A Study of Turn Bias in People with Idiopathic Parkinson’s Disease

Abstract

Objective:

To explore whether people with Parkinson’s disease (PwPD) display a preferential turn bias dependent upon disease asymmetry, and whether specific disease features predict turn bias.

Methods:

PwPD and age-matched controls were instructed to walk on an instrumented gait mat making “normal” turns. Trials were analyzed using ProtoKinetics Movement Analysis Software (PKMAS) and time-locked video recordings to obtain turn directionality and spatiotemporal turn measures. Turn bias was estimated using previously defined formulas.

Results:

Seventy-two PwPD and 28 controls were included. One hundred percent of controls and 85% of PwPD had left turn bias. Turn bias was not significantly associated with age, gender, handedness, disease asymmetry, cognition, or disease severity. The Freezing of Gait Questionnaire (FOGQ) questions 5 and 6 showed linear-by-linear association with turn bias. In binary logistic and ordinal regression models, FOGQ question 6 (average duration of turn freezing) and turn width were predictive of turn bias. Rightward turns had greater frequency of freezing episodes.

Conclusions:

Turn bias in our PwPD cohort does not appear related to disease asymmetry or other disease features, except gait freezing. Whether freezing severity on turning leads to non-left turn bias or vice versa requires more focused studies.

Significance:

Physical therapy interventions targeting turning direction in PwPD could reduce freezing severity.

1. INTRODUCTION

Turning behaviors are an important component of daily walking and can comprise up to 50% of the time spent in activities such as walking around a cafeteria (Glaister et al. 2007). In the elderly, falls during turning were 8 times more likely to lead to hip fractures than falls during straight walking (Cumming and Klineberg 1994). In people with Parkinson’s disease (PwPD), turning can be especially affected. As opposed to a sequenced cranio-caudal order of turning (Hong et al. 2009), PwPD turn en-bloc, initiating head, trunk and pelvis rotation simultaneously. Freezing of gait (FOG), in which some PwPD have the feeling of their feet sticking to the ground(Fahn 1995), also commonly occurs during turning (Giladi et al. 1992; Schaafsma et al. 2003), and can lead to falls (Stack et al. 2006), and decreased quality of life (Rahman et al. 2008).

Turning is an inherently asymmetric task. In rodents, rotational or circling behavior is well established as a sign of dopamine asymmetry (Zimmerberg et al. 1974; Loscher 2010). Results in healthy human turning behavior are less clear. Leftward (Lenoir et al. 2006; Taylor et al. 2007; Toussaint and Fagard 2008) and rightward (Mead and Hampson 1996) turning bias have been reported. Turning direction has been linked to sex differences (Bracha et al. 1987a; Mead and Hampson 1996) including hormonal regulation (Mead and Hampson 1996), handedness (Bracha et al. 1987a; Mohr et al. 2003a; Taylor et al. 2007) and scores on tests of visuospatial function (Gordon et al. 1992). Asymmetry in the basal ganglia in healthy individuals has also been suggested to play a role in turning directionality as a larger left globus pallidus size (Kooistra and Heilman 1988) and higher left globus pallidus dopamine levels (Glick et al. 1982) have been reported. Additionally, in blindfolded healthy individuals given levodopa, right veering behavior during straight line walking was reduced compared to a placebo group (Mohr et al. 2003b). In a more recent study utilizing a virtual reality version of the Morris Water Maze task, Yuan et al. found that older age was associated with greater rightward turning preference, men and women showed opposite turning preference (leftward in men, rightward in women), and participants turned preferentially towards the hemisphere with larger putamen and cerebellum (Yuan et al. 2014).

A number of studies have explored spatiotemporal dynamics of turning in PD although the effect of turning bias has not been explored (for review see (Hulbert et al. 2015)). PwPD have been shown to take a greater number of steps when turning (El-Gohary et al. 2013), have a reduction in step length (Huxham et al. 2008). PwPD have been reported to turn more on-the spot, especially when turning in their non-preferred direction (Stack and Ashburn 2008) as opposed to pivoting or advancing towards the target while turning. These deficits appear more pronounced in those with FOG compared to those without FOG (Bhatt et al. 2013; Bengevoord et al. 2016; Mancini et al. 2018) (for review see (Spildooren et al. 2018)). We previously reported that the length of foot-strike to the ground during turning was more variable in those with FOG compared to those without FOG (Pillai et al. 2022). Additionally, when distracted during dual-tasking, thereby reducing volitional cognitive control of gait, there are greater deficits in turning (Spildooren et al. 2010).

In PD where the asymmetric loss in dopamine leads to the clinical disease affecting one side of the body more than the other (Hoehn and Yahr 1967; Fahn and Elton 1987; Varrone et al. 2001), one would expect turning directionality to be affected. In support of this idea, a report on 5 participants with Hemi-parkinsonism, suggested that the majority turned towards the side with less dopaminergic activity (Bracha et al. 1987b). Since turn directionality in an individual could be modifiable, our goals were to explore whether PwPD have a preferential bias to turning leftward or rightward and if turn bias was present, to determine the disease or gait features that predicted the turn bias.

