With the aging of the driving population, the prevalence of medical conditions associated with loss of functional abilities needed to safely control a motor vehicle has increased. Considerable evidence indicates that declines in cognition, particularly neurological diseases such as dementia, lead to driving impairments and increase crash risk among older drivers (Breker et al., 2003; Hakamies-Blomqvist, 2004; Sims, McGwin, Allman, Ball, & Owsley, 2000). Many people with dementia continue to drive, and these drivers have a twofold increase in crash risk compared with those without cognitive impairment (Carr & Ott, 2010).

Dementia-related deficits that negatively affect driving performance include lack of insight, making compensation for limitations unlikely (Dickerson et al., 2007). Detection of the early stage of dementia is difficult, and even family members may not realize that cognitive changes are occurring. Driving skills may be preserved in this stage (Lundberg et al., 1997). Researchers have underlined the need for clinicians, family members, and people with mild cognitive impairment (MCI) to watch for changes in driving that may become increasingly problematic over time (Wadley et al., 2009).

Because clinicians’ assessment alone may not be accurate enough to determine driving competence in drivers with mild cognitive decline, research supports the use of on-road driving evaluations (Bédard & Dickerson, 2014; Classen, Dickerson, & Justiss, 2012; Wadley et al., 2009; Wheatley, Carr, & Marottoli, 2014) and experimental drives in simulators or instrumental vehicles as means to acquire information and knowledge regarding driving performance (Uc & Rizzo, 2011). The importance of identifying drivers with early dementia or MCI is underscored by their reduced capacity to self-regulate (Staplin, 2012; Staplin, Lococo, Gish, & Decina, 2003).

Wadley et al. (2009)  examined the driving performance of 59 cognitively normal older adults and 46 people with MCI using an on-road driving assessment. Although the MCI group was older and mostly male, differences in mean driving performance ratings were small, with participants in both groups receiving high mean ratings. However, the people with MCI were significantly more likely to receive a less-than-optimal rating on left turns, lane control, and global ratings. The authors discussed these specific difficulties in people with MCI as resulting from the greater demands in executive function associated with these maneuvers.

Frittelli et al. (2009)  examined the impact of Alzheimer’s disease and MCI on driving ability using an interactive driving simulator (STISIM Driving Simulator, Hawthorne, CA). Twenty participants had mild Alzheimer’s, 20 had MCI, and 19 were neurologically normal control participants. Drivers with Alzheimer’s disease performed significantly worse than those with MCI and control participants on three driving behaviors: (1) length of the run in seconds, (2) mean time to collision, and (3) number of off-road events (e.g., crossing the lateral border of the road). The only statistically significant difference between participants with MCI and control participants was a shorter mean time to collision (i.e., the time to contact the preceding vehicle if the test car kept moving under constant velocity) for the MCI group.

In a similar study, Devlin, McGillivray, Charlton, Lowndes, and Etienne (2012)  examined the performance of matched older drivers with and without MCI when approaching intersections. Results indicated that drivers with MCI exhibited behaviors that were less situationally appropriate than control drivers when approaching controlled intersections and critical light-change intersections. Specifically, healthy drivers demonstrated a greater number of foot hesitations on approach to stop-controlled and critical light-change intersections compared with the MCI group, indicating greater readiness in the event rapid braking was required.

Drivers with a suspected cognitive impairment may be referred to their licensing authority for medical review. Although observed or reported deficits in performing basic activities of daily living (ADLs) contribute to a diagnosis of dementia, preservation of ADLs can help differentiate MCI from mild dementia. At the same time, evidence that people experience difficulty in performing more complex instrumental activities of daily living (IADLs), such as driving, is a more reliable marker to distinguish people with MCI from healthy older people (Devlin et al., 2012). This finding is consistent with evidence suggesting that observation and evaluation of IADL performance can assist in determining who might be an at-risk driver (Dickerson, Reistetter, Davis, & Monahan, 2011). Thus, neuropsychological testing results and reports by the driver or caregivers of difficulties with ADLs and IADLs may be indispensable to this review (Gold, 2012). Nevertheless, the on-road assessment is seen as the gold standard. When driving performance measures are sought to supplement these indicators of cognitive impairment, however, safety and cost considerations could dictate the use of a driving simulator rather than the more expensive, time-intensive, and potentially risky on-road evaluation.

