Home-Use Evaluation of a Wearable Collision Warning Device for Individuals With Severe Vision Impairments

Key Points

Question

Does a wearable collision warning device reduce collisions of visually impaired individuals in their daily mobility activities?

Findings

In this double-masked randomized clinical trial including 31 blind or severely visually impaired individuals, participants used the device over a period of 4 weeks during daily mobility tasks. The rate of contacts with obstacles reduced significantly by approximately 37% when the device was actively warning for hazards (active condition) compared with when the device was silent (control condition).

Meaning

This study provided evidence that the device can be of value for daily walking mobility of visually impaired individuals, including those who are blind.

This randomized clinical trial evaluates the effect of a collision warning device on the number of contacts experienced by blind and visually impaired people in their daily mobility.

Abstract

Importance

There is scant rigorous evidence about the real-world mobility benefit of electronic mobility aids.

Objective

To evaluate the effect of a collision warning device on the number of contacts experienced by blind and visually impaired people in their daily mobility.

Design, Setting, and Participants

In this double-masked randomized clinical trial, participants used a collision warning device during their daily mobility over a period of 4 weeks. A volunteer sample of 31 independently mobile individuals with severe visual impairments, including total blindness and peripheral visual field restrictions, who used a long cane or guide dog as their habitual mobility aid completed the study. The study was conducted from January 2018 to December 2019.

Interventions

The device automatically detected collision hazards using a chest-mounted video camera. It randomly switched between 2 modes: active mode (intervention condition), where it provided alerts for detected collision threats via 2 vibrotactile wristbands, and silent mode (control condition), where the device still detected collisions but did not provide any warnings to the user. Scene videos along with the collision warning information were recorded by the device. Potential collisions detected by the device were reviewed and scored, including contacts with the hazards, by 2 independent reviewers. Participants and reviewers were masked to the device operation mode.

Main Outcomes and Measures

Rate of contacts per 100 hazards per hour, compared between the 2 device modes within each participant. Modified intention-to-treat analysis was used.

Results

Of the 31 included participants, 18 (58%) were male, and the median (range) age was 61 (25-73) years. A total of 19 participants (61%) had a visual acuity (VA) of light perception or worse, and 28 (90%) reported a long cane as their habitual mobility aid. The median (interquartile range) number of contacts was lower in the active mode compared with silent mode (9.3 [6.6-14.9] vs 13.8 [6.9-24.3]; difference, 4.5; 95% CI, 1.5-10.7; P < .001). Controlling for demographic characteristics, presence of VA better than light perception, and fall history, the rate of contacts significantly reduced in the active mode compared with the silent mode (β = 0.63; 95% CI, 0.54-0.73; P < .001).

Conclusions and Relevance

In this study involving 31 visually impaired participants, the collision warnings were associated with a reduced rate of contacts with obstacles in daily mobility, indicating the potential of the device to augment habitual mobility aids.

Trial Registration

ClinicalTrials.gov Identifier: NCT03057496

Introduction

People with vision impairments are at a higher risk of collisions and falls.^{1,2,3,4,5,6,7} To assist with obstacle detection and avoidance, long canes and guide dogs are commonly used mobility aids. However, the limited scanning range of the long cane and the costs associated with training and keeping a guide dog are major limitations. Although many electronic mobility aids that use electronic sensors and processors have been developed over the past few decades in an attempt to overcome such limitations,^8,9 none have supplanted traditional mobility aids for various reasons.^10,11

Rather than using electronic sensors to replace traditional mobility devices, we took the approach of addressing mobility safety by developing a wearable collision warning device that could be used together with a habitual mobility aid, such as a long cane. A chest-mounted video camera detected impending collisions based on the relative motion between the user and obstacles in the immediate surroundings. It provided alerts via 2 vibrotactile wristbands when the collision risk was high^12,13 based on time to collision rather than distance to objects. In our previous laboratory-based studies, the device helped reduce the number of contacts of individuals with blindness or severe peripheral vision loss in indoor obstacle courses.^14,15

