## Introduction

Impaired balance and gait are associated with an increased risk of falls and significant morbidity and mortality in those aged 65 and older1.

Early detection of higher risk for falls is critical to begin timely intervention prior to fall episodes2. Gait assessments are used extensively to identify those at higher risk for falling3. Traditional gait assessments are principally based on volitional movements. However, most falls occur when an individual fails to recover from an unexpected loss of balance4. Recovering from gait perturbation elicits very specific neural control strategies and people at higher risk for falling usually show a reduced ability to recover5,6. Therefore, gait assessments that challenge balance have been proposed as a more effective method for early identification and remediation of gait deficits7,8. The majority of research performed using simulated balance-demanding tasks have included mechanical perturbations using either moveable platforms, obstacles, slippery surfaces, treadmill accelerations, and waist-pull perturbations5,7,8.

Recent advancements in virtual reality (VR) technology have allowed access to novel paradigms for studying balance. Specifically, VR provides a safe and controlled environment for studying how humans react and adapt to the virtual environment (VE) and to conflicting sensorimotor stimulation18. VR systems that include sensory discordance in terms of unpredictable oscillations of the visual field can challenge an individual’s ability to maintain stability, thus targeting the neuromuscular skills required for fall prevention. Studying the effect of visual perturbations during gait allows for understanding the ways in which individuals respond to challenges to their balance and may be useful for diagnostic and training purposes.

Studies that analyzed gait in response to visual perturbations have mostly been conducted using treadmills rather than overground walking with virtual screens and domes rather than virtual reality headsets. Virtual reality headsets, however, have many potential advantages including: (1) providing immersive binocular depth cues and a first-person point of view for better responses to visual stimuli, (2) increasing the area through which a user can navigate which allows for overground-based rather than treadmill-based interventions, and (3) allowing for comparatively cheaper and portable exposure to the virtual environment in comparison with virtual reality domes or screens18,19.

Previous studies have showed that young adults walking on treadmills in the virtual environment display a shorter stride length, increased step width, and increased variability in stride velocity and step width in comparison to walking in the real environment20. Furthermore, during treadmill walking, young adults have increased gait instability when exposed to anteroposterior (AP) or mediolateral (ML) perturbations but are more affected by ML perturbation than the AP perturbation, with increased instability in the direction of the perturbation21. Exposure to visual perturbations during treadmill walking is also associated with shorter step length and increased step width compared to the unperturbed state, with increased effect of ML perturbations compared to AP perturbations. ML and AP visual perturbations are also associated with increased variability in step width and step length compared to the unperturbed state, while ML visual perturbations are associated with increased variability in step width and stride length compared to AP visual perturbations. ML perturbations are associated with increased stride time variability and decreased stride time compared to the unperturbed state22.

We have recently developed a VR platform that allows us to analyze the effect of visual perturbations and the prevalence of visuomotor adaptation during unperturbed overground walking or in response to continuous multidirectional visual oscillations in a VR headset18. A study conducted on 12 young participants showed that overground walking in the VE was associated with shorter stride length and higher stride width variability compared to the real environment (RE). When exposed to visual perturbations, young adults displayed decreased stride length, increased stride width, and increased stride variability, particularly with visual perturbations in the ML direction. However, they appeared to adapt to both the VE and the perturbations by increasing stride length and decreasing stride width, stride time variability, stride length variability, and stride time18. The aims of this pilot study were to investigate age-related modifications and propensity for visuomotor adaptations over time of spatio-temporal gait parameters due to prolonged walking in a VE, and during continuous antero-posterior (AP) and medio-lateral (ML) oscillations of the visual field. We hypothesize that: (1) the spatiotemporal gait characteristics of older adults will be more affected by the virtual environment and visual perturbations when compared to younger adults, (2) older adults will adapt to the virtual environment and visual perturbations but to a lesser degree than young adults, and (3) effects would be larger for ML perturbations than AP perturbations.

