Sensor‐based gait analysis in atypical parkinsonian disorders

Abstract Background and Objectives Gait impairment and reduced mobility are typical features of idiopathic Parkinson's disease (iPD) and atypical parkinsonian disorders (APD). Quantitative gait assessment may have value in the diagnostic workup of parkinsonian patients and as endpoint in clinical trials. The study aimed to identify quantitative gait parameter differences in iPD and APD patients using sensor‐based gait analysis and to correlate gait parameters with clinical rating scales. Subjects and Methods Patients with iPD and APD including Parkinson variant multiple system atrophy and progressive supranuclear palsy matched for age, gender, and Hoehn and Yahr (≤3) were recruited at two Movement Disorder Units and assessed using standardized clinical rating scales (MDS‐UPDRS‐3, UMSARS, PSP‐RS). Gait analysis consisted of inertial sensor units laterally attached to shoes, generating as objective targets spatiotemporal gait parameters from 4 × 10 m walk tests. Results Objective sensor‐based gait analysis showed that gait speed and stride length were markedly reduced in APD compared to iPD patients. Moreover, clinical ratings significantly correlated with gait speed and stride length in APD patients. Conclusion Our findings suggest that patients with APD had more severely impaired gait parameters than iPD patients despite similar disease severity. Instrumented gait analysis provides complementary rater independent, quantitative parameters that can be exploited for clinical trials and care.

including motor and non-motor symptoms of parkinsonian disorders is commonly defined by semi-quantitative rating scales such as the Movement Disorder Society Unified Parkinson's Disease Rating Scale (MDS-UPDRS) (Goetz, 2010), Unified MSA Rating Scale (UMSARS) (Wenning et al., 2004), and PSP Rating Scale (PSP-RS) (Golbe & Ohman-Strickland, 2007), showing good to excellent construct and face validity. However, non-metric rating scales do not provide objective, metric measures for clinical studies. These considerations led to the development of complementary quantitative assessment tools of motor (in particular locomotor) function in iPD (Espay et al., 2016;Maetzler, Klucken, & Horne, 2016;Lees, Hardy, & Revesz, 2009).
From a biomechanical point of view, gait is a highly regular and cyclic movement, which makes it ideal for automated sensor-based detection and subsequent quantitative and qualitative analysis with a high biomechanical resolution. Body-worn inertial measurement units, comprising of the biosensors 3D accelerometer, gyroscope, and magnetometer are able to objectively measure changes of gait patterns in PD Klucken et al., 2013;Kegelmeyer, Parthasarathy, Kostyk, White, & Kloos, 2013;Schlachetzki et al., 2017). A new era in medical engineering is emerging, where objective real-time motion metrics in iPD could be obtained in virtually any environmental scenario by placing lightweight wearable sensors in the patient's clothes, and connecting them to a medical database through mobile devices such as cell phones or tablets (Pasluosta, Gassner, Winkler, Klucken, & Eskofier, 2015;Schlachetzki et al., 2017). This approach provides comprehensive, objective and metric data, enabling an assessment for clinical studies in iPD which is free from the confounds of observer bias (Espay et al., 2016;Lees et al., 2009). It also allows continuous patient monitoring even in unsupervised, habitual environments (Del Din, Godfrey, Mazzà, Lord, & Rochester, 2016). Furthermore, bringing healthcare technology into clinical practice might improve diagnostic accuracy and provide the base for multidisciplinary care concepts using raterindependent quantitative measures of motor signs.
Whereas numerous studies showed the feasibility of wearable sensors in iPD, there are only a few studies about the use of instrumented technology in APD (Baston, Mancini, Schoneburg, Horak, & Rocchi, 2014;Hatanaka et al., 2016;Sale et al., 2014;Sanchez-Ferro et al., 2016). To our knowledge, there are no published studies that compare results of instrumented gait analysis between cohorts of MSA, PSP, iPD patients, and healthy subjects.
The goal of the present cross-sectional study was to test whether a mobile, objective, and simple to use gait assessment system is able to detect differences in gait parameters between APD (Parkinsonvariant MSA = MSA-P, PSP) vs. iPDs patients vs. controls. Also, to decipher the clinical value of the gait parameter alterations, we correlated them to clinical ratings in standardized scales (MDS-UPDRS part 3 = MDS-UPDRS-3 (Goetz, 2010; Movement disorder society task force, 2003), UMSARS part 1 and 2 (=UMSARS 1/2) (Wenning et al., 2004), and PSP-RS (Golbe & Ohman-Strickland, 2007). We not only identified distinct gait parameters that differed between matched patient/control cohorts, but also their correlation to the different clinical characteristics, underlining the complementary diagnostic value of sensor-based gait assessment.