While there have been evaluation studies of various electronic mobility aids conducted in well-controlled environments,^{16,17,18,19,20} there is a serious lack of evidence about the effectiveness of these aids in the real world. Any evidence, positive or negative, regarding the role of electronic devices in augmenting habitual mobility aids could be beneficial in guiding low-vision and mobility rehabilitation practice and future development of electronic mobility aids. We therefore conducted a randomized clinical trial to rigorously evaluate the mobility benefit of the collision warning device during daily usage in naturalistic settings over a period of approximately 4 weeks in blind and visually impaired individuals with independent mobility.

Methods

Study Design

This was a double-masked randomized clinical trial of the collision warning device where the intervention (device gave warnings) and control condition (device did not give warnings) alternated randomly throughout the duration of the trial for each participant. The clinical trial was approved by the Institutional Review Board of Mass Eye and Ear as well as the US Army Medical Research and Materiel Command, Human Research Protection Office. Written informed consent was obtained from all participants. The trial followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline. The trial protocol can be found in Supplement 1.

Participants

The main inclusion criteria were age of 18 years or older, visual acuity worse than 20/200 OU or restricted visual field (tunnel vision less than 40° diameter remaining field), and independently mobile using a mobility aid without a sighted guide. Exclusion criteria were current participation in a mobility training program, diagnosed dementia, or substantial cognitive decline. Eligibility was determined during a screening visit that included assessment of visual acuity (eMethods 2 in Supplement 2), visual fields, and the administration of a brief questionnaire addressing ocular history and mobility and the Short Portable Mental State Questionnaire²¹ for cognitive status.

The study was conducted at the Schepens Eye Research Institute of Mass Eye and Ear, Boston, Massachusetts, from January 2018 to December 2019. Participants were recruited via referrals from the Carroll Center for the Blind, Newton, Massachusetts, and via referrals from practitioners at vision rehabilitation clinics. Additionally, participants were recruited via a pool of volunteers who had participated in prior studies. Study participants were reimbursed for travel and time related to the study visits.

Visit Schedule

Informed consent was obtained at the first visit before screening measures were administered. Participants who met inclusion criteria were then given device training and instruction over 2 training visits. The first training visit focused on the basics of device operation and interpreting its signals while walking. The second training visit included a review of the device operation followed by a walk along an indoor and outdoor route with an orientation and mobility specialist, who advised the participants on using the device together with their habitual mobility aid. After training, participants took the device home and used it at their discretion in their everyday mobility over a period of approximately 4 weeks. They were told to always use the device together with their habitual mobility aid and were informed that the device camera recorded scene videos whenever the device was on. They received weekly telephone calls to check whether there were any problems with the device. Device repairs were completed as needed. At the end of the 4 weeks, they returned the device and completed a postuse questionnaire (eResults 4 in Supplement 2).

Intervention, Randomization, and Masking

The collision warning device (eMethods 1 and eFigure 1 in Supplement 2) provided active warnings via the vibrotactile wristbands in the intervention condition (active mode). In the control condition (silent mode), the device worked as usual, except the wristbands did not vibrate, so no feedback was given to the user. The chest-mounted camera of the device recorded videos of the scene in both modes. The videos also contained embedded information about the operating mode (active or silent) and the detected collision warning hazard (eFigure 2 in Supplement 2). The device was programmed to switch intermittently between the active and silent modes on a randomized schedule that was not disclosed to the participants (eMethods 2 in Supplement 2). Once initiated, the switching schedule was not traceable by the research team. Duration of device operation in a given mode was approximately 30 minutes before it switched to the other mode.

Participants were masked to device operating mode. They were informed that it was a prototype device and might not provide warnings in some situations. They were not explicitly informed of the presence of the silent mode and mode switching. Research staff who reviewed the recorded videos were also masked to device operating mode. Information about the operating mode was removed so they only reviewed videos with scene information and an annotation to indicate the object for which the device had issued collision warnings.

