Surrogates for rigidity and PIGD MDS-UPDRS subscores using wearable sensors

Abstract

Background:

Telemedicine has the advantage of expanding access to care for patients with Parkinson’s Disease (PD). However, rigidity and postural instability in PD are difficult to measure remotely, and are important measures of functional impairment and fall risk.

Research Question:

Can measures from wearable sensors be used as future surrogates for the MDS-UPDRS rigidity and Postural Instability and Gait Difficulty (PIGD) subscores?

Methods:

Thirty-one individuals with mild to moderate PD wore 3 inertial sensors at home for one week to measure quantity and quality of gait and turning in daily life. Separately, we performed a clinical assessement and balance characterization of postural sway with the same wearable sensors in the laboratory (On medication). We then first performed a traditional correlation analysis between clinical scores and objective measures of gait and balance followed by multivariable linear regression employing a best subset selection strategy.

Results:

The number of walking bouts and turns correlated significantly with the rigidity subscore, while the number of turns, foot pitch angle, and sway area while standing correlated significantly with the PIGD subscore (p<0.05). The multivariable linear regression showed that rigidity subscore was best predicted by the number of walking bouts while the PIGD subscore was best predicted by a combination of number of walking bouts, gait speed, and postural sway.

Significance:

The correlation between objective sensor data and MDS-UPDRS rigidity and PIGD scores paves the way for future larger studies that evaluate use of objective sensor data to supplement remote MDS-UPDRS assessment.

1. Introduction

Telemedicine has the advantage of expanding access to care for patients with Parkinson’s Disease (PD). However, it also has barriers, particularly with regard to ability to assess rigidity and of postural instability aspects of the MDS-revised Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) neurologic examination [1]. These components are particularly important as worsening of these scores predicts an increased risk of falls in PD [2].

There is a pressing need for a means to administer all aspects of the MDS-UDPRS remotely, both for routine neurological care and for clinical trials. A modified UPDRS that excludes rigidity and postural instability has been demonstrated to be generally feasible for telehealth visits in PD [3], [4]. However, secondary to the Covid-19 pandemic, Goetz et al have subsequently emphasized that the MDS-UPDRS Part III can only accommodate the consistent loss of 3 values on any given visit and still allow the total calibrated score to be valid [5]. Therefore, the loss of 5 rigidity scores and a postural instability score during a telehealth evaluation questions the validity of using the standard MDS-UPDRS without supplemental data from a remote setting. Furthermore, the examiner may want to know about rigidity and PIGD scores specifically, and the modified UDPDRS cannot provide that information.

There is growing evidence supporting the use of wearable inertial sensors (that include a tri-axial gyroscope, accelerometer, and magnetometer) to characterize mobility during daily life in older adults and in other neurological populations [6], [7], [7], [9], [10], [11], [12]. Our group has examined the quality and quantity of gait [13], [14] and of turning [15], [16] during daily life activities with sensors placed on the feet and waist [16]. We have reported that quality, but not quantity, of gait and turning differentiates mobility in people with moderate PD from healthy older adults, in keeping with studies from other groups [17], [18]. In addition to characterizing walking and turning, inertial sensors can quantify postural sway as a measure of standing balance [19] which has been shown to be independent from gait [20].

Here, we sought to preliminarily identify and extract variables from inertial sensor data that most associate with rigidity and postural instability and gait difficulty (PIGD) scores of the MDS-UPDRS. Our a priori hypotheses were: 1) the rigidity subscore may be associated with measures of turning in daily life, as turning was associated with measures of axial and neck rigidity [21]; 2) the PIGD subscore may be associated with measures of postural sway while standing, as sway is often increased in people at risks for falls [22].

2. Materials and Methods

2.1. Participants

Thirty-one subjects with PD participated in this study. The diagnosis of idiopathic PD was made by a movement disorders specialist according to the UK Parkinson’s disease Society Brain Bank criteria [23]. Exclusion criteria were dementia, other factors affecting gait such as hip replacement, musculoskeletal disorders, uncorrected vision or vestibular problems, or inability to stand or walk without an assistive device. The protocol was approved by the Ethical Committee of Oregon Health & Science University (eIRB #15578). All the participants provided informed written consent.