## Methods

Twelve young and twenty-six older subjects participated in the experiment. The older subjects consisted of adults who visited the Otolaryngology Clinic at Columbia University Irving Medical Center while the younger adults were healthy volunteers recruited at Columbia University’s undergraduate campus. Data for young participants was the same used in our previous study18. All methods/experiments were carried out in accordance with relevant guidelines and regulations. All participants signed a written informed consent form. Research received ethical approval from Columbia University’s Institutional Review Board prior to subject participation. All subjects had normal or corrected-to-normal vision. Given that we were interested in older adults without diagnosed vestibular or neurological disorders who ambulated independently, the following subjects were excluded: four subjects with a diagnosis of vertigo, one patient who walked with a cane, and one patient with a history of Parkinson’s disease. Thus, twenty older subjects and twelve young subjects were included in data analysis. Table 1 summarizes anthropometric characteristics for the twenty elderly and twelve young subjects included in the study.

The protocol below is the same of that described in Martelli et al.18 Subjects were instructed to walk for either 5 min (older adults) or 6 min (younger adults) on a 6.0 m × 0.6 m instrumented walkway (Zeno Walkway, Protokinetics) at their self-selected speed under four conditions: real environment (RE), virtual environment (VE), virtual environment with medio-lateral (ML) visual perturbations, and virtual environment with antero-posterior (AP) visual perturbations. A 1-min walk on the instrumented walkway in the RE followed each virtual environment condition to washout possible residual effects of walking in the virtual environment. Older adults walked 1 min less than younger subjects to reduce the risk of fatigue. Moreover, in our previous study we noticed that most of the adaptations happened in the first 5 min of walking. Subjects could take breaks between sessions as necessary to minimize confounding effects of exhaustion or poor effort on gait. A virtual reality headset (HTC VIVE) was worn during all conditions in the virtual environment. A Unity3D program was created to display a virtual walkway within the VR headset, with the virtual walkway calibrated to align with the physical walkway in the RE. The 3D virtual scene was set outside with trees and grass along the pathway. Two stationary cameras were used to track the position and orientation of the VR headset in space. The VR headset cables were connected to the laptop containing the virtual scene and the laptop was placed in a backpack worn by a researcher as the subject walked on the walkway to prevent tangling of the wires. Subjects had a first-person view of themselves during walking trials in the VE, as their bodies were not rendered in the VE. To account for the possibility of learning effects, the order in which subjects experienced AP and ML visual perturbations was randomized, with some subjects experiencing AP followed by ML perturbations and vice versa. Randomization was not performed for the RE and VE conditions in accordance to the protocol followed in Martelli et al.18. We wanted the participants to get used to the unperturbed VE before starting with the AP and ML conditions in order to ensure that gait modifications were mostly due to the visual perturbations rather than the VE itself. Although subjects were informed that they might encounter AP or ML visual perturbations, they were not aware of the magnitude or direction of the perturbations. Perturbations were presented to subjects by superimposing oscillations of the visual field on normal visual flow; oscillations consisted of a pseudo-random sum of sines of four different frequencies using the equation:

$${\text{D(t)}} = {\text{A}}[{\sin}(0.{16} \cdot {2}\uppi {\text{t}}) + 0.{\text{8sin}}(0.{21} \cdot {2}\uppi {\text{t}}) + {1}.{\text{4sin}}(0.{24} \cdot {2}\uppi {\text{t}}) + 0.{\text{5sin}}(0.{49} \cdot {2}\uppi {\text{t}})]$$

where D(t) represents the superimposed displacement of objects in meters that subjects perceive during visual perturbations in the VE, A represents a scaling factor of 0.5, and t represents time in seconds18.

PKMAS gait recording and analysis software was used to measure spatiotemporal gait parameters at 120 Hz from the instrumented walkway. Specifically, stride length (SL), stride width (SW), stride time (ST), and stride velocity (SV) were calculated for both the left and right feet. Here, SL is defined as the distance between successive heel strikes of the same foot in the direction of gait progression, SW is defined as the perpendicular distance between a line connecting two ipsilateral heel strikes and the contralateral heel strike, stride time is calculated as the time between consecutive heel strikes of the same foot, and stride velocity is stride length/stride time. These spatio-temporal parameters were selected as they were analyzed in many similar studies22,24,29,30 and showed to be the most meaningful ones that we were able to extract with PKMAS software.