Review of Collision Warning Events

A collision event was defined as a group of collision warning instances issued within a window of 2 seconds. The video clip for each reviewed event included 2 seconds before and after the event. Each event was independently reviewed by 2 masked reviewers (V.B. and M.M.) in a hierarchical manner, judged in the following order: (1) whether the event was associated with a true collision hazard, (2) whether there was any kind of unintentional collision contact with obstacles, and (3) in case of collision contact, whether there was a body contact. Collision contacts were visually judged based on walking speed, nearness of the object in the camera, sudden pause or sharp change in walking direction, and shaking of the camera due to the impact. They included cane contacts and body contacts. Normal intentional cane probing contacts for environmental awareness were not counted as contacts for the purposes of this study. Disagreements between the reviewers were resolved via consensus. Details regarding the methods for event detail annotation, definitions of various categories, and the overall review process have been previously published²² (eMethods 3 in Supplement 2).

For each participant, the events were randomly sampled for review. Some participants had large numbers of events (for example, more than 1500). To keep review to manageable levels, the number of events reviewed were capped at 400 per participant per operating condition (eMethods 3 in Supplement 2). When the total recorded events were below this threshold limit, all events were reviewed for that participant.

Outcomes

The primary outcome measure was the rate of contacts per 100 true hazards (reviewed) per hour. The exposure variable took into account both duration of use and number of true hazards encountered, since depending on the surrounding environment, some participants might walk for longer but encounter fewer hazards while others might encounter more hazards in a shorter duration. The rate of body contacts was included as a secondary rather than a primary outcome because it was sometimes challenging to unambiguously determine whether the participant’s body (part) bumped into the obstacle, due to limitations in the field of view of the scene video. Subjective ratings from the postuse device questionnaire were included as secondary outcomes (eResults 4 in Supplement 2). Primary and secondary outcomes were prespecified in the statistical analysis plan.

Sample Size

The sample size calculation was based on the findings of our previous evaluation of the collision warning device in an obstacle course,¹⁴ where the mean (SD) effect size in participants with tunnel vision (n = 13) was 0.4 (0.38; pairwise difference between with-device and without-device conditions). Assuming the effect size is halved (0.2) to account for real-world mobility, the same standard deviation of change, α of .05, and power of 0.8, we needed a sample size of 31. Further accounting for 30% attrition rate in study participants, we estimated a final sample size of 42.

Statistical Analysis

Since the number of contacts represented overinflated count data, negative binomial regression models—considered to be more appropriate than Poisson regression models for such distributions²³—were used for evaluating the association between rate of contacts and device operating condition (active vs silent), the main predictor of interest. In actual implementation of the regression model (eMethods 4 in Supplement 2), the outcome was set as the number of contacts, while the exposure (accounting for duration of device use and the number of true hazards) was the offset, and the estimated regression coefficients were expressed as the rate ratio (β). The role of other possible predictors that could potentially affect the rate of contacts was analyzed. These predictors came from 3 broad categories: demographic characteristics, vision characteristics, and mobility characteristics. Each predictor was evaluated individually in a generalized mixed-effects modeling framework (because we wanted to estimate within-participant effects) that also included device mode condition. Interactions between categorical predictors and the device mode were examined. Potentially significant predictors and possible confounders (causing more than 10% change in β for the device condition predictor) were selected to be included in the final model for rate of contacts. For pairwise comparison (paired differences), P values were computed using Wilcoxon signed rank test. When reporting regression results, P values were obtained from the z scores of the coefficients. Two-sided P values less than .05 were considered statistically significant, and no adjustments were made for multiple analyses. Statistical analyses were conducted using R version 4.0.4 (The R Foundation).

