Naturalistic Driving Techniques and Association of Visual Risk Factors With At-Fault Crashes and Near Crashes by Older Drivers With Vision Impairment

Key Points

Question

What are visual risk factors associated with at-fault crashes and near crashes by older drivers with age-related eye conditions objectively assessed by naturalistic driving methods using video-confirmed events?

Findings

This cohort study of 154 drivers 70 years or older found that vision impairment in contrast sensitivity, processing speed, and motion perception at baseline was moderately associated with the rate of at-fault crashes and near crashes during the subsequent 6 months of driving.

Meaning

Motor vehicle crash reports used in the study of driver safety are frequently biased and incomplete; this study avoids these shortcomings by documenting vision impairment characteristics associated with collision risk among older drivers with age-related eye conditions by using naturalistic driving methods.

Abstract

Importance

Government motor vehicle crash reports used in the study of driver safety can be biased and incomplete. Naturalistic driving methods using in-vehicle instrumentation have been developed in recent years to objectively measure crashes and near crashes as they occur on the road using video and vehicle kinematic data.

Objective

To examine visual risk factors associated with at-fault crashes and near crashes among older drivers, most of whom have age-related eye conditions associated with vision impairment.

Design, Setting, and Participants

This prospective cohort study was conducted at an academic ophthalmology clinic from January 1, 2015, to November 10, 2018, among 154 adults 70 years of age or older who were legally licensed in Alabama and who reported currently driving at least 4 days per week; 137 of 151 participants (90.7%) had an age-related eye condition in at least 1 eye. Drivers participated in a baseline visual function assessment followed by installation of a data acquisition system recording multichannel synchronized video and vehicle kinematics in their personal vehicle. Drivers were instructed to drive for 6 months as they normally would during everyday life. Statistical analysis was performed from June 15 to September 15, 2020.

Main Outcomes and Measures

The rate of combined incident at-fault crashes and near crashes, defined by the number of events and the number of miles driven.

Results

The sample consisted of 154 drivers (85 men [55.2%]; mean [SD] age, 79.3 [5.1] years). Visual functions associated with crash and near-crash involvement included impaired contrast sensitivity (rate ratio [RR], 2.7; 95% CI, 1.3-5.5), moderate (RR, 2.3; 95% CI, 1.1-4.9) and severe (RR, 5.0; 95% CI, 2.2-11.7) slowing in visual processing speed, and elevated motion perception thresholds for a drifting grating (RR, 1.9; 95% CI, 1.1-3.5). Those with impaired peripheral visual field sensitivity had increased rates of crashes and near crashes (RR, 1.8; 95% CI, 1.0-3.3); however, this finding was not statistically significant (P = .07).

Conclusions and Relevance

With the use of naturalistic driving methods in which crashes and near crashes involving older drivers are objectively measured as they occur on the road, associations have been identified between impaired contrast sensitivity, slowed visual processing speed, and impaired motion perception and an increased rate of a combined total of at-fault crashes and near crashes.

This cohort study examines visual risk factors associated with at-fault crashes and near crashes in drivers 70 years or older, most of whom have age-related eye conditions associated with vision impairment.

Introduction

Older adults have the highest prevalence of vision impairment in the US.¹ Epidemiologic studies on older drivers have examined the visual requirements for driver safety, focusing on the types of vision impairment associated with increased collision risk.^2,3,4,5,6 The primary outcome in many studies is crash involvement, defined by motor vehicle crash reports providing details on the collision (eg, driver demographic characteristics, time, and place). These reports are completed by a police officer who attends the scene after the crash has been reported. The police officer does not actually see the crash take place but collects information about it from witnesses and by observing the vehicle(s) and roadway after the crash. Although a motor vehicle crash report is an objective government document, it is prone to bias from the police officer and witnesses.⁷ Another limitation is that some crashes are not reported to the police and go uncounted in assessing risk factors associated with crash involvement.⁸

During the past 15 to 20 years, a novel method for studying driver safety has been developed—naturalistic driving using instrumented vehicles.^9,10,11 Multichannel video cameras and sensors are installed in participants’ vehicles recording several data streams. Participant drivers go about their everyday driving for an extended time period (months or years), creating an objective driving record that can be used to identify crashes after the study. In computing the crash rate (crashes per mile driven), the actual mileage of the car is used, rather than the driver’s self-reported mileage, which is unreliable.^12,13,14 Near crashes can also be identified, occurring more commonly than crashes with similar associated factors, which allows them to be used as crash surrogates.^9,15 In our novel prospective cohort study, we studied the association between visual function at baseline in drivers 70 years of age or older and at-fault crash and near-crash involvement during the subsequent 6 months. We specifically focused on older adults, most with age-related eye conditions that engender vision impairment, with the goal of identifying the visual risk factors associated with objectively defined at-fault crashes and near crashes.