| SUBJEC TS AND ME THODS
In all, 50 patients were enrolled in the outpatient clinics of the Department of Neurology at the Medical University Hospital Innsbruck, Austria, and the Department of Molecular Neurology at the University Hospital Erlangen, Germany. iPD and APD (probable MSA-P n = 11, possible MSA-P n = 2, probable PSP n = 12 patients) were defined according to standard diagnostic criteria (Movement disorder society task force, 2003; Gilman et al., 2008;Litvan et al., 1996). Exclusion criteria consisted of non-PD-related gait impairments (e.g., spinal or orthopedic surgery), spasticity, stroke, neuropathy, myelopathy, hydrocephalus, and severe dementia. All patients were investigated in stable ON medication without the presence of motor fluctuations. This study has been approved by the local ethics committees in Erlangen, Germany iPD, MSA-P, and PSP patients were rated by MDS-UPDRS-3. Some iPD patients were rated by UPDRS-3 (Rampp et al., 2015). For these patients, the conversion method of UPDRS-3 into MDS-UPDRS-3 was applied (Hentz et al., 2015). Furthermore, MSA-P patients were scored using UMSARS (Wenning et al., 2004) and PSP patients were assessed by PSP-RS (Golbe & Ohman-Strickland, 2007).  (Schlachetzki et al., 2017). Walking performance was captured using a sensor-based gait analysis system. This system consists of wearable SHIMMER 2 sensors (Shimmer Research Ltd., Dublin, Ireland) laterally attached to the posterior lateral portion of both shoes. Gait signals were recorded within a (tri-axial) accelerometer range of ±6 g (sensitivity 300 mV/g), a gyroscope range of ±500 degree/s (sensitivity 2 mV/ degree/s), and a sampling rate of 102,4 Hz. Sensor signals were transmitted via Bluetooth ® to a tablet computer and stored for subsequent data analysis (Kegelmeyer et al., 2013). Inertial sensor data were processed with a pattern recognition algorithm for calculating clinically relevant spatiotemporal gait parameters (e.g., stride length, F I G U R E 1 Spatiotemporal gait parameters (Mean ± SD) in patients with atypical Parkinson disorders (APD), patients with Parkinson's disease (iPD)-(matched by age, gender, age of onset, and Hoehn & Yahr disease stage), and healthy controls (matched by age and gender). Max toe clearance (cm), Maximum toe height during swing phase; Heel strike angle (°), Angle of heel contacting the floor at initiation of stance phase; Toe off angle (°), Angle of toe during push-off at end of stance phase. * p < 0.05, ** p < 0.01, *** p < 0.001 Bonferroni post-hoc test gait speed, maximum toe clearance) Rampp et al., 2015). Participants performed standardized overground walking tests on a 10-m long corridor in the hospital in self-chosen walking speed. Only straight strides were automatically detected by the stride detection algorithm  and used for gait parameter calculations as described (Rampp et al., 2015). Calculated gait speed, stride length, cadence, and maximum toe clearance were normalized to the height of the participants.
A one-way ANOVA was used to detect differences in spatiotemporal gait parameters between groups. For all parameters that were compared between the three groups (iPD, APD, Controls), Bonferroni post-hoc test was used to account for multiple comparisons. Mann-Whitney-U Test was performed for the subgroup analysis in MSA patients in which participants with and without impairment in body sway and walking were compared in terms of gait parameters. In a second step, correlation analysis was used to evaluate associations between clinical scores (MDS-UPDRS, UMSARS, and PSP-RS) and spatiotemporal gait parameters. Spearman's Rank correlation was used to evaluate correlations in this small cohort.

| RE SULTS
Patient characteristics are shown in Table 1. A detailed description of the gait parameters in each group is shown in Figure 1 and an overview of calculated gait parameters is shown in the supplemen- p > .05). We also analyzed MSA-P and PSP patients separately, According to our second goal, we correlated gait speed and stride length of patients with MSA-P with the UMSARS total score, with the part 1 (=historical review of motor and non-motor symptoms including walking, falling, and orthostatic symptoms; UMSARS-1) and 2 (=Motor Examination without rating non-motor symptoms, UMSARS-2) (Figure 3a-c). Here, we observed a significant inverse correlation of gait speed and stride length with UMSARS total and UMSARS-1 but not with UMSARS-2 scores. According to the item 13 of the UMSARS-2 (body sway), we divided MSA patients into two subgroups, namely patients who recovered unaided (e.g., 0-1 rating points) and patients who would fall if not caught (e.g., 2-4 rating points) and we compared these subgroups in terms of gait speed (p = .013) and stride length (p = .040), observing a statistically relevant difference (Figure 4a). Similarly, we divided ratings for the item gait (14 of UMSARS-2) into normal/mildly impaired (e.g., 0-1 rating points) and moderately/severely impaired (e.g., 2-4 rating points).
Here, a significant difference of the subgroups with gait speed (p = .011) and stride length (p = .011) was also shown (Figure 4b).
Moreover, stride length correlated with PSP-RS scores in the PSP patients (p = .021) (Figure 3d). Finally, we observed a significant correlation of maximum toe clearance with MDS-UPDRS-3 (r = −.444, p = .026) in APD. In contrast, maximum toe clearance did not correlate with MDS-UPDRS-3 in iPD patients, whereas gait speed and stride length did.

