Background
Driving is a highly visual task [
1,
2]. Yet vision impairment is common in older adults [
3,
4], and thus an important question is how vision impairment impacts older driver safety. Older drivers aged ≥70 years old have among the highest rates of motor vehicle collisions (MVC) compared to drivers in other age groups [
5]. Research over the past 2–3 decades indicates that some types of vision impairment are associated with elevated MVC risk in older drivers including slowed visual processing speed [
6‐
8], visual field defects [
8‐
10], and contrast sensitivity impairment [
11]. The majority of studies addressing risk factors for crash involvement, including population-based studies, have utilized accident reports as the primary outcome, which are submitted by police who typically do not directly witness the crash. While accident reports document that a crash occurred and provide a wealth of information on the circumstances (e.g. place, weather, vehicles involved, driver’s age), they cannot always be used to provide an accurate description of what actually happened or the causes of the crash. Additionally, accident reports do not provide information on the occurrence of crash events where police do not attend the scene, collisions occurring on private property (e.g., parking lots), and near-misses; thus outcome events are likely incomplete and biased [
2]. Previous studies using accident reports to identify risk factors for collision involvement do not objectively measured driving exposure (miles driven), but rather have relied on the driver’s self-report of driving exposure. Such studies cannot address mechanistic questions about how impaired visual function directly impacts driving performance. They provide little to no information about how visual function is related to driver behaviors and vehicle control, such as lane control and turning, or the impact of secondary tasks on driver behavior and vehicle control. All these issues undermine the goal of achieving a comprehensive understanding of visual mechanisms underlying driver performance and safety.
Instead of relying on accident reports as the outcome of interest in studying the relationship between vision impairment and driver safety and performance, in the present study we use naturalistic driving research techniques [
12,
13]. This approach involves installing a high-tech yet unobtrusive data acquisition system (DAS) in a participant’s own vehicle. The DAS continuously records multi-channel video of the driver and roadway environment, sensor-based kinematics data, GPS location, and presence of nearby objects near the front of the vehicle. The DAS’s unobtrusive design is facilitated by advances and miniaturization of computer, sensor, data storage, communications, and video technology. It is designed to automatically and continuously collect data whenever the instrumented vehicle is driven by the research participant (i.e., from key on to key off), and remains installed in the vehicle for a lengthy period of time (i.e., months or even years). The advantages of naturalistic driving techniques are striking in contrast to other driving research methodologies. Naturalistic methods avoid the short snapshot (e.g., 1 h) of a standardized course of on-road driving typical of most on-road studies, where drivers know they are being evaluated by study personnel and thus are likely to be on their “best behavior” [
2]. On the other hand, such in-person evaluation may also result in poorer performance if the perceived pressure of the study participation causes stress. Also, the route driven during the on-road evaluation may not be representative of the typical driving trips (e.g., traffic density, types of roadways) made by study participants in their everyday life. In naturalistic research, the driver chooses all driving routes in the course of everyday life. On-road evaluation includes explicit instructions by the CDRS on when and where to make turns (e.g., “at the next traffic light, turn left”). Naturalistic driving methods allow for the study of not only crash events but near-crash events, which are similar in terms of driver behavior and vehicle kinematics to actual crashes, yet occur at a rate 2–10 times higher than crash events [
14,
15], thus creating a larger number of outcome events to analyze in risk factor modeling.
Naturalistic driving methods have been successfully employed in the study of driver safety and performance for over 10 years, a body of work that establishes their feasibility as a measurement approach. The literature on naturalistic driving research specifically focused on older drivers is, however, small and is summarized here. Older adults who experience a decline in contrast sensitivity over 12 months are more likely to be involved in rapid deceleration events while driving [
16]. Visual function in older drivers was found to be unrelated to involvement in lane changing errors [
17] but narrowing of the visual attentional field was associated with a higher risk of failing to stop at red lights [
18]. Older drivers who restrict their night driving tended to be those with worse visual fields and contrast sensitivity [
19]. With respect to head movement while driving through intersections, older drivers had a greater degree of lateral head rotation than middle-aged drivers [
20], with the authors suggesting that it may be a compensatory mechanism for older adults’ reduced visual attention skills. Only a few studies thus far have used naturalistic driving data to study visual risk factors for crash and near-crash involvement by older drivers. Older drivers with worse contrast sensitivity had a higher rate of crash and near-crash events [
21], however this finding was based on only 20 drivers. Using the Strategic Highway Research Program 2 (SHRP 2) data [
13], two studies using different analytic approaches both found that impaired contrast sensitivity and peripheral vision were related to elevated rates of collision involvement [
15,
22]. Also using SHRP 2 data, Guo et al. [
23] found that secondary-task-induced distractions posed a greater safety threat for older drivers than for middle-aged drivers, however older drivers were less likely to be engaged in secondary tasks while driving. Studies based on the SHRP 2 data included participants with normal or near-normal visual sensory and visual-cognitive skills, thus making it difficult to evaluate associations between visual dysfunction and driver safety and performance.