Methods

This study, conducted from January 1, 2015, to November 10, 2018, is based on the Alabama VIP Older Driver Study using naturalistic driving to study driver safety and performance among older adults.¹⁶ Institutional review board approval was obtained from the University of Alabama at Birmingham (UAB) and Virginia Tech. Written informed consent was obtained from the participants. The target population was older adults who recently had an appointment at the Callahan Eye Hospital Clinics at UAB. Our goal was to recruit approximately 90% of our sample having age-related eye conditions, including age-related macular degeneration, cataract, diabetic retinopathy or macular edema, glaucoma, and other conditions that have the potential for impairing vision. Our goal was to recruit the other 10% from persons who did not have age-related eye conditions. Eye diagnoses were obtained from the electronic health record. Inclusion criteria at baseline were aged 70 years or older, legally licensed in Alabama, reported driving 4 days or more per week, owned a motor vehicle, and were willing to have a data acquisition system (DAS) installed in their vehicle.

Study visits took place at the Clinical Research Unit at Callahan Eye Hospital. Potential participants were contacted by letter and received a telephone call to schedule an in-person screening visit. This visit was to examine the person’s vehicle to assess its suitability for DAS installation and to discuss installation details and the study process. If the vehicle was suitable for installation, a second visit was scheduled within 1 to 2 weeks for installation and completion of the baseline protocol. Enrollees were compensated for their participation.

At baseline, demographic variables were obtained through interview. Several visual and visuocognitive tests were conducted under binocular conditions unless otherwise noted. Tests were those associated with crash involvement using motor vehicle crash reports.^3,5,17,18,19 Visual acuity testing was included because it is a screening method for driving licensure in all states. Distance visual acuity with habitual distance correction while driving (if any) was measured using the electronic visual acuity tester²⁰ and expressed as logMAR. Impaired acuity was defined as worse than 20/40. Contrast sensitivity was assessed using the Pelli-Robson chart²¹ and scored as log sensitivity using the letter-by-letter method.²² Impaired contrast sensitivity was defined as worse than 1.5 log sensitivity.²³ Visual processing speed was assessed by the Useful Field of View subtest 2,²⁴ measuring the amount of time needed (in milliseconds) to discriminate 2 targets in central vision and simultaneously localize a peripheral target at 10° eccentricity in any of 8 radial directions. Processing speeds from 150 to 350 milliseconds were defined as moderately impaired, and those greater than 350 milliseconds were defined as severely impaired.⁶ Visual field sensitivity was assessed for each eye separately using a custom test for the Humphrey Field Analyzer Model II-I developed for a previous study.⁵ Visual field sensitivity was assessed monocularly so that the fixation tracking system could be used. Light sensitivity was measured using the full-threshold procedure and white stimulus-size III targets presented at 20 visual field locations selected to be those that fall within the visual field area when a driver gazes through a vehicle’s windshield or to the vehicle’s dashboard.^5,25 Because driving is performed using both eyes together, the monocular fields from each participant were combined to form a binocular field of 21 points spanning 60° to the right and left, 15° to the superior field, and 30° to the inferior field.²⁶ The sensitivity at each test location was defined by the more sensitive point (higher value in decibel units) of the 2 eyes (target grid published previously).^5,27 The visual field was divided into specific regions, and those in the lowest quartile of sensitivity were defined as having impaired vision.⁵ Motion perception was assessed using a drifting Gabor test.²⁸ This test presented a 3-cycle per degree vertical sinusoidal grating filtered with a gaussian envelope. Participants were asked to identify the grating’s drifting direction (right vs left), with the drift rate (in units of hertz) varying during a 2-down and 1-up staircase with 8 reversals. Thresholds were defined as the mean of the last 6 reversals. Impaired motion perception was defined as thresholds worse than the median value.¹⁹