Results

In total, 49 individuals were enrolled, of which 36 finished training, took the device home, and completed the home-use trial (Figure). However, because of device malfunctions, data were not recorded at all for 3 participants and for only one of the conditions in 2 participants. Thus, a total of 368 hours of walking video data from 31 participants was available for the main analysis of rate of contacts. Of the 31 included participants, 18 (58%) were male, and the median (range) age was 61 (25-73) years. Three of these 31 participants had better vision than specified in the original inclusion criteria but were nevertheless included since they used a long cane, reported mobility difficulties, and their collision event data were within the range of the other participants (eResults 2 and eTables 3 and 4 in Supplement 2).

Figure. CONSORT Flow Diagram.

^aTwo participants who enrolled in an earlier pilot study followed all protocol procedures for this trial, and their device data are included in analyses.

^bOne participant completed data collection with the device but left the trial before the final feedback questionnaire.

Of the 31 participants included in the analysis, 19 (61%) had a visual acuity of light perception only or worse (Table 1) and 6 (19%) had visual acuity better than 20/200 (eResults 3 and eTable 5 in Supplement 2). Onset of visual impairment was at birth or during early childhood for 17 participants (55%), with a median (range) duration of 38 (5-73) years of vision loss. A vast majority of participants used a long cane as their habitual mobility aid (28 [90%]) and had previously received orientation and mobility training (29 [94%]).

Table 1. Characteristics of 31 Participants Who Were Included in the Analysis of Rate of Contacts.

Characteristic	No. (%)
Demographic characteristics
Age, median (IQR; range), y	61 (48.0-65.5; 25-73)
Sex
Female	13 (42)
Male	18 (58)
Self-reported race
White	21 (68)
Black	8 (26)
Asian	1 (3)
Multiple	1 (3)
Vision characteristics
Visual acuity
<LP	11 (35)
LP only	8 (26)
Between LP and 20/200	6 (19)
Between 20/200 and 20/100a	3 (10)
> 20/100b	3 (10)
Visual acuity better than LP
No	19 (61)
Yes	12 (39)
Disease conditions
Retinitis pigmentosa	8 (26)
Retinopathy of prematurity	5 (16)
Other	15 (48)
Unknown	3 (10)
Visual impairment onset
Birthc	17 (55)
Adultd	14 (45)
Duration of vision loss, median (IQR; range), y	38 (18-62; 5-73)
Mobility characteristics
Habitual mobility aidse
Long cane	28 (90)
Guide dog	3 (10)
Mobility training
Yes	29 (94)
No	2 (6)
Any falls in the past 1 y, No./total No. (%)
No	19/29 (66)
Yes	10/29 (34)
No. of falls in the past 1 y
0	19 (68)
1	3 (10)
2	3 (10)
3	4 (14)

Abbreviations: IQR, interquartile range; LP, light perception.

^aIncludes 2 participants who met the visual field criteria, having tunnel vision with remaining fields of 20° horizontal by 15° vertical, and 17° horizontal by 10° vertical diameter, and 1 participant with no peripheral field loss who did not meet the criteria (eResults 2 in Supplement 2).

^bIncludes 1 participant who met the visual field criteria, having tunnel vision with remaining field of 25° horizontal by 25° vertical diameter, and 2 participants did not meet the visual field criteria—one with a remaining field larger than 40° diameter and one with a ring scotoma (eResults 2 in Supplement 2).

^cOnset for 1 participant was at age 3 years.

^dMedian age of onset, 38.5 years.

^eHabitual mobility aid self-reported during screening. One of the habitual guide dog users, who was also trained for cane use, walked with a long cane during the trial.

The overall hours of device use and the number of detected events were higher in the active than the silent mode, but after normalization for exposure, the rate of detected events per hour was statistically matched (Table 2). The rate of reviewed events per hour was similar in the 2 modes. The number of valid events and true hazards were higher in the active mode. However, the rates of valid events and true hazards per hour of reviewed use and as the proportion of the number of reviewed events were statistically matched between the 2 modes.