2. METHODS

Participants with a diagnosis of idiopathic Parkinson’s disease (PD) based on UK brain bank criteria were recruited from the University of Arkansas for Medical Sciences (UAMS) Movement Disorders clinic. Spouses or family members who did not have Parkinson’s disease were asked to participate as relatively age-matched controls, in order to determine if normal aging itself played a role in any preferential turn bias that may have been uncovered in this study. The project was approved by the institutional review board (UAMS IRB# 203234), performed in compliance with the Declaration of Helsinki and written informed consent was obtained from each participant before study assessments were performed. Exclusion criteria included a Montreal Cognitive Assessment Score (MoCA) score of <10, falls>1/day and prior antidopaminergic medication use in the year prior to enrollment.

2.1. Clinical and cognitive assessments

PD participants performed all assessments in their usual medicated (or levodopa ON-state) without any change to their dose routine. Most recent dose and time information was recorded and daily equivalent levodopa dose was calculated using previously reported equivalencies (Schade et al. 2020). To assess disease severity, participants received a complete Unified Parkinson’s disease Rating Scale (UPDRS) (Fahn and Elton 1987) assessment and assigned a Hoehn and Yahr (H&Y) staging score (Hoehn and Yahr 1967) by a movement disorders fellowship trained neurologist (TV). As PD is asymmetric disease, the disease more affected side was calculated based on the ratio of the summated right/left UPDRS item scores with individual measures for each body side (tremor, rigidity, bradykinesia), with a score >1 indicating right more affected side. Turning is a common cause of falls in the elderly (ref) and therefore fall severity was judged based on the UPDRS Item 13 score (falls independent of freezing).

The Giladi Freezing of Gait Questionnaire (Giladi et al. 2000) is a six-item questionnaire of which questions (Q) 1 and 2 assess overall walking abilities from normal walking to inability to walk (Q1) to the impact of walking difficulties on activities of daily living (Q2). Questions 3–6 are related specifically to freezing and measure frequency (Q3) and duration (Q4–6) of the longest perceived freeze (Q4) and average freezing on gait initiation (Q5) and turning (Q6). In our analysis, we used the FOGQ total score, individual question scores, and a freezing severity sub-score (sum of Q3–6 only) to determine whether turn bias was predicted by these features of gait and freezing severity. These items have been individually validated against items of the UPDRS during development of the FOG-Q and subsequently (Giladi et al. 2000; Giladi et al. 2009).

Participants were also administered previously validated cognitive screening tests including the Montreal Cognitive Assessment (MoCA)(Nasreddine et al. 2005), the Frontal Assessment Battery (FAB)(Slachevsky et al. 2004), and the Scales for Outcome in Parkinson’s Disease - Cognition (SCOPA-COG)(Marinus et al. 2003).

2.2. Gait assessments

A 20 foot x4 foot Zeno Walkway (Protokinetics LLC, Haverton, PA), two video cameras, and ProtoKinetics Movement Analysis Software (PKMAS)(Egerton et al. 2014; Lynall et al. 2017; Virmani et al. 2018) were synchronized to capture data. PKMAS captures pressure sensor data at a frequency of 120 Hz (120 cycles/second) samples/second, and video was captured at a rate of 30 frames/second. The instrumented gait mat (Zeno walkway) was placed in a room with adequate distance to walk around on all four sides, with the longer length closer to one wall. Two small traffic cones placed approximately one foot from both ends of the mat served as visual cues to avoid off-mat turns. Trained test administrators used pre-defined vocabulary to instruct participants to turn at or before the cones while also demonstrating the task. Participants walked at least eight lengths of the mat at their comfortable pace making “normal” turns at or before the cones, in order to obtain at least seven turns to analyze per participant. As turns were performed on both ends of the mat, proximity to a specific wall forcing turning in one direction was accounted for, as a leftward turn would be away from the closer wall in one direction and towards the closer wall in the other direction. If it appeared during the walking trial that participants may have stepped off the mat, or were talking or otherwise distracted during a turn, they were requested to perform additional walking lengths and turns.

2.3. Spatiotemporal measurements of turns

Footprint processing and video reviews were done within the PKMAS environment. Footfalls, either not identified or incorrectly assigned by PKMAS, were edited manually. Captured data included straight walk and turn segments and for the purposes of this analysis, we deleted straight walks and retained only the turn segments as previously described (Pillai et al. 2022). Turn segments were defined as the first contact of the last normally angled footstep (pre-turn) and the last contact of the first normally angled footstep (post-turn)(Pillai et al. 2022). PKMAS has inbuilt algorithms that automatically calculate standard gait measures once foot steps are correctly identified as being the right or left foot. These measures are shown in Table 1 (source: PKMAS). We also used the foot location (X and Y coordinates) on the mat and foot contact times provided for each footstep by the PKMAS software to calculate and calculated the mean and percent coefficient of variation (CV) of spatiotemporal measures of turning (Table 1, author defined) as we have also previously reported (Pillai et. al. 2022). Directionality of each turn was determined using the time-locked video recordings.