General cognitive status, depressive symptoms, and general health were evaluated. The Mini-Mental State Examination (MMSE)²⁹ was used to assess general cognitive status, and the Center for Epidemiological Studies–Depression (CES-D) scale³⁰ was used to assess depressive symptoms. A general health condition checklist asked about medical problems in 17 areas (eg, heart disease and diabetes).³¹

During the baseline visit, the DAS was installed in the participant’s vehicle by trained personnel at an automotive garage. The DAS was developed by the Virginia Tech Transportation Institute (VTTI) and is identical to that used in the Strategic Highway Research Program 2 study.¹⁰ The following components of the system were mounted into the vehicle: 5-channel video cameras of the roadway environment and driver (views of 83° forward, 55° left side, 99° rear, 99° front seat passenger and cabin snapshot, and 55° of steering wheel and driver controls), accelerometers, a global positioning system, and a hard drive installed in the trunk or within the vehicle. The vehicle network provided accelerator position, brake actuation, and speed. After installation was complete, the driver was informed that the DAS would not impact the operation of the vehicle; their experience in driving and controlling the vehicle would be identical to when the DAS was absent. The DAS started 30 seconds after the ignition was turned on and powered down when the ignition was off. Participants were asked to drive their instrumented vehicle as they would normally for the 6-month duration of the study. During this period, the “health” of the DAS was checked remotely by VTTI through cell phone networking; if there was a malfunction, VTTI notified UAB staff, who then scheduled a return visit for maintenance. When the DAS data drive approached being full, the drive was retrieved by UAB staff who transmitted the recorded data to VTTI for processing; a new data drive was then installed in the vehicle. The DAS was deinstalled at the end of the 6-month follow-up period, restoring the vehicle to its prestudy condition.

The outcome of interest was based on the total number of at-fault crashes and at-fault near crashes combined. Given that at-fault crashes and near crashes are significantly associated with each other,³² combining the events into a single outcome was appropriate. Previous work has shown that the roadway circumstances underlying crashes and near crashes are similar.¹⁵ Trained VTTI analysts identified at-fault crashes and near crashes from the data streams; they were masked to the participant’s status on variables collected at UAB. Crashes were defined as events in which the participant’s vehicle had contact with any object (other vehicles, pedestrians, cyclists, or animals), at any speed, including nonpremeditated departures from the roadway where at least 1 tire leaves the paved or intended travel surface of the road.³² Near crashes were defined as any circumstance that required a rapid evasive maneuver by the participant’s vehicle or any other vehicle, pedestrian, cyclist, or animal to avoid a crash.³² Methods for crash and near-crash identification were identical to those used in the Strategic Highway Research Program 2 study.^10,32 Analysts first identified trip files in which the driver was the consenting participant, destroying files from nonconsenting drivers.³³ Software with defined triggers was used to detect sudden and extreme changes recorded by physical sensors and the vehicle network (eg, longitudinal acceleration and deceleration, lateral acceleration, yaw rate, and advanced safety system activation) to identify potential safety-critical events. Analysts then reviewed a temporal window of video around the event to confirm that it was indeed a crash or near crash. Crashes and near crashes could also be detected by analysts while viewing individual video files of trips. Participants were also asked to telephone the UAB coordinator when a crash occurred, and then the crash was communicated to VTTI for confirmation by the DAS data streams. Motor vehicle crash reports from governmental jurisdictions were not used to identify crashes.

Only crashes and near crashes deemed at fault by raters were considered. These events were coded as “at fault” when there was demonstrable video evidence that the driver committed an error leading to the event. A system for establishing intrarater and interrater coding agreement was established previously at VTTI.³² The mean intrarater agreement was 91% (range, 73%-94%); agreement between each rater against an expert rater was 88% for crash and near-crash events.