Table 2. Summary of Collision Warning Event Data Recorded by the Device for the 31 Study Participants.

Measure	Total, No. (%)	Median (IQR)
		Active mode	Silent mode	P valuea
Overall data
Total time of device use, h	368.17	4.18 (2.91-10.27)	2.81 (1.89-7.75)	<.001
All detected events, No.	28 733	330 (150-721)	193 (92-666)	<.001
Detected events per h of device use	78.0	68.6 (59.1-87.8)	78.3 (48.8-97.8)	.22
Reviewed datab
Reviewed time of use, h	214.7 (58.3)	3.77 (2.38-5.24)	2.48 (1.53-4.71)	<.001
Reviewed events	16 341 (56.9)	332 (152-427)	193 (92-403)	<.001
Reviewed events per h of device use	76.11	78.8 (60.6-95.0)	78.4 (55.3-108.9)	.20
Valid eventsc
Valid events	14 123	278 (122-374)	183 (75-325)	.001
Valid events per h of reviewed use	65.78	70.8 (42.2-85.6)	68.7 (41.8-98.4)	.35
Proportion of valid events to reviewed events	0.86	0.92 (0.86-0.96)	0.90 (0.80-0.97)	.29
True hazardsd
True hazards	4067	60 (32-118)	46 (19-84)	<.001
True hazards per h of reviewed use	18.94	21.89 (9.74-28.67)	17.79 (9.86-30.65)	.34
Proportion of true hazards to reviewed events	0.25	0.24 (0.18-0.33)	0.26 (0.17-0.35)	.61
Contactse
Body contacts	171 (1.05)	2 (1-5)	2 (1-4)	.29
Body contacts per 100 true hazards per h	0.02	1.09 (0.27-1.97)	1.13 (0.45-3.08)	.22
All contacts	1454 (8.9)	22 (11-41)	14 (7-32)	.19
All contacts per 100 true hazards per h	0.17	9.26 (6.56-14.97)	13.79 (7.28-24.30)	<.001

Abbreviation: IQR, interquartile range.

^aWilcoxon signed rank test.

^bReviewed events were capped for 12 participants because they had more than 400 events in either mode.

^cValid event is defined as an event when the camera view was unobstructed and device operation was as expected.

^dTrue hazard is defined as a valid event associated with a mobility hazard that would have resulted in a collision if the trajectory of motion between the user (camera) and the entity was maintained.

^eContact with a mobility hazard is defined as unintentionally bumping into an object or making contact with the body or with long cane. Contacts do not include normal probing contacts made to be aware of the environment.

The median (interquartile range) number of contacts reduced in the active mode (active, 9.3 [6.6-14.9]; silent, 13.8 [6.9-24.3]; difference, 4.5; 95% CI, 1.5-10.7; P < .001), and the same was found via regression analysis with device condition as the only predictor (β = 0.66; 95% CI, 0.56-0.78; P < .001). Results of the analyses for the individual predictors of rate of contacts modeled together with the device condition are shown in Table 3. Presence of visual acuity better than light perception was found to be a significant predictor of rate of contacts (β = 0.37; 95% CI, 0.21-0.66; P = .001). Interaction between the device operating mode and fall history (whether the participant self-reported a fall in the 1 year prior to the trial) was also found to be significant (β = 1.47; 95% CI, 1.05-2.05; P = .02). Visual impairment onset (either at birth/childhood or during adulthood) was not a significant predictor of rate of contacts, but its inclusion led to a large change in the rate ratio for the device condition relative to the initial value without including any other covariates (0.66 [95% CI, 0.56-0.78] to 0.73 [95% CI, 0.57-0.93], a change of more than 10%). Therefore, these predictors (visual acuity, impairment onset, and fall history), in addition to the device condition, were selected to be analyzed together in the final model (Table 4).

Table 3. Possible Factors Affecting Rate of Contacts, Each Examined Together With Device Condition.

