A total of 30 participants with PD and 17 age-matched participants without PD were recruited. All participants provided written informed consent as approved by the institutional review committee of the Nanjing Brain Hospital. The inclusion criteria for the PD cohort were: having idiopathic PD as diagnosed by experienced clinicians based on the Chinese Diagnostic Criteria of Parkinson's Disease (2016), a surgery plan of DBS within 2 months, and able to stand and walk unassisted for more than 10 minutes. The exclusion criteria were: being younger than 40 years, having any other major neurological diseases (eg, stroke, dementia), and having ongoing psychiatric disturbances such as hallucinations.

Before and after the treatments (ie, the acute levodopa challenge test [ALCT] or DBS), each participant in the PD group completed one 10-meter walking test for each of the following conditions: walking normally (ie, single task) and walking while performing a cognitive task (ie, dual task). The cognitive task was the serial subtraction of 3 or 7 from a random three-digit number. The UPDRS-III was also completed before and after the treatments and was used to assess the severity of PD. The healthy cohort completed one study visit consisting of the same single and dual task walking trials.

All trials were completed in the same room. One study personnel stood at the end of the walkway, and their position was fixed. No markers were used in the room, as the marker may give a cue to the participants in the PD group, and the pathway was not approaching a wall or doorway, which could interfere with participants’ gait. Each participant stood in front of a wooden chair on one side of the pathway at the beginning of the trial and was instructed to walk along the pathway and stop at the same position as the study personnel. During each walking trial, two wearable sensors were used and attached to each ankle. The kinematic signals of walking including the angular velocity were then recorded and used to quantify the gait metrics.

The wearable sensors (Figure 1) used in the study were developed by our team and embedded with inertial measurement units (chip MPU-9250, IvenSense Inc, San Jose, CA). The sensors captured triaxial acceleration, angular velocity, and magnetic field intensity signals. The size of the sensors was 52 mm by 37 mm by 13 mm, and they each weighed 26.3 g. The average operating current of the sensors was 38.1 mA allowing for 13 hours of continuous recording. The sampling rate was 100 Hz, and the range of measurement for the gyroscope was ±1000 °/sec with a resolution of 0.06 °/sec/least significant bit.

The ALCT was performed in the morning using the established formulation to observe the clinical improvement in the PD group following withdrawal of all antiparkinsonian medication and overnight fasting [21]. The ALCT was used in clinics to screen patients who could use DBS based on their response to medication [21], that is if patients’ UPDRS-III score reduced by more than 30% after the ALCT, they would benefit from DBS.

DBS is a type of neurosurgical surgery that sends electrical impulses through implanted electrodes to specific brain nuclei and alleviates the burden of multiple movement disorders such as tremors in PD [22]. In this study, the DBS surgery was conducted after a minimum 1-week break from the ALCT. The pulse of the DBS was set up using the width of 60 μs and a frequency of 130 Hz. On the DBS visit, which was during participants’ perioperative period, participants completed the walking test with the DBS off and on without taking any medicine.

Figure 2 shows the pipeline of the data processing. First, we removed isolated noise points by using a 5-point median filter, and a band-pass filter was applied to remove fluctuations of frequency greater than 12 Hz and lower than 0.1 Hz. Second, the angular velocity of the axis with the largest variance was selected (Figure 3A). We observed that the angular changes in the sagittal plane during walking were obvious and much greater than the changes in other planes, so we focused on the angular characteristics of gait in this plane. Third, to divide gait cycles accurately, a threshold was set to exclude the influence of small displacements of the legs, and only the peaks and valleys that were greater than the threshold (ie, the heel strikes or toe-offs) were identified and used. The threshold was calculated as the mean of this signal plus or minus √2 / 2 SD (Figure 3B). The next step is the transformation from angular velocity to the degree of angle. The degree of angle was converted by the integral of the processed velocity signal. Finally, the step angle of each step was calculated as the difference of angle, A_(i), between a peak of the wave, that is the angle of the maximum threshold=T_max(i), and the adjacent valley, that is the angle of the minimum threshold=T_min(i), in the angle degree series (Figure 3C).