2.2. Laboratory data collection:

All subjects were tested in the laboratory 1 hour after their regular dopaminergic medication intake (ON-medication state). PD severity was assessed by a certified researcher using the MDS-UPDRS, Part III [2]. The rigidity subscore in the MDS-UPDRS, Part III reflects a summation of rigidity measures in the neck, bilateral upper limbs and bilateral lower limbs. We used 4 items from MDS-UPDRS Part III to calculate the PIGD score (Part III: 3.9 sit-to-stand, 3.10 gait, 3.12. postural stability, and 3.13 posture) [24]. Subjects wore 3 wireless inertial sensors (weight: 25g, Opals by APDM Wearable Technologies, Portland, OR, USA) on the lower lumbar area, applied with an elastic velcro belt, and on each foot. The sensors on the feet were embedded in instrumented socks (APDM Wearable Technologies prototype). The instrumented socks are lightweight, elastic ankle wraps with velcro. The opal sensor is placed on the top of the foot and the battery is moved up from the sensor to a separate pocket on the ankle within a washable, lightweight ankle wrap. This allowed subjects to comfortably wear the instrumented socks in their shoes or slippers, and without a visually-distracting external strap attachment. Each sensor included triaxial accelerometers, triaxial gyroscopes, and magnetometer recording at 128 Hz. To measure postural sway, only the sensor on the lumbar area was used for analysis during two standing balance tasks of 30 seconds each: 1) feet together with eyes open standing on a firm surface; and 2) feet together with eyes open on a foam surface (Airex, Inc).

2.3. Daily life data collection:

Details are reported in Shah et al. [13]. Briefly, subjects wore the same instrumented socks and Opal on the waist for a week of continuous at-home/normal daily life monitoring of at least 8 hours per day. Data were stored in the internal memory of the Opals. Subjects mailed back the sensors using a pre-paid box after completion of one week of data collection. Data were uploaded offline to an approved secure cloud service upon return of the devices and downloaded to a local computer for further processing.

2.4. Outcome measures:

The gait and turning algorithms used for extracting spatial and temporal measures in the home have been described previously [13], [15], [25]. Briefly, measures related to activity quantity (number of walking bouts, number of turns) were calculated in windows of 30-minute, then averaged across the days [15]. Similarly, quality of gait (gait speed, foot pitch angle at heel strike and variability of both), and quality of turns (turn duration, turn velocity, turn angle, and the variability of each) were calculated by taking an average of all strides across a week. Variability was calculated as a coefficient of variation (CV = standard deviation/mean). In the laboratory, postural sway area was calculated as the area of an ellipse covering 95% of the sway acceleration in both the antero-posterior and medio-lateral planes [26].

2.5. Statistical analysis:

To investigate the association between the MDS-UPDRS III total, rigidity subscore, PIGD subscore, and wearable sensor measures, we used the Spearman correlation coefficient. As exploratory analysis, because various combinations of objective measures can give similar regression results [27], we used multivariable linear regression employing a best subset selection strategy (adopted from [25]). The best subset selection strategy selects the best model from all possible subsets according to goodness-of-fit criteria. For feature selection, we applied best subset selection with leave one out cross-validation. To assess the goodness-of-fit, we used the mean square error (MSE) between the measured and predicted values for a range of predictors and 31 (no. of subjects) models [27]. We selected the number of predictor (feature selection) based on the minimum of MSE averaged across 31 models. After feature selection, we trained and tested a model on selected features using leave one out cross-validation. Finally, a Bland–Altman analysis was conducted. This graphical method analyzes the agreement between two different measurement methods. It uses the mean and the SD to create a scatterplot with limits of agreement, in which the difference between the two paired measurements is plotted against the mean of the two measurements. All statistical analysis was performed using R Version 1.1.456 software with statistical significance set to p< 0.05. The best subset selection method was performed using the package “leaps” with “regsubsets” function in R.

3. Results

3.1. Participant characteristics:

All participants had mild to moderate PD (age mean (SD): 68.9 ± 5.9 years; Female: 12 and Male: 19; MDS-UPDRS part III total tested ON medication: 31.7 ± 9.8), mean MoCA score 27 (Table 1). Levodopa equivalent dose was available for 30 of 31 participants and ranged from 250mg to 3000mg (mean (SD): 1106 ± 668.1mg). The MDS-UPDRS PIGD subscore was 4.7 ± 2.9, and the rigidity subscore was 7.7 ± 2.9. Seventeen of 31 participants reported dyskinesia and 13 of 31 reported freezing of gait with average freezing of gait score 16.3 ± 3.9. Table 1 also lists the number of hours of mobility recording during daily life for each subject. We were able to monitor quality and quantity of gait in home for a range of 5 to 9 days (average of 6.8 days).

Table 1.