Table 1: –

Spatiotemporal gait measures – sources and definitions

Measure	Source	Definition
Turn time (s)	author defined	The time (in seconds) taken to complete each turn. Calculated as the difference in time# between the first contact of the last normal-angled, pre-turn footstep, and the last contact of the first normal-angled, post-turn footstep.
Turn length (cm)	author defined	The maximum longitudinal distance (in centimeters) taken for turning along the length of the 20’x4’ mat. Calculated as the distance (difference) between the maximum and minimum value of the X coordinates# of foot placement on the mat.
Turn width (cm)	author defined	The maximum horizontal distance (in centimeters) taken for turning along the width of the 20’x4’ mat. Calculated as the distance (difference) between the maximum and minimum value of the Y coordinates# of foot placement on the mat.
Turn area (cm2)	author defined	turn length * turn width
Step count (no)	author defined	Total number of steps taken to complete a turn segment. Calculated as the sum of right and left steps in each turn segment.
Integrated Pressure (p × s)	PKMAS	The aggregate of pressure applied by a footstep in its contact area at each sampling time (120 Hz sampling rate).
Foot length (cm)	PKMAS	The length (in centimeters) of the major axis of the ellipse created by PKMAS to enclose each footstep.
Foot width (cm)	PKMAS	The length (in centimeters) of the minor axis of the ellipse created by PKMAS to enclose each footstep.
Stride length (cm)	PKMAS	The distance (in centimeters) between heel strikes of two consecutive footsteps of the same foot.
Stride time (s), also known as gait cycle time	PKMAS	The difference in time (in seconds) between the initial heel contacts with the mat of two consecutive footsteps of the same foot.
Stride velocity (cm/s)	PKMAS	The stride length divided by stride time, calculated for each gait cycle.
Stance percent	PKMAS	A percentage measure of time spent in stance phase of the gait cycle and calculated as stance time/gait cycle time.
Swing percent	PKMAS	A percentage measure of time spent in swing phase of the gait cycle and calculated as swing time/gait cycle time.

Source column: All “PKMAS” calculations were auto-generated by the PKMAS software.All “author defined” turn measures were created by authors and calculated using Microsoft Excel.

^#time stamps and mat location coordinates for each footstep were auto-generated by PKMAS. Standard PKMAS terminologies are defined here based on our understanding of PKMAS Measurements and Definitions manual provided by Protokinetics.

2.4. Turn bias

A turn bias index was calculated as the percentage of the total turns that were in a leftward direction (Yazgan et al. 1996). Participants were categorized as having a right bias (indices ≤40%) no bias (>40% but <60%) or left bias (≥60%) (Taylor et al. 2007). We defined non-left bias as either a right bias or no bias (indices <60%).

2.5. Statistical Analysis

All statistical analysis was performed with SPSS Version 27 (IBM) and assessed for significance at alpha (0.05) and two-tailed p values are reported. Mann-Whitney U test was used to test for significant differences in continuous variables categorized as left or non-left bias. Association between turn bias and categorical variables were tested with Chi-Square tests for association and Chi-square values reported are either Pearson’s Chi-Square or Likelihood Ratio, as applicable. Association between turn bias and ordinal variables were tested with linearity of relations. Binary logistic regression to test effect of variables on likelihood of turning leftward was performed using Forward Logistic regression methods in SPSS. In order to perform an initial feature reduction, only variables with p<0.1 were entered into the regression models. However, independent ordinal variables such as the FOGQ items would be treated as categorical variables during binary logistic regression analysis. To overcome this problem, ordinal regression analysis was performed to validate our findings from binary logistic regression, while retaining the ordinal nature of the FOGQ questions. In order to perform ordinal regression, we had to use the FOGQ questions as the dependent variables and turn bias (the dependent variable for linear regression) as an independent variable. Ordinal regression was carried out using the standard Ordinal regression output and selecting for 10% CI, default commands in Output Management System in SPSS and PLUM (Polytomous Universal Model) parameter estimates to generate odds ratios. Model improvements were driven by overall model significance, selection of parameters that improved the previous model odds ratio at p = 0.1, goodness of fit, and liberality of relations.

2.6. Identification of freezing episodes

PD participants who self-reported the presence of freezing of gait (FOG) as the sensation of feet “sticking to the ground” on Q3 of the FOGQ after the freezing phenomena was demonstrated to them, or had witnessed freezing during examination (by TV) were assessed for the presence of FOG during turn trials. A movement disorders fellowship trained neurologist (TV) reviewed video footage of these 34 FOG participants and rated each turn for the presence or absence of freezing episodes.. The turns were also visually segmented (by LP) into thirds based primarily upon the pivot portion of the turn (active torso rotation) being the middle third, with the first and last third being turn initiation and ending respectively.

3. RESULTS

A total of 100 participants were enrolled of which 72 had PD and 28 were controls. PD participants were a few years older, had worse cognition, and as would be expected, had higher UPDRS and FOGQ scores than control participants (Table 2). None of the controls met clinical criteria for PD, despite reporting on occasion difficulty with activities of daily living (UPDRS Parts 1 and 2) and having subtle findings on motor testing (UPDRS Part 3).

Table 2:

Demographics and basic disease characteristics of enrolled participants

	HC (n=28)	PD (n = 72)
Age (years)	62.5 ± 9.1	67.0 ± 7.9*
Gender (percent female)	60.7%	40.3%
Right handed	93.1%	90.3%
MoCA score	27.4 ± 1.8	24.9 ± 3.7*
FAB score	16.5 ± 1.4	15.0 ± 2.9* (n=70)
SCOPA-COG score	29.1 ± 4.7	23.0 ± 5.3* (n=70)
Disease duration (years)	-	8.1 ± 5.6
Hoehn & Yahr score	-	2.1 ± 0.7
FOGQ score	0.1 ± 0.4	6.4 ± 5.7*
Subjects with falls	-	33.3%
Fall frequency (/month)	-	0.9 ± 4.0 (n=72)
UPDRS Part III Score (motor)	1.5 ± 1.7	15.3 ± 8.1*
Total UPDRS Score	3.5 ± 2.5	28.6 ± 12.7*
Total daily levodopa dose (mg)	-	645 ± 478 (n=72)
Last levodopa dose (mg)	-	182 ± 107 (n=65)
Time to last levodopa dose (hrs)	-	2.3 ± 1.9 (n=65)

^*p<0.05

3.1. Turn bias

Only one control participant made a single rightward turn, with all other turns in a leftward direction. The other 27 controls turned leftward only. Of our 72 PD participants, 51 turned only leftward while 21had bidirectional turns. Of these 21 bidirectional turners, 10 were categorized as having a left bias and 11 as non-left bias (2 right bias and 9 no bias). So overall 85% of the PD participants had a left turn bias while 100% of the control participants had a left turn bias.