Here we describe the Alabama VIP Older Driver Study, a naturalistic driving study designed to examine associations between vision impairment in adults ≥70 years old and crash and near-crash involvement as well as other driver behaviors. In order to overcome a major limitation of earlier studies as described above (i.e., most had normal or near-normal vision), our enrollment process targets older adults with a range of visual capabilities with respect to contrast sensitivity and visual processing speed. These aspects of vision were selected as enrollment criteria because they are two of the strongest visual risk factors for collision involvement and driving problems in older adults [
6‐
8,
11,
16,
24,
25]. Our recruitment strategy also targets older adults who are patients from an ophthalmology clinic since they are more likely to have chronic eye conditions that cause visual impairment. The study has three aims:
To examine the relationships between vision and naturalistic driving performance in older drivers ≥70 years old. Analyses will focus on the relationship between vision and safety critical events (crashes, near-crashes), lane-keeping, turning at intersections, driving performance under secondary task demands, and when a “co-pilot” (passenger in the front seat) is present. Visual function measurements will include assessments of contrast sensitivity, visual processing speed, visual acuity, visual field sensitivity, and visuo-spatial processing. These aspects of vision were selected because they have been widely related to older driver safety and performance [
6‐
11,
16,
24‐
29]. There are several hypotheses relevant to this aim. Older drivers with contrast sensitivity loss, slowed visual processing speed, and/or visual field impairment will be more likely to exhibit critical safety events, lane keeping deviations, and intersection turning errors, as compared to those without these impairments. Older drivers with these vision impairments will be more likely to exhibit these problems under secondary task demands than drivers without these vision impairments. Older drivers with vision impairment who have a co-pilot will be less likely to exhibit these problems than drivers with vision impairment who do not have a co-pilot.
To examine these relationships in light of potential effect modifiers, specifically, driver characteristics (e.g., other visual problems, cognitive status, medical conditions, recent history of MVC, physical function, medications); environmental factors (e.g., roadway type, weather, time of day); and vehicle factors (e.g., type of vehicle, tire condition). Our primary hypothesis here is that drivers with both visual impairment and cognitive impairment will be more likely to exhibit critical safety events, lane keeping deviations, and intersection turning errors, as compared to those without vision impairment only but not cognitive impairment.
To examine the relationships between driving performance as measured by naturalistic driving methods and driving performance ratings provided by a CDRS [
30] on a standardized driving route (the clinical gold standard). Our primary hypothesis is that worse CDRS ratings of driving fitness will be associated with more lane keeping deviations, intersection turning errors, and rapid decleration/acceleration events.
Statistical analysis plan
The primary aim of the proposed work is to examine the relationships between vision and naturalistic driving performance in older drivers ≥70 years old. We will examine associations between different types of vision impairment and crash and near-crash involvement, lane-keeping, turning at intersections, driving performance under secondary task demands, and when a “co-pilot” is present. For these analyses, study participants will be grouped according to vision impairment status (e.g., whether impairment is present or not), type of impairment, impairment severity, and groups will be compared with respect to demographic, health and functional (including vision), behavioral and driving characteristics using a variety of statistical tests including analysis of variance and chi-square tests or their non-parametric equivalents and/or small-sample size (e.g., Freeman-Halton extension of Fisher’s exact test), as deemed necessary. The objective of these analyses is to identify potential confounders that might subsequently be used to adjust associations between the vision groups and the dependent measures of naturalistic driving.
For comparisons between the participant groups and the DAS-derived dependent variables,
it is important to keep in mind that the diversity in the measures will require a wide range of statistical approaches, the most appropriate of which may not be clear until the characteristics of the actual data have been evaluated and the analysis process is underway. It is important to keep in mind that the measurement of naturalistic driving data proposed is relatively novel, so there is little precedent to draw from. Simply based upon the nature of the measures described above, several approaches are likely to be appropriate and be employed. Because all of the dependent variables can be enumerated as counts (i.e., the # of times each event occurs), Poisson regression would be a useful tool to model the count of these events per mile driven as a function of vision impairment status with and without adjustment for potential confounding characteristics. Using this approach would call for the calculation of rate ratios and associated 95% confidence intervals using the unimpaired group as the common reference. Another, related, approach would be the use of generalized estimating equations to similarly model the occurrence of these events as a function of vision impairment status. This approach would model each event as a binary occurrence but account for the clustering of events within participants and/or drives. Odds ratios and associated 95% confidence intervals would be estimated for vision impairment as well as for other variables of interest.