Participant Demographics

1	F	67	4	2	29	31	Y	N	675	54
2	F	73	3	2	29	26	N	N	250	24.5
3	F	58	3	2	28	33	N	N	1150	51
4	M	66	18	2	29	33	Y	Y	2162	53.5
5	M	69	10	2	24	36	N	N	1150	55.5
6	M	63	8	2	29	44	N	N	1313.6	43.5
7	F	68	2	2	26	34	Y	N	300	65.5
8	M	71	5	2	30	18	N	N	1327.5	58
9	F	53	2	2	23	24	Y	Y	1200	36
10	M	76	13	2	25	22	Y	Y	950	44.5
11	M	76	8	2	27	47	N	N	775	52.5
12	F	70	16	4	20	47	Y	Y	1000	53
13	M	72	16	2	26	25	Y	Y	1400	74
14	M	76	9	2	25	50	Y	Y	1855	11.5
15	F	68	14	2	29	37	Y	N	1120	39.5
16	M	64	5	2	29	25	Y	N	1430	52.5
17	M	74	6	2	29	48	Y	N	400	43.5
18	F	75	6	3	26	41	N	N	1200	41
19	M	77	-	3	28	35	Y	Y	725	46
20	F	76	5	2	27	23	Y	N	1200	40
21	M	61	7	2	30	18	N	Y	1516.5	63.5
22	M	62	4	2	28	28	N	N	650	51.5
23	M	71	8	2	30	24	N	N	798	49.5
24	F	69	10	2	29	23	N	N	3000	50
25	M	70	8	2	22	26	N	N	1425	53
26	F	63	3	2	26	27	N	N	450	39.5
27	M	71	12	2	22	28	Y	Y	1050	33.5
28	M	76	9	2	25	39	Y	Y	-	16.5
29	M	69	23	2	28	46	Y	Y	1350	29.5
30	M	66	7	2	30	31	N	Y	625	54.5
31	F	66	14	2	29	14	Y	Y	745	57
Mean		68.9	8.6	2.1	27.0	31.7			1106.4	46.4
SD		5.9	5.2	0.4	2.7	9.8			568.2	13.5

H&Y= Hoehn & Yahr Scale; MOCA= Montreal Cognitive Assessment; MDS UPDRS3= Part 3 Motor subsection of the Unified Parkinson’s Disease Rating Scale; Y= yes, N= no; -= missing data point

3.2. Correlations with total MDS-UPDRS, rigidity, and PIGD subscores:

The number of turns, walking bouts, pitch angle of the feet at heel strike, and gait speed significantly correlated with the total MDS-UPDRS III (Fig. 1). The number of walking bouts and number of turns significantly correlated with the rigidity subscore (Fig. 1), whereas the number of turns, number of walking bouts, pitch angle of the feet at heel strike and gait speed, and sway area while balancing on the firm and foam surfaces significantly correlated with the PIGD subscore.

Figure 1.

Radar plot summarizing the correlation between MDS-UPDRS Part III total score (green), PIGD subcores (blue) and Rigidity subscore (Yellow) with objective measures of quality and quantity of gait and balance. The dotted circle indicate significance (p<0.05).

3.3. Multivariable linear regression with best subset selection:

We then used a total of six objective measures (predictors) for the multivariate regression analysis employing the best subset selection strategy. The best subset selection with the minimum MSE for the total MDS-UPDRS Part III total score only included the number of walking bouts during the week. The predicted score with only number of bouts feature with leave one out cross validation was statistically significantly correlated with the clinically assessed total MDS-UPDRS Part III score (r=0.48; p=0.0069, Fig.2) with an average absolute prediction error of 6.5.

Figure 2.

Scatter plot between the measured clinical scores (MDS-UPDRS Part III, PIGD Subscore and Rigidity Subscore) and predicted clinical scores by combining objective measures of gait and balance using the best subset selection method.

The best subset selection with the minimum MSE for the rigidity subscore also only included the number of walking bouts during the week. The predicted score with only number of bouts feature with leave one out cross validation was statistically significantly correlated with the clinically assessed MDS-UPDRS rigidity subscore (r=0.49; p=0.0059, Fig.2) with an average absolute prediction error of 1.9. Finally, the best subset selection with the minimum MSE for the PIGD subscore included the number of walking bouts during the week, gait speed and sway area over a firm surface. The predicted score with these three features with leave one out cross validation was statistically significantly correlated with the clinically assessed total MDS-UPDRS PIGD subscore (r=0.61; p<0.001, Fig.2) with an average absolute prediction error of 1.8. In addition, Fig. 3 in supplementary materials showed Bland-Altman plot for the MDS-UPDRS Part III, Rigidity and PIGD. The plots reflect a good agreement between the measured clinical scores (MDS-UPDRS Part III, PIGD and Rigidity) and the predicted clinical scores (using objective measures of gait and balance).