3.2. Turn bias and disease characteristics

As it has been suggested previously that turn bias may be related to deficits in spatial reasoning, we ran Chi-Square tests for turning bias against individual visuospatial reasoning subscores from the MoCA (“Copy cube” and “Draw CLOCK”) and the SCOPA-COG (“Assembling patterns”) in PD participants and found no significant association (Table 3). There was also no difference in distribution between total MoCA (Man-Whitney U=329.5, p=0924), SCOPA-COG (Man-Whitney U=310, p=0.815) or FAB (Man-Whitney U=246, p=0.199) scores in people with left turn vs non-left turn bias.

Table 3:

Chi-Square Tests of association of turn bias with cognition and disease symptoms of PD subjects

Features	Chi-square	Linear-by-linear association	Phi/Cramer’s V
Bias * MoCA - Trails	X2(1) = 1.016, p = 0.313	X2(1) = 0.924, p = 0.336	Phi = −0.114, p = 0.333
Bias * MoCA - Cube	X2(1) = 0.107, p = 0.743#	X2(1) = 0.106, p = 0.745	Phi = 0.039, p = 0.743
Bias * MoCA - Clock: contour	No statistics - constant	No statistics - constant	No statistics - constant
Bias * MoCA - Clock: numbering	X2(1) = 1.526, p = 0.217	X2(1) = 1.277, p = 0.258	Phi = −0.134, p = 0.255
Bias * MoCA - Clock: hands	X2(1) = 0.608, p = 0.435	X2(1) = 0.529, p = 0.467	Phi = −0.086, p = 0.464
Bias * SCOPA-COG - Pattern A	X2(1) = 0.345, p = 0.557	X2(1) = 0.186, p=0.666	Phi = −0.052, p = 0.664
Bias * SCOPA-COG -Pattern B	X2(1) = 0.179, p = 0.673	X2(1) = 0.163, p = 0.687	Phi = −0.049, p = 0.684
Bias * SCOPA-COG -Pattern C	X2(1) = 0.108, p = 0.743#	X2(1) = 0.106, p = 0.744	Phi = −0.039, p = 0.743
Bias * SCOPA-COG -Pattern D	X2(1) = 0.003, p = 0.953	X2(1) = 0.003, p = 0.954	Phi = −0.007, p = 0.954
Bias * SCOPA-COG -Pattern E	X2(1) = 0.419, p = 0.517	X2(1) = 0.399, p = 0.527	Phi = −0.076, p = 0.524
Bias * FOGQ Q1	X2(3) = 3.065, p = 0.382	X2(1) = 1.429, p = 0.232	Cramer’s V = 0.162, p = 0.597
Bias * FOGQ Q2	X2(3) = 2.232, p = 0.526	X2(1) = 0.003, p = 0.956	Cramer’s V = 0.16 p = 0.605
Bias * FOGQ Q3	X2(4) = 2.751, p = 0.600	X2(1) = 0.926, p = 0.336	Cramer’s V = 0.185, p = 0.653
Bias * FOGQ Q4	X2(4) = 1.004, p = 0.909	X2(1) = 0.651, p = 0.420	Cramer’s V = 0.129, p = 0.88
Bias * FOGQ Q5	X2(3) = 6.984, p = 0.072	X2(1) = 4.162, p = 0.041#	Cramer’s V = 0.352, p = 0.031
Bias * FOGQ Q6	X2(3) = 6.235, p = 0.101	X2(1) = 4.218, p = 0.040#	Cramer’s V = 0.338, p = 0.042
Bias * Disease More/Less Affected Side	X2(3) = 1.026, p=0.795	X2(1) = 0.606, p = 0.436	Cramer’s V = 0.101, p = 0.866
Bias * UPDRS – Fall score (Item 13)	X2(3) = 1.199, p = 0.753	X2(1) = 0.041, p = 0.839	Cramer’s V = 0.116, p = 0.808
Bias * Gender	X2(1) = 0.143, p = 0.705	X2(1) = 0.143, p = 0.706	Phi = −0.045, p = 0.704
Bias * Handedness	X2(2) = 2.453, p = 0.293	X2(1) = 1.249, p = 0.264	Cramer’s V = 0.139, p = 0.497
Bias * Hoehn & Yahr scores	X2(4) = 5.134, p = 0.274	X2(1) = 0.452, p = 0.501	Cramer’s V = 0.222, p = 0.471

Bias categorized as Left or non-left bias (Right or no bias)

^#p<0.05

Turn bias was not associated with gender, handedness (Table 3), or age (Man-Whitney U=268, p=0.291). Turn bias was also not related to disease severity as measured by UPDRS motor or total scores (Mann-Whitney U=297.5, p=0.552 and U=313.5, p=0.730 respectively), Hoehn and Yahr scores (Table 3), fall severity (UPDRS item 13 fall score) (Table 3), or FOGQ total or freezing severity scores (FOGQ questions 3–6) (Mann-Whitney U=261.5, p=0.245 and U=269.0, p=0.259 respectively). Chi-Square analysis on individual FOGQ questions, did show a linear-by-linear association of bias with FOGQ Q5 & Q6 (Table 3); with participants with left turn bias having lower scores, i.e. shorter average duration of freezing episodes. Turn bias was also not associated with PD disease asymmetry when comparing bias direction to the disease more or less affected side (Table 3).