| D ISCUSS I ON
Our study is the first that uses sensor-based technology in APD patients, comparing objective spatiotemporal gait parameters with iPD and controls. All gait items except cadence showed gait and motor impairment in both parkinsonian cohorts compared to controls, similarly to a previous study in iPD patients (Schlachetzki et al., 2017). Among the different gait parameters, gait speed not only differentiated between controls and patients, but it was also more strongly reduced in APD compared to PD patients, despite similar global motor disability according to H&Y scores, indicating a more severe alteration of locomotor abnormality in APD patients. A similar tendency was shown for stride length. Our data demonstrate that these two gait parameters are a quantitative, metric measure for gait impairment and that they differ between APD and iPD. However, we were not able to discriminate MSA and PSP patients based on spatiotemporal gait parameters. This may reflect insufficient sample size, but also overlapping gait pathophysiology. Both MSA and PSP are characterized by levodopa refractory parkinsonism, impaired cerebellar outflow, and frontal lobe dysfunction all of which may contribute to gait disorders in APD patients (Wenning et al., 1999).
The second hypothesis of our study was that gait parameters correlate with clinical rating scores. Here, we observed a moderate correlation between MDS-UPDRS-3 and maximum toe clearance in APD patients. In contrast, we showed that gait speed and stride length correlated significantly with MDS-UPDRS-3 in iPD but not in APD. This finding may reflect the contribution of non-parkinsonian impairments to the gait disorder of APD patients such as ataxia, orthostatic dysregulation, and frontal lobe impairment.
Moreover, we correlated gait speed and stride length of 10 MSA patients to the UMSARS total and specific clinical scores UMSARS-1 F I G U R E 4 Comparison of gait parameters between (a) MSA patients with and without postural instability (BODY SWAY) and (b) between gait impairment levels rated by item gait of the UMSARS-2. No/mildly (0/1) impaired gait and moderately/severely (>1) impaired gait in MSA patients were compared to objective gait parameters and UMSARS-2, respectively. We demonstrated a moderate inverse correlation with the UMSARS total score and a strong inverse correlation with the UMSARS-1 but not with the UMSARS-2. The UMSARS-1 includes the historical review of motor and non-motor symptoms including walking, falling, and orthostatic symptoms, which are main features of MSA and, in addition, markedly influence gait. On the contrary, the UMSARS-2 includes the motor examination without considering non-motor symptoms. We hypothesized that only UMSARS-2 items "gait" and "body sway" are likely directly mirrored by objective gait parameters (Schlachetzki et al., 2017).
Indeed, we observed a significant difference of gait speed and stride length between mildly and severely impaired MSA patients divided according to the single items "body sway" (UMSARS-2 item 13) and "gait" (UMSARS-2 item 14). However, these preliminary results in this small MSA cohort should be carefully interpreted and need further investigation.
Intriguingly, PSP in contrast to MSA and iPD patients revealed a positive correlation of disease severity as determined by PSP-RS with gait speed and stride length. However, it should be noticed that three PSP patients used a gait support device, two of them a wheeled walker, the other one crutches. It has been shown that a four-wheeled walker improves gait in iPD patients (Kegelmeyer et al., 2013) and in geriatric patients (Schülein et al., 2017). In our study, these two patients that used a wheeled walker showed the largest strides and highest gait speed within the PSP cohort indicating that the correlation is biased by the wheeled walker gait patterns.
We acknowledge that our APD patients' cohorts were rather small because of the rarity of these disorders. However, the significant difference of objective gait parameters among patient groups suggests that sensor-based technology may support and complement the clinical assessment provided by validated rating scales. Longitudinal follow-up studies in larger cohorts are needed to establish sensor-based technology as outcome in trials and homecare.

ACK N OWLED G M ENTS
We acknowledge support by Deutsche Forschungsgemeinschaft and Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) within the funding program Open Access Publishing. KG, Athenion GmbH, and Thomashilfen GmbH; as well as lecturing