Within recognized limitations, simulators permit objective measures of driver performance and driver errors in a controlled environment for a wide range of driving tasks. A major challenge in using a simulator to assess driver performance is choosing effective and well-defined performance measures in scenarios that are most likely to elicit behaviors with clear safety relevance and that provide information about the specific mechanisms of impairment that underlie the behaviors (Vardaki, Yannis, & Papageorgiou, 2013).

Although research investigating the effects of functional deficits associated with MCI often finds small decrements in safety-relevant driving behavior, the simulated driving tasks used in such studies strongly influence their outcomes. Tasks presenting demands that tax a driver’s capacity for serial information processing (e.g., working memory) while also involving executive functions arguably are more efficient in revealing differences that are operationally significant. In other words, using the simulator with significant demands effectively increases the opportunity to differentiate between people who have retained safe driving behaviors and those who have not.

Working memory, a cognitive ability of key importance in driving, allows a driver to remember and use information while processing and responding to the real-time demands of performing moment-to-moment vehicle control tasks (Staplin, 2012; Staplin et al., 2003). Executive functions (e.g., decision making, planning, judgment) all strongly interact with working memory and attention (Fisher, Rizzo, Caird, & Lee, 2011).

To further explore the results of a pilot study, this study used an improved experimental methodology to examine how varying levels of operational (e.g., steering, braking) and tactical (e.g., deciding when to change lane) driving task demands might differentially affect message recall for older drivers with MCI versus a group of age-matched, healthy control participants. This study examined the extent to which differences between drivers with MCI and control participants on a sign recall task in a fixed-base driving simulator could better predict a driver’s MCI diagnosis compared with a self-reported decline in driving proficiency, frequency of avoiding driving, or age alone.

The participants were administered a questionnaire and participated in simulated driving tasks in which task demands varied at the operational and tactical levels. We hypothesized that recall performance would be significantly worse in participants with MCI because of greater interference of varying task demands on working memory.

This study was part of a larger driving simulator experiment (Yannis et al., 2013). All participants met the following criteria: held a valid driving license, had driven for more than 3 yr and at least 1,500 mi, had driven at least once a week and at least 6 mi/wk during the previous year, and had a Clinical Dementia Rating score of <2 (mild or very mild dementia). Exclusion criteria included psychiatric history of psychosis, kinetic movement disorder, motion sickness, alcohol or drug addiction, or disorder of vision or of the central nervous system.

The MCI group (n = 12) had 8 men and 4 women with a mean (M) age of 64.8 yr (standard deviation [SD] = 8.9, range = 51–76). The control group (n = 12) had 6 men and 6 women with a M age of 59.5 yr (SD = 7.2, range = 51–78). The two groups did not differ significantly on age (p = .18), driving experience (p = .18), driving exposure (p = .35), education (p = .84), total accidents (p = .67), and accidents in the past 2 yr (p = .71).

The diagnosis of MCI was based on the criteria of Petersen and Morris (2005), which included complaints about memory impairment by the person or a family member and verified impairment in at least one cognitive domain but with preserved functional ADLs and absence of dementia. Exclusion criteria included a Clinical Dementia Rating score ≥1 and a history of neurological or psychiatric disorder or depression.

Results of standard neuropsychological tests revealed significant differences between the MCI and control groups in measures assessing verbal episodic memory (Hopkins Verbal Learning Test–Revised; Benedict, Schretlen, Groninger, & Brandt, 1998; MCI group M = 3.73 ± 2.94, control group M = 7.42, SD = 3.12; p = .012) and information processing speed (Symbol Digit Modalities Test; Smith, 1991; MCI group M = 33.2, SD = 17.7, control group M = 46.0, SD = 10.1; p = .045). In contrast, the two groups did not differ significantly on measures of general cognitive functioning (Mini-Mental State Examination; Folstein, Folstein, & McHugh, 1975), working memory (Letter Number Sequencing; Wechsler, 1997), visuospatial memory (Brief Visuospatial Memory Test–Revised; Benedict, 1997), psychomotor speed (Trail Making Test Part A; Reitan, 1979), mental flexibility (Trail Making Test Part B; Reitan, 1979), and visuospatial perception (Judgment of Line Orientation; Benton et al., 1994).

All participants were administered a questionnaire to gather demographic data; driving experience, exposure, and self-restrictions; self-assessment of driving skills; driving difficulties (Vardaki, & Karlaftis, 2011); presence of specific driving behaviors associated with early dementia (Eby et al., 2009); presence of distracted driving behaviors; emotions while driving; and history of accidents. Family members also completed a questionnaire about their impressions of participants with MCI.