Statistical Analysis

Statistical analysis was performed from June 15 to September 15, 2020. Crash and near-crash data identified by VTTI analysts were linked to data collected at UAB using participant numbers. The primary outcome was the rate of combined incidents, defined by the total number of at-fault crashes and near crashes during the 6-month follow-up period and number of miles driven for each participant. We did not focus on all events regardless of fault because at-fault events are more closely associated with drivers’ visual characteristics than all events.³⁴ Associations between visual function and at-fault crashes and near crashes were assessed using rate ratios (RRs) and 95% CIs from Poisson regression models scaled for deviance. An offset for the natural log of miles driven was included. All models were adjusted for age. Visual function measures were categorized because the assumption of a linear association was not met. The level of significance was set at P < .05 (2-tailed). All analyses were completed in SAS, version 9.4 (SAS Institute Inc).

Results

In all, 321 drivers met eligibility criteria after completing a telephone screening; 280 participated in an in-person screening visit. Of these 280 persons, 162 met inclusion criteria and passed the vehicle screening. A total of 159 drivers enrolled in the study, representing a 49.5% participation rate. Shortly after installation of the DAS, garage personnel determined that the vehicles for 5 drivers prevented proper DAS or vehicle operation or presented driver inconvenience (radio static); therefore, the analysis sample was reduced to 154 drivers. Of these drivers, 146 (94.8%) participated for the 6-month follow-up period, for a mean (SD) 177 (6) days of follow-up. Eight drivers (5.1%) left the study before 6 months for the following reasons: vehicle needed repairs with implementation delayed (n = 4), a serious medical issue prevented driving (n = 2), death (n = 1), and newly developed DAS problems (n = 1). These 8 drivers participated for a mean (SD) of 105 (40) days. Participants drove a mean (SD) of 3846 (4375) miles (6189.5 [7040.9] km). There were 26 at-fault crashes and 55 at-fault near crashes, totaling 81 events. Of the 81 events, most crashes and near crashes (71 [87.7%]) occurred during the day. In terms of the number of vehicles involved, 55 crashes or near crashes (67.9%) involved other vehicles.

Approximately half the sample of drivers were in their 70s (85 [55.2%]), with the rest in their 80s (66 [42.9%]) and 90s (3 [1.9%]) (Table 1). More than half of the participants were men (85 [55.2%]). In terms of race/ethnicity, 28 participants (18.2%) were Black, and 126 (81.8%) were White, representative of the racial/ethnic makeup of north central Alabama. Cognitive status (MMSE scores) was normal in 152 participants (98.7%), and 151 participants (98.1%) did not exhibit depression. There was a range of chronic medical conditions. The common age-related eye conditions (age-related macular degeneration, cataract, diabetic retinopathy or macular edema, and primary open-angle glaucoma) and other ocular conditions were represented in the sample; 137 of 151 participants (90.7%) had an ocular condition in at least 1 eye, with the balance of the sample having normal eye health.

Table 1. Demographic and Health Characteristics of the Sample.

Characteristic	Participants, No. (%) (N = 154)
Age group, y
70-79	85 (55.2)
80-89	66 (42.9)
90-99	3 (1.9)
Sex
Women	69 (44.8)
Men	85 (55.2)
Race/ethnicity
Black	28 (18.2)
White	126 (81.8)
Educational level
Less than high school graduate	6 (3.9)
High school graduate	76 (49.4)
College graduate	61 (39.6)
Professional or graduate school	11 (7.1)
MMSE score
≥24 (Not impaired)	152 (98.7)
<24 (Impaired)	2 (1.3)
Medical conditions, No.
0-1	20 (13.0)
2-3	44 (28.6)
4-5	60 (39.0)
≥6	30 (19.5)
CES-D score
≤16 (Not depressed)	151 (98.1)
>16 (Depressed)	3 (1.9)
Eye diagnoses (1 or both eyes), No./total No. (%)a
Age-related macular degeneration	28/151 (18.5)
Cataract	63/151 (41.7)
Diabetic retinopathy or macular edema	10/151 (6.6)
Primary open-angle glaucoma	43/151 (28.5)
Pseudophakia	93/151 (61.6)
Otherb	68/151 (45.0)
Age, mean (SD), y	79.3 (5.1)
Educational level, mean (SD), y	14.9 (2.8)
MMSE, mean (SD), total score	27.7 (1.8)
CES-D, mean (SD), total score	3.8 (4.2)

Abbreviations: CES-D, Center for Epidemiological Studies–Depression; MMSE, Mini-Mental State Examination.

^aParticipants could have more than 1 diagnosis. Medical records could not be located for 3 participants.