The VR system is only able to track a limited area, which restricted the length a subject could walk continuously in the hallway. The VR working space was only slightly longer than the walkway (with a total length of 6.6 m, including both sensorized and non-sensorized areas). In order to ensure that the steps used in the analysis were collected during steady state walking, for each lap the first and last three steps were excluded from the analysis. Furthermore, (1) the first and last steps were outside the sensorized area of the walkway and were not recorded; and (t) the second and penultimate steps were inside the sensorized area of the walkway but were excluded from the analysis. Accordingly, gait recordings started at the end of the third step (i.e., the first analyzed stride was based on the fourth and fifth steps). Moreover, data processing excluded data points if subjects stepped off the walkway or if footprints were unrecognizable by the software or manually. Fifteen random strides for each minute of walking taken in the middle of the walkway were analyzed in each walking condition. This was the maximum number of strides that would allow for each subject in each condition and minute of walking to have the same number of strides to be included in gait analysis.

Mean and standard deviation values were calculated for each gait parameter (SL, ST, SW, SV), where variability in each parameter is represented by the coefficient of variation ([standard deviation/mean]  100) expressed as a percentage of the mean. Thus, SLV is stride length variability, SWV is stride width variability, STV is stride time variability, and SVV is stride velocity variability. Increased gait variability has been associated with an increased risk of future falls, even when the systems that contribute to gait variability only show subtle decline31. Thus, gait variability as well as other spatiotemporal gait parameters assessed in the VE may be particularly important for early identification of gait abnormalities in the elderly and for therapy24.

Data was analyzed using mixed design 3-way repeated-measures ANOVAs (rmANOVAs) to assess main and simple interaction effects of group (two levels: elderly and young), condition (four levels: RE, VE, AP, and ML) and time (two levels: minutes 1and 5) on gait variables. For analysis, group was the only between-subject variable. If significant, main and interaction effects were followed with Tukey’s Honest Significant tests. The Lilliefors, and Mauchly’s tests were performed to check the normality and sphericity assumptions of data. The Huynh–Feldt correction was applied if data violated the sphericity condition. Statistical significance was set at p < 0.05.

## Results

Figures 1 and 2 show spatiotemporal gait parameters (SL, SW, ST, SV) and their variabilities (SLV, SWV, STV, SVV) for each group (E and Y), condition (RE, VE, AP, and ML) and minute of walking (1, and 5). Tables 2 and 3 report the results of the statistical analysis.

SL and SV showed a significant main effect of group. Older adults walked with a shorter SL (E: 107.6 ± 21.8 cm; Y: 127.1 ± 19.6 cm, p = 0.011, Fig. 1A) and slower SV (E: 95.3 ± 23.2 cm; Y: 117.8 ± 21.3 cm, p = 0.018, Fig. 1A) than young adults.

All outcomes showed significant main effects of condition (p < 0.034). SWV and SV showed a significant group × condition interaction term (p < 0.035). While walking in the VE, participants took significantly shorter (VE: 117.9 ± 21.7 cm; RE: 126.0 ± 19.1 cm, p = 0.002, Fig. 1A) and wider (VE: 8.1 ± 3.7 cm; RE: 6.8 ± 3.4 cm, p = 0.005, Fig. 1B) strides than in the RE. While walking in the VE, only young adults took slower strides than in the RE (VE: 121.6 ± 14.5 cm/s; RE: 129.2 ± 12.7 cm/s, p = 0.038, Fig. 1D). During AP perturbations, participants took wider (AP: 9.5 ± 3.9 cm; VE: 8.1 ± 3.7 cm, p = 0.031, Fig. 1B) strides than unperturbed walking (i.e., VE).

During ML perturbations, participants took significantly shorter (ML: 102.4 ± 23.0 cm; VE: 117.9 ± 21.7 cm, p < 0.001, Fig. 1A), wider (ML: 11.0 ± 4.2 cm; VE: 8.1 ± 3.7 cm, p < 0.001, Fig. 1B), and slower strides (ML: 96.0 ± 25.9 cm/s; VE: 104.5 ± 23.4 cm/s, p < 0.001, Fig. 1B) with higher SLV (ML: 11.4 ± 8.5%; VE: 4.7 ± 2.8%, p < 0.001, Fig. 2A), STV (ML: 9.4 ± 8.6%; VE: 3.6 ± 2.1%, p < 0.001, Fig. 2C) and SVV (ML: 8.9 ± 6.2%; VE: 6.0 ± 2.8%, p = 0.007, Fig. 2D) than unperturbed walking (i.e., VE).