The self-assessment asked older drivers to assess whether their driving performance had declined over the previous 5 yr, to indicate whether they avoided driving, and to provide the reasons for avoiding driving (e.g., self-regulation, discouraged by family). The inclusion of the items relating to self-assessment of global driving ability and driving ability in specific driving situations was based on findings of a previous study investigating perceptions of safe driving ability in relation to actual and self-assessed performance of a group of older adults during an on-road trial (Vardaki & Karlaftis, 2011). Specifically, drivers reported on changes in global driving proficiency by comparing “your ability now to 5 years ago” using a 5-point Likert scale (much worse, a little bit worse, the same, a little bit better, much better). Responses were consolidated into two categories because of the small sample size. For the question about frequency with which they avoided making trips because of concerns about driving, the four responses (never, rarely, sometimes, often) were consolidated into two categories: never and rarely or sometimes because no one responded often.

The driving simulator, a FOERST Trainer model (FOERST GmbH, Wiehl, Germany), consisted of an adjustable driver seat, fixed support base, and three LCD 40-in.-wide screens, which provided a 170° field of view. Drivers viewed the LCD displays from a distance of 125 cm. Display resolution for the LCD screens was full high definition (1920 × 1080 pixels). The full dimensions were 230 × 180 cm, with a base width of 78 cm. The simulator featured an adjustable driver seat, a steering wheel 27 cm in diameter, pedals (throttle, brake, clutch), dashboard, and two external and one central mirror that appeared on the side and main screens. Performance data were extracted directly from the simulator and logged 30 times per second.

The present investigation built on tentative conclusions from a pilot study by using an improved experimental methodology (i.e., by controlling the time drivers were exposed to the message) to examine how varying levels of operational and tactical driving task demands might differentially affect recall of messages on road signs for older drivers with MCI versus a group of matched, healthy control participants. Experimenters were blind to the results of the neuropsychological tests. All participants gained a degree of familiarity with the simulator through participation in a prior experiment that lasted approximately 45 min. This prior experience allowed participants to accommodate to using a driving simulator and practice their driving skills (e.g., distance judgment, pedal and steering control) and also served as a screen for susceptibility to simulator adaptation syndrome for the study sample.

The experiment, completed in a single laboratory session, included three test conditions (TCs) using repeated measures for all participants in three separate simulator drives of approximately 2 min each. The experiment measured the effect of different levels of demand on intervening driving tasks (i.e., displayed between sign message presentation and recall) on the recall of the sign information. The time between the presentation and recall of the sign message was roughly equivalent across conditions (100 s). The sign message was presented for a fixed interval (∼8 s) that was constant across study participants. Three equivalent messages were constructed for presentation during the three TCs, using a common format: Each sign displayed three units of information, indicating (1) the type of situation ahead, (2) its distance, and (3) a driver action required, using standard wording (Campbell et al., 2012). Before each of the three drives in the simulator, participants were instructed to respond to traffic control information, maintain safe gaps with other vehicles, and maintain a constant speed at the posted speed limit unless doing so was not possible because of other traffic, road construction, and so forth.

All driving scenarios involved driving along straight sections and gentle curves on a limited-access divided roadway. Each scenario began with a period of low-demand driving requiring minimal steering input and with the only other traffic being two vehicles ahead with the lead vehicle at a safe distance ahead. These low-demand driving conditions persisted throughout the first condition. The three TCs were as follows:

1. In TC1, drivers experienced the lowest level of demand and were required to respond only to operational-level driving tasks.

2. In TC2, drivers made a double lane change that involved driving through a roadwork section containing large blocks (barriers) on each side of the road, causing the road to progressively narrow (1:20 taper ratio; lane width 3 m). These requirements were designed to produce an intermediate level of demand; thus, demand in this scenario was higher than in TC1.

3. In TC3, drivers were presented with the same roadwork section and associated steering requirements as in TC2, but after the forced lane changes, they were required to execute an additional lane change if a discriminative stimulus (activation of the brake lights on a lead vehicle) was presented. This decision rule was included in the predrive instructions. The addition of this working memory task was designed to result in the highest level of demand in this scenario.