The focus of our second aim is to explore the modifying effect of driver, environmental and vehicle factors on the association between vision impairment status and driving measures procured from the DAS. Therefore, to evaluate the presence of effect modification, the aforementioned statistical models will be stratified according to potential effect modifiers and the relevant measures of association (e.g., ORs) will be compared across strata.
Aim 3 seeks to examine the relationships between driving performance as measured by naturalistic driving methods and driving performance ratings provided by a CDRS on a standardized driving route (the clinical gold standard). This will provide the opportunity to examine the validity of CDRS ratings against objective measures of performance. Based upon prior work, many of the on-road driving performance ratings are ordinal variables, some of which may be used to classify drivers on a binary basis as “safe” or “unsafe”. However, in the context of the present study, the ordinal measures are likely of greater interest. As a result, we will calculate correlation coefficients (both Pearson’s and Spearman’s) for the association between the DAS- and CDRS-derived measures of driving performance which are common to both approaches, for example, lane-keeping. We will explore both the confounding and modifying influence of driver (including vision), environmental and vehicle factors on these associations using regression models using the DAS-derived measures as dependent variables and the CDRS-derived measures as independent variables. As noted above, the exact nature of these models will be highly reliant on the nature of the dependent variables; however, Poisson and logistic regression are two likely approaches.
Discussion
Older drivers are the fastest growing group of drivers on the road in the US [
74]. There are approximately 40 million adults aged ≥70 years old in the US (69) and 4 out of 5 of them (32 million), are drivers [
74]. Older adults have a crash rate nearly equal to that of younger drivers whose crash rate is the highest among all age groups [
75]. Once in a crash, older adults are more likely to be injured or die than are young drivers [
76]. Removing the driver’s license of an older adult has negative consequences for the individual and the society (2–10). Identifying ways to enhance driver safety among older adults as well as identifying drivers who are unsafe behind the wheel has become a pressing public health issue, garnering much media attention.
Researchers focused on driver safety and performance have had access to several approaches:
epidemiological methods utilizing national crash databases, population-based surveys, statistical simulations, closed road circuits, laboratory-based studies on the characteristics of drivers, and driver simulator studies. The relative strengths and limitations of these research methods have been discussed at length previously [
2]. These and other approaches have contributed substantially to the knowledge base. However, until recently there has been no feasible way to examine real-world, on-road driver behavior and vehicle kinematics in detail over extended periods of time. The naturalistic driving study paradigm has emerged to fill this gap facilitated by advances and miniaturization of computer, sensor, data storage, communications, and video technology. The measurement of actual, real-world driving behavior over extended periods of time, rather than short duration “snap-shots” of on-road driving, is the primary strength of the naturalistic driving approach. Limitations must also be acknowledged. For example, in our study generalization of findings from our older driver cohort to others remains unknown, an issue to be explored in future research. Volunteer bias is also present, however this is a problem for all older driver performance studies and is not unique to naturalistic driving studies. Some volunteers’ vehicles cannot be installed with a DAS because of incompatibilities between the designs of the DAS and vehicle. However, this rate is expected to be low, based on previous research.
In summary, the Alabama VIP Older Driver Study is the first naturalistic driving study whose design focuses on the examination of the association between vision impairment in older drivers and actual on-road driving, including both safety measures (crashes and near-crashes) and driver behaviors. Although a limited number of studies have examined the relationship between vision and driving in older adults using naturalistic techniques, the vast majority of drivers in previous studies had normal or near normal vision [
15‐
21,
23], thus hindering an examination of the relationship. By design our study is not population-based but instead, focuses on those older drivers with vision impairment. Thus, the Alabama VIP Older Driver Study will provide novel information on how various types of vision impairments (e.g., contrast sensitivity loss, visual field impairment, slowed visual processing speed) and the severity of those impairments impact actual on-road performance and safety. Study findings have the potential to stimulate the development of improved methods for on-road evaluation of older drivers, rehabilitation interventions for visually impaired older drivers, and evidenced-based vision standard policies for licensure.