3.3. Spatiotemporal dynamics of turning in relation to turn bias

The spatiotemporal dynamics for the 11 PD participants that did not have a left turn bias were compared to the 61 that had a left turn bias. There were no statistically significant differences, setting a significance value of p<0.05, between the mean or CV in spatiotemporal dynamics of turning (Table 4).

Table 4:

Spatiotemporal gait measures in participants with Parkinson’s disease

	Mean			Percent CV
Gait measure	Left bias	Non-left bias	p-value	Left bias	Non-left bias	p-value
Turn Time (s)	3.8 ± 1.1	3.7 ± 0.8	0.931	11.6 ± 6.0	11.1 ± 2.2	0.808
Turn Length (cm)	84.5 ± 18.2	81.3 ± 13.4	0.931	21.9 ± 10.6	23.5 ± 8.6	0.360
Turn Width (cm)	31.2 ± 10.3	25.4 ± 7.4	0.078	28.3 ± 12.6	31.3 ± 10.7	0.284
Turn Area (cm2)	2616 ± 942	2073 ± 766	0.089	37.6 ± 17.0	38.3 ± 11.0	0.748
Step Count (no)	6.3 ± 2.9	6.2 ± 2.0	0.950	13.9 ± 6.7	14.9 ± 4.3	0.389
Integrated Pressure (p × s)	154.9 ± 49.2	146.6 ± 47.4	0.667	30.3 ± 17.4	33.9 ± 14.1	0.197
Foot Length (cm)	27.2 ± 3.0	25.9 ± 2.1	0.068	14.2 ± 6.9	14.8 ± 7.3	0.701
Foot Width (cm)	11.5 ± 1.5	11.3 ± 1.1	0.501	9.7 ± 4.3	10.1 ± 3.2	0.526
Stride Length (cm)	44.5 ± 15.7	40.2 ± 13.3	0.377	71.8 ± 17.3	79.7 ± 25.1	0.377
Stride Time (s)	1.1 ± 0.2	1.1 ± 0.2	0.754	24.0 ± 15.9	28.5 ± 13.5	0.191
Stride Velocity (cm/s)	38.3 ± 11.5	34.4 ± 9.8	0.225	68.9 ± 22.6	68.6 ± 16.0	0.772
Stance percent	74.3 ± 6.0	75.2 ± 8.0	0.736	13.5 ± 6.3	14.1 ± 3.7	0.377
Swing percent	25.7 ± 6.0	24.8 ± 8.0	0.736	43.9 ± 26.8	52.9 ± 35.3	0.394

3.4. Factors best categorizing non-left turn bias

We performed regression models (Table 5, Figure 1) to see the effects of different variables on turn bias with bias rated as left or non-left (i.e., right or no bias) category. As an initial factor reduction, we only included variables in the model that had an alpha <0.10 on chi-square or Mann-Whitney tests. Five variables met this criteria (Table 3 and Table 4), FOGQ Q5 & Q6, turn width, turn area and mean foot length.

Table 5:

Regression models for turn bias prediction

Effect of FOGQ and spatiotemporal measures on likelihood of non-left turn bias Model Analysis Method: Binary Logistic Regression, Method: Forward LR Independent Variables: FOGQ Q5, FOGQ Q6, mean turn width, mean turn area, mean foot length.		Effect of turn bias and spatiotemporal measures on FOGQ Q5 & Q6 Model Analysis Method: Ordinal Regression Factor: Turn Bias Covariates: mean turn width, mean turn area, mean foot length Dependent variable: Model 1) FOGQ Q6          Model 2) FOGQ Q5
Model 1: Overall model significance (Chi-Square, p-value) X2(2) = 9.554, p = 0.008	Hosmer & Lameshow Test: X2(1) =1.451, p = 0.228; Pseudo R-square: Nagelkerke R2 = 0.216; Overall Percentage Correct: 87.5% Variables in equation and contribution to increased non-left turn bias: FOGQ Q6: p =0.019 Wald: 5.526, B = 0.855, 90% CI Exp(B) = 2.352 (lower:1.293, upper 4.279); Mean Turn Width: p = 0.039 Wald = 4.260, B =, −0.101 90% CI Exp(B) = 0.904 (lower: 0.834, upper 0.980); Variables not in equation FOGQ Q5; Mean turn area; Mean foot length	Model 2, Iteration 1: Effect of turn bias, turn width, turn area and foot length on FOGQ Q6 Overall model significance: X2(4) = 9.987, p = 0.041	Goodness of Fit (Pearson): X2(209) = 213.737, p = 0.396 Pseudo R-square: Nagelkerke R2 = 0.147 Test of parallel lines: X2(8) = 4.457, p = 0.814 Parameter estimates and odds ratio non-left bias: p = 0.048, Wald = 3.919, B = 1.291, 90% CI Exp(B) = 3.637 (lower: 1.244, upper: 10.636); Turn width: p = 0.202, Turn area: p = 0.978; Foot length: p = 0.165
		Model 2, Iteration 2 Effect of turn bias, turn width and foot length on FOGQ Q6 Overall model significance: X2(3) = 9.986, p = 0.019	Goodness of Fit (Pearson): X2(210) = 213.716, p = 0.416 Pseudo R-square: Nagelkerke R2 = 0.147 Test of parallel lines: X2(6) = 6.517, p = 0.368 Parameter estimates and odds ratio non-left: p = 0.048, Wald = 3.923, B = 1.292, 90% CI Exp(B) = 3.638 (lower: 1.245, upper: 10.634); Turn width: p = 0.026, Wald = 4.949, B = 0.055, 90% CI Exp(B) = 1.056 (lower: 1.014, upper: 1.1); Foot length: p = 0.152
		Model 2, Iteration 3: Effect of turn bias and turn width on FOGQ Q6 Overall model significance: X2(2) = 7.651, p = 0.022	Goodness of Fit (Pearson): X2(211) = 212.934, p = 0.450 Pseudo R-square: Nagelkerke R2 = 0.114 Test of parallel lines: X2(4) = 6.245, p = 0.182 Parameter estimates and odds ratio non-left: p = 0.028, Wald = 4.821, B = 1.412, 90% CI Exp(B) = 4.102 (lower: 1.425, upper: 11.81); Turn width: p = 0.030, Wald = 4.698, B = 0.052, 90% CI Exp(B) = 1.054 (lower: 1.013, upper: 1.097);
		Model 2, Iteration 4: Effect of turn bias and foot length on FOGQ Q6 Overall model significance: X2(2) = 4.844, p = 0.089	Goodness of Fit (Pearson): X2(211) = 197.708, p = 0.735 Pseudo R-square: Nagelkerke R2 = 0.074 Test of parallel lines: X2(4) = 5.157, p = 0.272 Parameter estimates and odds ratio non-left bias: p = 0.112, Foot length: p = 0.174
		Model 3: Effect of turn bias, turn width, turn area and foot length on FOGQ Q5 Overall model significance: X2(4) = 8.684, p = 0.070	Goodness of Fit (Pearson): X2(209) = 198.891, p = 0.681 Pseudo R-square: Nagelkerke R2 = 0.128; Test of parallel lines: X2(8) = 12.392, p = 0.135 Parameter estimates and odds ratio Foot length: p = 0.047, Wald = 3.948, B = −0.170, 90% CI Exp(B) = 0.844 (lower:0.733, upper: 0.971) Turn width p = 0.783 Turn area p = 0.565 Non-left bias p = 0.101

Fig. 1:

An analysis flowchart depicting the modeling rationale for determining the disease parameters that affect turn bias in PwPD. Features studied were grouped broadly into clinical disease features, participant fixed characteristics and spatiotemporal dynamics of turning. The categories were further subdivided and the specific measurements used in each subdivision are shown in the bottles with the cap labels depicting the evaluation instruments. Each measurement was analyzed statistically for association using Chi-square or Mann-Whitney U tests. Measurements that did not meet statistical significance are marked with an X. Variables used for further regression analysis models to predict turn bias are represented in the clouds. The final model best predicting turn bias is shown in the banner at the bottom of the figure.

The results of binary logistic regression with bias as the binary dependent variable and results representing the effects of the selected variables on the likelihood of PD participants having a non-left bias are shown in Model 1 (Table 5, left column). FOGQ Q6 and mean turn width were the two significant features in the model with overall statistical significance (p = 0.008) with a good fit (p = 0.261; Hosmer and Lameshow test). The model explained 21.6% (Nagelkerke R²) of the variation in turn bias and correctly classified 87.5% of cases. PD participants with higher scores on the FOGQ Q6 were 2.352 times more likely to have a non-left bias while turning than a left bias. Increasing turn width decreased the chances of non-left bias by a factor of 0.904 (Table 5).

Since FOGQ Q5 and Q6 are ordinal variables and would be used as categorical variables in binary regression models, the degree of dysfunction encoded in these variables (0–4 scale) would be lost on conversion to a categorical, ungraded scale. We therefore also used ordinal regression to see whether turn bias (used now as an independent variable for the purposes of modeling) predicted the FOGQ question scores (used as dependent variable). We continued to use the other variables that had p<0.1 as additional possible predictive variables. Model 2, Iteration 1 (Table 5, right column), tested for the effect of turn bias, turn width, turn area, and mean foot length on FOGQ Q6. This model was significant (p=0.041), with non-left bias as the significant variable (p=0.048). The odds ratio of PD participants with non-left bias rating higher on FOGQ Q6 (longer average turn freeze duration) was 3.637 (90% CI, 1.244–10.636) times that of PD participants with left bias. As turn-width was also a significant feature in binary regression models, we iteratively removed the inputs from our ordinal model. With turn area removed (Table 5, right column; Model 2 Iteration 2), significance of the model improved (p=0.019) and non-left bias (p=0.048) and turn width (p=0.026) were significant variables. Excluding turn width from the model (Table 5, right column; Model 2, Iteration 4) reduced the significance of the model (p=0.089).