ML perturbations caused greater reductions in SL (p < 0.001, Fig. 1A) and SV (p = 0.010, Fig. 1D) and increments in SW (p < 0.001, Fig. 2B), SWV (p = 0.014, Fig. 2B), and STV (p = 0.010, Fig. 2C) than AP perturbations.

Surprisingly, SWV only increased significantly in the younger participants during ML perturbations with respect to the VE (p = 0.049) and AP (p = 0.032) conditions. Despite this, no differences in SWV were found between younger and elders in any walking conditions (p > 0.195). SV only significantly decreased in younger participants from VE to ML (p = 0.002). As a result, significant differences in stride velocity were found in the RE and VE conditions (p < 0.007).

Except for ST and SWV, all outcomes exhibited significant main effects of time (p < 0.002). Only SVV showed a significant group × time interaction (p = 0.048). SW, ST, SLV, and STV showed a significant condition × time (p < 0.043). Over time, participants took longer (SL—Minute 1: 112.0 ± 23.5 cm; Minute 5: 118.8 ± 22.0 cm, p < 0.001, Fig. 1A), and faster (SV—Minute 1: 100.8 ± 24.8 cm/s; Minute 5: 107.8 ± 24.8 cm/s, p < 0.001, Fig. 1D) strides with lower SVV (Minute 1: 8.7 ± 7.7%; Minute 5: 6.2 ± 4.0%, p = 0.001, Fig. 2D). Further analysis revealed that: (1) both groups reduced SLV (Minute 1: 14.4 ± 10.3%; Minute 5: 8.2 ± 4.0%, p < 0.001, Fig. 2A) and STV (Minute 1: 12.6 ± 10.7%; Minute 5: 5.9 ± 2.7%, p = 0.001, Fig. 2A) over time only during the ML perturbations; and (2) narrower SW over time only during VE (Minute 1: 8.5 ± 3.6 cm; Minute 5: 7.7 ± 3.8 cm, p = 0.009, Fig. 1B) and AP (SW—Minute 1: 10.5 ± 3.7 cm; Minute 5: 8.5 ± 3.9 cm, p = 0.001, Fig. 1B) conditions.

## Discussion

In this study, we sought to utilize a virtual reality headset to analyze age-related differences in gait patterns during overground walking when exposed to the virtual environment and to visual perturbations, and to understand to what extent older adults are able to modify their gait over time in response to visual perturbations. To the best of our knowledge, this is the first study examining these age-related differences utilizing continuous oscillations of the visual field during overground walking with a virtual reality headset.

As expected, we found that the VE was associated with decreased stride length and stride velocity and increased stride width in both older and younger adults when compared to the RE18,20,23. The introduction of visual perturbations posed a gait challenge in both groups. Specifically, AP perturbations were associated with increased SW and ML perturbations were also associated with decreased SL and SV. Exposure to visual perturbations increased variability in SL, ST, and SV with ML perturbations18,24,25,26. Moreover, ML perturbations caused further reductions in SL and SV and greater increases in SW, SWV, and STV than AP perturbations. We expected this anisotropic response and it is in line with previous studies10,18,22. It is inherently more challenging to maintain walking balance in the ML direction than it is in the AP direction32. Walking requires step-by-step, integrative control as it is passively unstable in the ML direction, whereas balance is passively stabilized with a series of controlled falls in the AP direction10. Moreover, the AP disturbances visually superimposed a velocity to the forward walking velocity whereas ML perturbations superimposed a sideway velocity in a direction orthogonal to walking. Thus, AP perturbation could be less perceivable than ML perturbation22.

Our hypothesis that the spatiotemporal gait characteristics of older adults would be more affected by the VE and visual perturbations than younger adults was not supported by the results. Despite the shorter stride lengths and slower stride speeds than their younger counterparts in all conditions25, the older adults seemed less affected by visual perturbations, especially in the ML direction than the young participants. While older adults maintained a slow walking velocity for all conditions, young participants significantly slowed down during the visual perturbations. Young adults also showed a larger increase in SWV with ML perturbations than old adults, in contrast to previous findings24,25.