Covariates examined			Mixed-effects model estimate
Variable name	Variable type	Category	Covariate		Device condition		Interaction
			Rate ratio	P value	Rate ratio	P value	Rate ratio	P value
Device condition only	Binary (active or silent mode)	Active mode	NA	NA	0.66 (0.56-0.78)	<.001	NA	NA
Visual acuity >LP	Binary (present or absent)	Presence	0.37 (0.21-0.66)	.001	0.65 (0.52-0.82)	<.001	1.01 (0.72-1.43)	.94
Duration of vision loss	Continuous (y)	Per decade	0.98 (0.85-1.13)	.78	0.66 (0.56-0.78)	<.001	NA	NA
Impairment onset	Binary (birth or adult onset)	Birth	1.27 (0.66-2.45)	.48	0.73 (0.57-0.93)	.01	0.83 (0.60-1.15)	.26
Age	Continuous (y)	Per decade	0.92 (0.72-1.18)	.50	0.66 (0.56-0.78)	<.001	NA	NA
Sex	Binary (female or male)	Male	1.76 (0.93-3.34)	.08	0.71 (0.56-0.92)	.008	0.86 (0.62-1.21)	.39
Habitual mobility aid	Binary (cane or dog)	Dog	1.30 (0.41-4.14)	.65	0.65 (0.54-0.77)	<.001	1.23 (0.67-2.24)	.51
Fall history	Binary (falls or no falls)	Did fall	0.72 (0.34-1.49)	.37	0.63 (0.54-0.73)	<.001	1.47 (1.05-2.05)	.02
No. of falls	Continuous (count)	Per fall	0.97 (0.70-1.33)	.82	0.68 (0.58-0.79)	<.001	NA	NA

Abbreviations: LP, light perception; NA, not applicable.

Table 4. Final Model for Rate of Contacts After Including Significant Predictors and Potential Confounders.

Variable	Variable type	Category	Rate ratio (95% CI)	P value
Intercept	NA	NA	0.23 (0.13-0.38)	<.001
Device operating mode	Binary (active or silent mode)	Active mode	0.63 (0.54-0.73)	<.001
Visual acuity >LP	Binary (present or absent)	Presence	0.38 (0.21-0.66)	.001
Impairment onset	Binary (birth or adult onset)	Birth	1.06 (0.60-1.86)	.84
Fall history	Binary (did fall or no falls)	Did fall	0.70 (0.37-1.31)	.26
Device operating mode × fall history	NA	Active mode × did fall	1.48 (1.06-2.07)	.02

Abbreviations: LP, light perception; NA, not applicable.

Presence of visual acuity better than light perception reduced the rate of contacts significantly by about 62% in the final model (β = 0.38; 95% CI, 0.21-0.66; P = .001), but visual impairment onset and fall history were not significantly associated with the rate of contacts (Table 4). The interaction between fall history and device mode was significant (β = 1.48; 95% CI, 1.06-2.07, P = .02). For participants who did not report recent falls, the rate ratio between the active and silent device condition was 0.63, whereas for participants with falls, it was 0.93. Controlling for all these factors, the rate of contacts decreased significantly in the active mode compared with the silent mode by about 37% (β = 0.63; 95% CI, 0.54-0.73; P < .001).

With device condition as the only predictor, the rate of body contacts reduced in the active mode (β = 0.63; 95% CI, 0.42-0.94; P = .02) (eResults 1 and eTable 1 in Supplement 2). In the final model (controlling for visual acuity better than light perception, sex, and fall history), the rate of body contacts decreased in the active mode compared with the silent mode by about 36% (β = 0.64; 95% CI, 0.41-0.99; P = .04) (eResults 1 and eTable 2 in Supplement 2).

Average qualitative ratings from the postdevice use questionnaire were mostly positive and are summarized in eTable 6 in Supplement 2. A total of 8 minor adverse events (minor contact/brushing against an object when walking) were reported by 6 participants during the home-use trial. There were no serious adverse events.