We performed similar ordinal regression modeling for FOGQ Q5 (Table 5, right column; Model 3), and confirmed that FOGQ Q5 was not significantly affected by turn bias (p=0.070), with turn width, turn area and mean foot length were also included as potential predictive variables.

3.5. Turn bias and witnessed freezing events

As the freeze duration on turning was a factor categorizing left and non-left turn bias participants in our regression models, we also used the recorded video of the gait trials to isolate freezing episodes during turning. Thirty-four participants reported freezing of gait (FOG) based on FOGQ Q3 or had witnessed freezing on exam and were scored for the presence or absence of freezing of gait during each turn (Table 6). Thirty eight percent of the FOG participants (13/34) had witnessed freezing during turns, 10 with left turn bias and 3 with non-left turn bias. Stated another way, 34% of FOG left bias participants and 60% of FOG non-left turn bias participants had witnessed freezing on turning. Out of a total of 245 rated turns, 50 turns had freezing episodes; 20% of which were right turns and 80% left turns. However, 19% of the total left turns had freezes while 33% of the total right turns had freezes. Since only 5 FOG participants did not have a left turn bias, it is difficult to determine any trends but 41% of the rated turns in non-left turn bias FOG participants had freezing compared to 17% of turns in left bias FOG participants. Additionally, 93% of the freezes were in the middle 1/3 of the turn in non-left turn bias FOG participants, compared to 62% in left turn bias FOG participants.

Table 6:

Freezing during turns

		Turn bias		Rightward turns			Leftward turns			No. of freezes
Part. No.	Turn direction (L=leftward; R= rightward)	% left turns	Bias classification (L=leftward, N=non-leftward)	No. rightward turns	No. rightward turns with FOG	% rightward turns with FOG	No. leftward turns	No. leftward turns with FOG	% leftward turns with FOG	First 1/3	Middle 1/3	Last 1/3
29	L	100%	L	0	0		7	0	0%
9	LRLRLLLLR	67%	L	3	0	0%	6	2	33%		2
30	L	100%	L	0	0		7	1	14%			1
59	L	100%	L	0	0		7	0	0%
18	L	100%	L	0	0		7	0	0%
46	L	100%	L	0	0		7	0	0%
62	L	100%	L	0	0		7	0	0%
88	L	100%	L	0	0		7	7	100%	1	7	3
61	LLLLRRRLL	67%	L	3	0	0%	6	0	0%		1
86	L	100%	L	0	0		7	0	0%
68	L	100%	L	0	0		7	6	86%	1	5
98	L	100%	L	0	0		7	0	0%
63	L	86%	L	1	0	0%	6	0	0%
23	LLLLLLR	86%	L	1	0	0%	6	0	0%
34	L	100%	L	0	0		9	0	0%
78	L	100%	L	0	0		7	0	0%
32	L	100%	L	0	0		5	1	20%		1
37	LRLLLRL	71%	L	2	1	50%	5	2	40%	1	2
81	L	100%	L	0	0		7	3	43%		3	1
72	L	100%	L	0	0		7	0	0%
97	LRLxLLLxL	89%	L	1	0	0%	8	0	0%
80	L	100%	L	0	0		7	0	0%
100	L	100%	L	0	0		7	0	0%
10	L	100%	L	0	0		7	0	0%
15	L	100%	L	0	0		7	4	57%		3	3
69	L	100%	L	0	0		7	0	0%
31	L	100%	L	0	0		7	2	29%		1	1
65	L	100%	L	0	0		7	0	0%
55	LRLLLR	67%	L	2	2	100%	4	4	100%	3	6	4
36	LLLRRRR	43%	N	4	0	0%	3	1	33%		1
71	LRLRLRLRL	56%	N	4	4	100%	5	5	100%		8	1
53	LRLRLRL	57%	N	3	0	0%	4	0	0%
54	LRLRLRL	57%	N	3	0	0%	4	0	0%
12	LLRLRLR	57%	N	3	3	100%	4	2	50%		5

4. DISCUSSION

In this study, we explored whether PwPD have a preferential bias to turning left or non-left and the disease features that may associate with turn bias. We found that turn bias was not associated with disease asymmetry, handedness, disease severity, sex, disease duration or cognitive function. Turn freeze duration (based on FOGQ Q6) and turn width were most predictive of a non-left turn bias while rightward turns had a higher frequency of freezing episodes. These results provide evidence for targeted therapy to help improve turning gait stability in PwPD, when falls are most common.

It has previously been suggested by a small study of people with hemi-parkinsonism (Bracha et al. 1987b) and in studies on healthy individuals with asymmetry in basal ganglia size (Kooistra and Heilman 1988) and dopamine levels (Glick et al. 1982) that asymmetry in dopamine levels may determine turn bias. Our results however do not support this hypothesis as both left-sided and right-sided disease more affected participants in our cohort had a predominant leftward turn bias (89% and 82% respectively) using previously defined turn bias calculations (Yazgan et al. 1996; Taylor et al. 2007). While we did not directly measure dopamine levels in our participants, disease asymmetry in PwPD is due to asymmetric loss of dopamine in the basal ganglia (Hoehn and Yahr 1967; Fahn and Elton 1987; Varrone et al. 2001). Taken together our results suggest that asymmetry in dopamine may not be the primary cause of turn bias in PwPD.

