditions 3 and 6 were most challenging in the AP direction, suggesting a greater impairment using the hip strategy under such conditions. Although healthy older adults frequently employ the hip strategy (Brown et al., 1999), ineffective use of this strategy by individuals with PD is consistent with previous research and may impair automatic postural responses and lead to falls (Baston et al., 2014; Bloem et al., 1996; Horak et al., 1992). Thus, in the PD population, and on a per-subject basis, it is imperative to quantify not only the net movement, but also movement in each sway direction to discern impairments between hip and ankle strategies and to better focus rehabilitation strategies on underlying causes that remain refractory to dopaminergic treatment.

The CC-PSI metric relates to the widely accepted clinically used NeuroCom SOT’s Equilibrium Score for the following reasons: (1) both measures give an overall score measuring P2P of sway throughout a balance trial (although the CC-PSI measures P2P across sway directions whereas the NeuroCom measures P2P only in the AP direction) and (2) clinically abnormal scores are taken as those worse than 95% of the control population (Nashner, 2001). Such performance measures of postural stability are more sensitive than clinical rating scales, while having the ability to discriminate between various populations. Although the PD group’s CC-PSI scores showed significantly greater postural instability compared to the control group across all balance conditions, a greater number of individuals with PD were identified with severe postural instability in at least one sway direction than depicted in the CC-PSI for that condition (Fig. 2A). Therefore, normalizing sway across movement directions separately accounts for additional information to aid clinicians in targeting subtle areas of deficits to improve the method in which individuals with PD perform certain tasks.

By increasing the sensitivity and objectivity to identify subtle postural deficits over time, clinicians and physical therapists can capture key biomechanical aspects in maintaining balance across a wider net of the PD spectrum with a range of balance impairments. There is growing evidence that intensive personalized physical therapy regimens performed under various sensory manipulations and other exercise programs such as tai chi may improve sensory integration and resultant motor control for maintaining postural stability in individuals with PD (Alberts, Linder, Penko, Lowe, & Phillips, 2011; Li et al., 2012). Understanding the biomechanics of patient-specific balance abnormalities would allow such interventions to be implemented in an individualized, prescriptive manner.

Ongoing studies are establishing age-based normative data in a larger cohort of older adults and evaluating additional biomechanical metrics that may be useful in further characterizing postural instability in individuals with PD.