^bExamples of other conditions include dry eye disease, hypertensive retinopathy, Fuchs dystrophy, lattice degeneration, macular cyst hole, macular pucker, ocular prosthesis, and retinal tear.

Table 2 summarizes the visual function status in the sample. Three drivers (1.9%) had visual acuity worse than 20/40, and 17 (11%) had contrast sensitivity worse than 1.5 log sensitivity. Almost half of participants had slowed visual processing speed, with 47 of 125 (37.6%) having moderate slowing and 15 of 125 (12.0%) having severe slowing.³⁵ The visual field sensitivity for the overall field and by region illustrates the lowest decibel quartiles; the median value (0.14 Hz) is provided for the drifting grating motion test.

Table 2. Visual Function at Baseline in the Sample.

Visual function	Participants, No. (%) (N = 154)
Visual acuity, Snellen
20/40 or better	151 (98.1)
Worse than 20/40	3 (1.9)
Contrast sensitivity, log sensitivity
≥1.5 (Better)	137 (89.0)
<1.5 (Worse)	17 (11.0)
UFOV subtest 2, msa
<150 (Better)	63 (50.4)
150-350	47 (37.6)
>350 (Worse)	15 (12.0)
Visual field sensitivity, 3 upper quartiles vs lowest quartile, dB
Overall
>22.4 (Better)	116 (75.3)
≤22.4 (Worse)	38 (24.7)
Peripheral (≥45°)
>19.3 (Better)	114 (74.0)
≤19.3 (Worse)	40 (26.0)
Superior
>22.2 (Better)	111 (72.1)
≤22.2 (Worse)	43 (27.9)
Inferior
>22.1 (Better)	113 (73.4)
≤22.1 (Worse)	41 (26.6)
Left
>21.7 (Better)	112 (72.7)
≤21.7 (Worse)	42 (27.3)
Right
>21.7 (Better)	118 (76.6)
≤21.7 (Worse)	36 (23.4)
Gabor drifting grating threshold, median split, Hz
<0.14 (Better)	77 (50.0)
≥0.14 (Worse)	77 (50.0)

Abbreviation: UFOV, Useful Field of View.

^aData were available for 125 participants.

Age-adjusted associations between visual function and incident at-fault crash and near crash involvement are provided in Table 3. Only 3 participants had impaired visual acuity (worse than 20/40); as a result, the association between visual acuity and crash and near-crash involvement was not evaluated. Impaired contrast sensitivity was associated with crash and near-crash involvement (RR, 2.7; 95% CI, 1.3-5.5). Slowed visual processing speed (Useful Field of View subtest 2) was associated with increased crash and near-crash risk, both for moderate slowing (150-350 milliseconds [RR, 2.3; 95% CI, 1.1-4.9]) and more strongly for severe slowing(>350 milliseconds [RR, 5.0; 95% CI, 2.2-11.7]). Elevation in drifting grating thresholds was associated with crash and near-crash involvement (RR, 1.9; 95% CI, 1.1-3.5). Visual field sensitivity, overall and by region, was not significantly associated with crash and near-crash involvement; those with impaired sensitivity in the peripheral visual field had increased rates of crash and near-crash involvement (RR, 1.8; 95% CI, 1.0-3.3); this finding was not statistically significant (P = .07). Confounding was assessed but not present for sex, race/ethnicity, MMSE score, CES-D score, educational level, number of medical conditions, lens status, glaucoma, age-related macular degeneration, and diabetic retinopathy or diabetic macular edema.

Table 3. Age-Adjusted Associations Between Baseline Visual Function and At-Fault Crash and Near-Crash Involvement During 6 Months After Baseline Based on Naturalistic Driving Data (N = 154).