Five gait metrics were calculated from the angular velocity signal, including 2 conventional gait metrics (step time and step number) and 3 new metrics (mean step angle, initial step angle, and last step angle). The definition of step angle is the angular change of the ankle in the sagittal plane within 1 gait cycle. The definitions of all 5 gait metrics are provided in Textbox 1. All the metrics here were obtained by averaging the left and right legs.

The statistical analysis was performed using SPSS 20 (IBM Corp, Armonk, NY). To examine the effects of group and task on step angle, 2-factor multivariate analyses of variance (MANOVAs) were used. The outcomes measured before the ALCT were used as the baseline outcomes of the PD group, as there was no influence for the implanted DBS electrodes. The dependent variables were mean step angle, initial step angle, and last step angle measured before the treatment (ie, baseline) in each model, and the model effects included the groups (ie, PD vs healthy), tasks (ie, single vs dual task), and their interactions. Similar models were also used to examine the effects of group and task on the two traditional metrics (ie, step time and number). To explore the effects of the ALCT on the UPDRS-III score and gait metrics, a 2-way MANOVA was used. The dependent variables were the 5 gait metrics and the UPDRS-III score. The model effects were time (before and after), task condition (single and dual task), and their interaction. The effects of DBS were then examined using the same models. If the interaction effect was significant, the analysis of simple effect was applied to further analyze the differences of the 5 gait metrics (dependent variables) between two levels of one independent variable while fixing another factor. The Bonferroni correction was used for multiple comparisons, where the significance level was set at P<.01. Partial eta-square was calculated for the effect size of the MANOVA. For partial eta-squares, 0.01, 0.06, and 0.14 were considered as small, medium, and large effect sizes, respectively [23].

Then a partial correlation analysis adjusted for age, gender, and status of treatment was used to explore the association between the UPDRS-III score and gait metrics at baseline. In addition, the independent-samples t test (two-tailed) and the Mann-Whitney U test were used to examine the differences in age and gender, respectively, between the two groups. A Bonferroni correction was used for the multiple comparisons, and the significance was set as P<.01.

The significance level was set at P<.05, not including those corrected by the Bonferroni correction.

A total of 18 participants in the PD group completed the tests at baseline (ie, before the ALCT), after the ALCT, and after the DBS surgery. In addition, 25 participants in the PD group completed the tests when the DBS was on and off. All participants completed the study. Table 1 shows their demographic and clinical information. No significant differences in age (F_2,57=0.18, P=.84) or gender (χ²₂=1.66, P=.44) between the PD cohort and non-PD control group were observed. There were no significant differences of disease duration (t₃₆=0.16, P=.88), the Hoehn-Yahr stage (t₃₂=0.04, P=.97), and UPDRS-III score (t₂₂=–0.24, P=.81) between the ALCT and DBS groups.

A significant main effect of group was observed, but no significant difference was observed for the effect of task and their interaction (Table 2). The mean step angle (F_1,48=69.75, P<.001, partial eta-square=0.59), initial step angle (F_1,48=15.56, P<.001, partial eta-square=0.25), and last step angle (F_1,48=61.99, P<.001, partial eta-square=0.56) in participants with PD were all significantly smaller than healthy people. Similar results were shown in those conventional metrics that the PD cohort had significantly larger step times (F_1,48=7.52, P=.009, partial eta-square=0.14) and more step numbers (F_1,48=12.05, P=.001, partial eta-square=0.20) than the control.

We observed that the UPDRS-III score significantly decreased after the treatments (ie, the ALCT and DBS) compared to the score before the treatments (ACLT:t₁₁=–7.81, P<.001; DBS:t₁₅=–15.22, P<.001).

Table 5 presented the association between the gait metrics and the UPDRS-III score. Age, gender, and condition were controlled in this analysis, and the degree of freedom in single and dual walking tasks were 58 and 51, respectively. The mean, initial, and last step angles in both single and dual walking tasks were all significantly correlated with the UPDRS-III score (r>–0.35, P<.01). A weaker correlation was observed between the step time and number and the UPDRS-III scores. The scatter plots of the UPDRS-III score and gait metrics are shown in Multimedia Appendix 1.