We did not see any dependence of handedness (Yazgan et al. 1996; Mohr et al. 2003a) or gender (Yazgan et al. 1996) on turn bias as has also been previously suggested. Our cohort had 7 left-handed and 2 ambidextrous participants, and none of these participants made any rightward turns during the trials. All our healthy controls (93% right handed, 60% female) also had a left bias, although one participant (right handed female) did make one rightward turn. In contrast, Yazgan et al. (Yazgan et al. 1996) in a younger, healthy cohort, reported 43% of left-handed/ambidextrous participants had a right turning bias. We did not however explicitly administer a handedness questionnaire as these prior studies had.

It has been reported in a small cohort of young university students that higher scores on visuospatial tasks (localization of objects on a frame, orientation of objects, contacting objects and symbol-digit mapping tasks) were correlated with a leftward turn bias (Gordon et al. 1992). As cognition is affected in PwPD, including in ability to perform visuospatial tasks (Emre et al. 2007; Goldman et al. 2015) we explored whether these components of cognitive testing could be related to a turn bias. We did not see any correlation in leftward bias either to scores on the subset of visuospatial tasks on the MoCA and SCOPA-COG, or to the total global scores on these cognitive tests, even though scores were lower in our PD cohort than the controls. This suggests that in PwPD, turn bias is not related to the associated decline in cognition that can occur as the disease progresses.

FOGQ Q6 was predictive of non-left turn bias. This question specifically asks people about the average time spent in a freeze episode during turning, ranging from 0 to greater than 30s average turn freezing. Based on our models longer turn freezing was more predictive of non-left turn bias. This could be due to intrinsic or extrinsic factors. Perhaps, leftward turns are their pre-disease norms and asymmetric dopamine loss, leads to turning in other directions, which are less “learned” or “patterned”, leading to longer duration of freezing. Alternatively, a conscious effort to try alternative turn directions due to a perception of freezing on turning leftward, could also lead to choosing the less patterned turn paradigm, and in turn to longer freezing episodes. Diaries to document whether there was patient selection at the time of turns could help, but then also would potentially bias the participant towards consideration of a rightward turn. Alternatively cuing participants to turn right or left in a random order and observing freezing frequency and duration could be performed. A prior study did not find a difference in turn steps, turn time, cadence or freezing episodes when participants were randomly guided to turn towards the disease more or less affected side (Spildooren et al. 2012). This could be due to participants focusing greater on their turns when guided to turn in a particular direction, compared to our study in which they were unguided.

While the spatiotemporal parameters of turning were not significantly different between participants with left turn or non-left turn bias, smaller turn width was predictive of greater incidence of non-left turn bias. Whether the smaller turn width is an adaptation or the consequence of turning in a less preferred direction is unclear from our data. The sequence effect model for development of freezing of gait suggests that in the setting of reduced stride length in PwPD, successively shorter steps could lead to a freezing episode (Iansek et al. 2006; Chee et al. 2009; Virmani et al. 2018). Following this logic, shorter steps during turning due to a smaller turn width could provoke freezing. In support of this, freezing episodes were more common in rightward turns where turn-width was shorter, although not reaching a statistically significant difference (p=0.078), with turn steps the same, which would mean shorter steps. Physical therapy interventions to re-train patients to turn leftward, and to take longer steps with wider turn width, could help improve turn stability and reduce freezing in those prone to freeze.

One limitation of our work is that we only evaluated PwPD in their levodopa medicated state and therefore cannot comment on whether there may have been a greater turn bias if participants had been evaluated in the levodopa unmedicated state. A number of factors still support our conclusions. The motor disease asymmetry in participants was still present on examination even in the levodopa-medicated state. Additionally, the effect of levodopa on turn parameters is unclear as one small study reported no improvement in step number and turn time while making 180 degree turns in place (Hong and Earhart 2010) while another did (McNeely and Earhart 2011). Even in the levodopa medicated state, 38% of participants with FOG in our study had freezing episodes during “normal” turning, comparable to prior studies performed in the levodopa unmedicated state (31–38%) (Spildooren et al. 2013; Bengevoord et al. 2016), suggesting that levodopa may not alter turn dynamics significantly. Future studies in both levodopa-unmedicated and medicated conditions in the same individuals would help resolve this issue.

5. CONCLUSIONS

Turning bias was predominantly leftward in PwPD irrespective of handedness or disease side predominance, arguing against disease asymmetry guiding turn directionality. Cognitive dysfunction in our cohort was also not related to turning bias. Longer turn freeze duration and shorter turn width were related to greater non-left turn bias. Our results therefore suggest that physical therapy interventions directed towards modifying turn directionality, increasing turn width and step length could help improve turning stability and decrease freezing of gait episodes.

Abbreviations:

PD

Parkinson’s disease

PwPD

People with Parkinson’s disease

FOG

freezing of gait

FOGQ

Freezing of Gait Questionnaire

UPDRS

Unified Parkinson’s Disease Rating Scale

H&Y

Hoehn and Yahr

MoCA

Montreal Cognitive Assessment Score

FAB

Frontal Assessment Battery

SCOPA-COG

Scales for Outcome in Parkinson’s Disease – Cognition

PKMAS

ProtoKinetics Movement Analysis Software

CV

Coefficient of Variation

PLUM

Polytomous Universal Model

Data Availability:

This data is part of an ongoing longitudinal study. However, de-identified data related to this publication can be made available to a qualified researcher upon contact with the corresponding author and in accordance to guidelines set by the protocol.