Visual function	At-fault crash and near-crash involvement (81 events), rate ratio (95% CI)
Contrast sensitivity, log sensitivity
≥1.5 (Better)	[Reference]
<1.5 (Worse)	2.7 (1.3-5.5)
UFOV subtest 2, ms
<150 (Better)	[Reference]
150-350	2.3 (1.1-4.9)
>350 (Worse)	5.0 (2.2-11.7)
Visual field sensitivity, dB
Overall
>22.4 (Better)	[Reference]
≤22.4 (Worse)	1.4 (0.8-2.8)
Peripheral
>19.3 (Better)	[Reference]
≤19.3 (Worse)	1.8 (1.0-3.3)
Superior
>22.2 (Better)	[Reference]
≤22.2 (Worse)	1.3 (0.7-2.5)
Inferior
>22.1 (Better)	[Reference]
≤22.1 (Worse)	1.4 (0.8-2.5)
Left
>21.7 (Better)	[Reference]
≤21.7 (Worse)	1.3 (0.7-2.5)
Right
>21.7 (Better)	[Reference]
≤21.7 (Worse)	0.9 (0.5-1.8)
Gabor drifting grating threshold, Hz
≤0.14 (Better)	[Reference]
>0.14 (Worse)	1.9 (1.1-3.5)

Abbreviation: UFOV, Useful Field of View.

Discussion

Using naturalistic driving data, this study has gone beyond the limitations of motor vehicle crash report data^7,8 to identify visual risk factors associated with at-fault crashes among older drivers with age-related eye conditions. This is the first study in this population, to our knowledge, that includes near crashes, expanding the number of events in the analysis.

Three visual risk factors associated with motor vehicle crash reports in previous studies were confirmed with naturalistic driving. Both moderate and severe slowing in visual processing speed elevated collision risk for older drivers, agreeing with accident report research.^3,4,18,36 This finding is also reflective of studies on driving performance problems among older adults for whom slowed processing speed is associated with performance errors.^37,38,39 Contrast sensitivity impairment was associated with crash and near-crash involvement, again agreeing with analyses based on motor vehicle crash reports,^17,19,40 and with a naturalistic driving study assessing the association between impaired contrast sensitivity and vehicle control on interstate highways at night.⁴¹ Elevated thresholds in motion perception of a drifting grating also increased crash and near-crash involvement, similar to a recent collision risk analysis using motor vehicle crash reports,¹⁹ and is consistent with previous research identifying motion perception impairment as associated with driving performance problems among older adults.^39,42,43

A novel aspect of this naturalistic driving study of drivers with vision impairment is our ability to use near crashes as a crash surrogate measure^9,15 because they are impossible to examine in motor vehicle crash report studies. As defined, a near crash requires a “last-second” evasive maneuver by the driver to avoid the crash. Near crashes and crashes have similar causal mechanisms, yet they have different safety outcomes because of the rapid evasive maneuver of a near crash.^9,15 In the present study, near crashes occurred at approximately twice the rate of crashes (RR, 1.9; 95% CI, 1.4-2.6).³²

Because there were only 3 drivers with visual acuity worse than 20/40, we were unable to evaluate the association between impaired visual acuity and at-fault crash and near-crash involvement. The prevalence of visual acuity impairment worse than 20/40 in a study of older drivers will be lower than in a population-based study of older adults, in general, because vision standards prevent drivers with significant acuity impairment from receiving a license. Such studies may be feasible in some states whose acuity standard extends to 20/60 or 20/100. Despite the government’s intent to provide a visual acuity standard that enhances safety, epidemiologic studies have largely failed to identify such an association with crash involvement. This is an interesting paradox; the primary vision screening test used by the government for licensure does not have a demonstrated association with driver safety.

Two previous studies have also used naturalistic driving to study rapid deceleration events among older drivers; the authors defined these events as near crashes.^44,45 However, these authors did not use video to confirm that the events were actually near crashes, unlike our study. Although rapid deceleration events can be near crashes, not all these events are near crashes, and some near crashes do not involve rapid decelerations.

Strengths and Limitations

This study has some strengths, including our use of naturalistic driving data to examine crashes by older drivers with age-related eye conditions to avoid the biases of motor vehicle crash reports. We used an outcome combining crashes and near crashes as an end point to improve our understanding of visual risk profiles. Earlier studies on near crashes were defined solely by rapid deceleration events,^44,45 whereas we validated near crashes by human observers viewing a video of the event. This study also has some limitations. With a small sample, we could not confirm a previously identified association between visual field impairment and crash involvement.⁵ In addition, our study was not designed to understand crash risk per diagnostic category.

Conclusions

Naturalistic driving by older drivers can identify and clarify the types of vision impairment most closely aligned with at-fault crash and near-crash events as they occur in everyday life. This information could inform licensure guidelines for persons with these functional problems and facilitate developing rehabilitation options.