Discussion

This study evaluated the effect of an electronic mobility aid on collision avoidance for blind and visually impaired individuals during their daily mobility activities in naturalistic settings. When the device was in the active mode (gave warnings), the rate of collision contacts (body contacts and cane contacts) and body contacts alone reduced significantly by about 37% and 36%, respectively, after controlling for various other confounding factors. These findings demonstrate a clear mobility benefit of using the device. Reducing the overall number of contacts may reduce the risk of injurious collisions and falls.

The results also showed that presence of visual acuity better than light perception was associated with a lower rate of contacts overall irrespective of the device operating mode, which makes sense as those with more severe visual impairments are more likely to make contact with obstacles not just with their bodies but also with their mobility aids. A significant interaction between the device operating mode and fall history indicated that participants who did not report any recent falls benefitted more from the device (difference in the rate of contacts between the silent and active modes was higher) than people who did report falling. This might be because people who reported falls were more likely to be cautious than those who did not report falls, as the former had relatively lower rate of contacts than the latter in the silent mode (Table 4).

Limitations

There are some limitations of this trial and the presented data. First, the results are based only on the review and analysis of events detected by the device. In reality, it is expected that the device might miss detection of some true hazards over the course of its use, so participants may have experienced more actual collision incidents than were recorded. Because it is not practically feasible to review every second of the hundreds of hours of videos (review takes much longer than the actual video duration), we could only review episodes identified by the device as potential collision hazards. In other words, our samples are those episodes. By setting a relatively liberal threshold for issuing collision warnings, the recorded episodes included more than just true hazards so that the sampling was not too limited, although this caused some false alarms. Another limitation was the possibility that if participants collided with an obstacle in the silent mode, they might become more cautious in their mobility behaviors until they received warnings again when the device went to active mode. If this happened, however, it would reduce the difference in collision rates between the active and silent modes and would not improperly drive the difference in the direction of being more significant.

Conclusions

In conclusion, our clinical trial provides evidence of the benefit of a collision warning device in reducing contacts in real-world mobility. Our results directly address the serious lack of evidence for the effectiveness of vision aids in habitual mobility identified in a number of reviews.^24,25,26 Few prior studies attempted to obtain natural mobility data in a visually impaired population,^{23,27,28,29,30} and to our knowledge, none have captured and processed video data on collision incidents. In the context of evaluating mobility aids for walking, this study has many novel methodological features, including the video review methods to characterize real-world mobility collision incidents and the use of the randomized switching of intervention and control conditions within participants, which allowed us to gather actual usage and collision incident data instead of relying on self-reported outcomes. These rigorous study design elements helped strengthen the validity of our findings about the positive effect of the collision warning device on the daily mobility of blind and visually impaired individuals.

Supplement 1.

Trial protocol.

Supplement 2.

eMethods 1. The wearable collision warning device.

eMethods 2. Randomization and study procedures.

eMethods 3. Event review details.

eMethods 4. Details of the regression model.

eFigure 1. Wearable collision warning device.

eFigure 2. Snapshot of the data recorded by the collision warning device.

eResults 1. Results of analysis of body contacts.

eResults 2. Study participants failing to meet original vision criteria.

eResults 3. Study participants with visual acuity better than 20/200.

eResults 4. Qualitative ratings for the device (postuse questionnaire).

eTable 1. Possible factors affecting rate of body contacts by device condition.

eTable 2. Final model for rate of body contacts after including significant predictors and potential confounders.

eTable 3. Comparison of the rate of contacts and body contacts for 3 participants who did not meet original inclusion criteria.

eTable 4. Model for all contacts after excluding 3 participants who did not meet original inclusion criteria.

eTable 5. Vision characteristics of the 6 participants with visual acuity better than 20/200.

eTable 6. Qualitative ratings for the collision warning device by the 30 trial participants.

eReferences.

Supplement 3.

Data sharing statement.