Discussion

We performed a preliminary analysis of the ability of objective, sensor-based gait and balance measures to approximate MDS-UPDRS Part III, and the MDS-UPDRS rigidity and PIGD subscores. This is the first study to specifically examine rigidity and PIGD MDS-UPDRS subscores in the context of objective gait and balance measures. Our analysis found that the MDS-UPDRS Part III as well as the combined rigidity score were best estimated by the number of gait bouts, whereas the PIGD score was best estimated by a combination of the number of gait bouts, gait speed, and sway area on a firm surface.

Although these findings are exploratory, both bivariate correlation and multiple regression analysis yielded similar results. Contrary to our a priori hypothesis, rigidity was best associated, and can be best predicted, by quantity of gait and turning (number of walking bouts). This result was surprising since previously we found that quality of gait and turning, rather than quantity, differentiated people with PD from healthy control subjects [13], [25]. However, the current study only looked at people with PD. Our current results suggest less overall activity during daily life in people with PD who had higher rigidity and MDS-UPDRS IIII scores. Rigidity, quantitatively assessed during standing, has been shown to be related to difficulty with gait and turning, so people with PD who have high rigidity may limit activities requiring gait and turning [28]. In contrast to rigidity, the PIGD subscore was best predicted by a combination of quantity and quality of gait, as well as postural sway, suggesting a more complex combination of different types of mobility.

We envision combining the remote, virtual MDS-UPDRS Part III with objective measures of gait, turning, and postural sway for a more comprehensive telehealth assessment in the future, creating a “MDS-UPDRS-mobile” score. Our analysis suggests that number of gait bouts during daily life should be explored as a surrogate for rigidity, and that a combination of number of gait bouts, gait speed, and sway area on a firm surface should be explored as a surrogate for PIGD. We want to emphasize that these potential surrogates for rigidity and PIGD are not yet ready for implementation in telehealth or in remote clinical trial administration, but are a starting point for creating data collection algorithms for a partially instrumented MDS-UPDRS-mobile score.

We thus acknowledge important limitations of our current data. First, this study assessed surrogate rigidity measures exclusively in the ON state which may reflect medication-refractory rigidity. We do know that our sensor-based measures are responsive to levodopa from previous studies [29], but further studies exploring responsiveness to medication are needed to establish agreement between clinical scores and the objective surrogates across a large cohort ranging from mild to severe individuals with PD. Second, we performed all mobility analysis by taking the mean of each measure for all the strides over a week for each subject and thus gave equal weight to each stride. But in reality, gait speed and other measures vary for gait bouts of different lengths [30], [31]. Hence, future work will focus on analyzing the effect of bout length on each mobility measure and how gait bout length affects the association with clinical measures. Third, studies will need to validate our findings in a larger population of people with PD through exclusively home-based assessments, as here postural sway data was collected in the lab rather than in the home. Nonetheless, it would be feasible to ask patients with PD to stand briefly on a firm surface in the home for collection of postural sway data, potentially with a home health physical therapist or trained care partner present to monitor safety during standing. Lastly, one of the limitations of capturing data at home and without direct physician monitoring is that during the 8 hours of data collection patients may sleep or engage in activities that do not necessarily involve walking. We asked the patients to use the sensors for at least 8 hours per day and during the time that they are active as we were limited by the battery life of the device. Future studies can separate day and night activity as patients with Parkinson’s disease may wake up several times at night, this will allow for development of separate algorithms for day and nighttime.

Conclusions

This analysis extracted and identified variables from inertial sensor data that most associate with rigidity and PIGD scores of the MDS-UPDRS. This preliminary study shows that objective sensor data for rigidity and PIGD scores may supplement remote MDS-UPDRS assessment in the future. Larger studies on patients in different stages of the disease and examination in both ON and OFF levodopa states as well as in home examination of balance are needed for validation of our findings and further characterization of this correlation with patterns of response to levodopa.

Highlights.

• Rigidity and postural instability in Parkinson’s are difficult to test remotely

• A combination of quantity and quality of balance and gait best relate to the PIGD subscore

• Quantity of gait best relate to the rigidity subscore

• Future, larger trial would need to confirm these preliminary findings

Data Availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.