In all TCs, participants relied on working memory to recall the sign message after completing each driving scenario. Immediately at the end of each drive, participants were asked to recall the sign message. The experimenter assigned a score of 0–3 to indicate whether none, 1, 2, or all 3 information units were recalled. The order of presentation of TCs was randomized.

Regarding self-reported changes in driving, 50% of the drivers in the MCI group responded that their current driving ability was worse than 5 yr earlier, whereas only 8% of control group drivers gave this response. Substantial majorities of drivers in both the MCI group (75%) and in the control group (83%) reported that they never avoided making trips because of concerns about driving.

Two sets of data analyses were performed. The first concerned differences in drivers’ speed choice under each test condition, as a manipulation check, to confirm that the demand of the driving task varied across TCs as revealed through drivers’ speed reductions to negotiate the roadwork section. The second analysis involved the sign recall scores to evaluate the hypothesized deficit for MCI drivers versus control participants and a potential interaction of sign message recall with task demand level.

The first set of analyses, which examined drivers’ speed reductions to negotiate the roadwork section, used a two-way mixed analysis of variance (ANOVA) to test for main effects of driver group, a between-subjects variable, and the level of demand for intervening driving tasks, a within-subjects variable, on drivers’ speed and also for a possible two-way interaction between these variables.

As shown in Table 1, in TC1 the mean speed was higher than in TC2 and TC3; in addition, the mean speed of the MCI group was lower than the mean speed of the control group across all levels of task demand. The ANOVA indicated that the effect of group membership on speed was not significant, F (1, 22) = 2.28, p > .05, but that differences in speed associated with the level of driving task demand were reliable, F (1.53, 35.57) = 32.09, p < .001. In other words, regardless of group membership, all participants reduced speed across TCs, suggesting that the level of demand was indeed varied by imposing different types of operational and tactical driving tasks on participants. A Bonferroni-corrected post hoc test showed that mean speed at TC1 was significantly different than at both TC2 and TC3 (both ps < .001); however, speeds at TC3 were not significantly different from speeds at TC2 (p > .05).

Sign message recall data are shown in Table 2 for each level of driving task demand. The MCI group performed more poorly in message recall, demonstrating higher percentages of low recall scores (0 and 1) than the control group. No clear pattern emerged indicating an interaction between group membership and driving task demand.

A general estimating equation (GEE) model (ordered multinomial logistic regression) was specified to examine the relationship between participant group and performance in the sign recall task, adjusting for potential intercorrelations among sign recall task for each participant in the three TCs. As shown in Table 3, the ordinal logistic GEE (applying a cumulative logit link function) indicated that control participants were significantly more likely to perform better than MCI participants in the sign recall task, exponential parameter estimate (Exp[b]) = 11.76, 95% confidence interval (CI) [2.73, 50.62], p = .001.

This analysis further revealed that, regardless of group membership, participants performed better in the recall of sign information in TC1 than in TC3, although this difference was not significant, Exp(b) = 2.46, 95% CI [0.99, 6.12], p = .052. Performance in the sign recall task was more likely to be higher in TC2 than in TC3, and this difference was statistically significant, Exp(b) = 4.85, 95% CI [2.08, 11.33], p < .001. The interaction effects were considered during the model building process, but they were not included in the final model specification because they were not significant at any reasonable level. Age was considered during the model building process, but the convergence criterion was not satisfied.

Building on the reliable differences between MCI and control participants on the sign recall task, we further explored the extent to which such evidence could better predict whether a driver would be diagnosed with MCI compared with self-reports of changes in driving proficiency or avoidance of driving or on the basis of age alone. We used stepwise logistic regression (Table 4) with a threshold of .05 for statistical significance. The dependent variable was group membership (a dichotomous variable), and the independent variables were age; self-reported changes in global driving proficiency (relative to 5 yr earlier), with 1 = worse and 2 = unchanged or better; self-reported frequency of avoiding driving, with 1 = never and 2 = rarely or sometimes; and recall score in the high demand test condition, TC3, with 1 = score of 0 or 1 and 2 = score of 2 or 3.

In the regression model (Table 4), predictor variables that failed to meet the criteria for inclusion were dropped, at each step in turn, until it was revealed that self-reported changes in global driving ability was the only statistically significant predictor. The associated odds ratio indicated that the odds of study participants who self-reported their driving ability as being “worse now than 5 years ago” also being diagnosed with MCI were 1/0.09, or 11 times more likely than the odds of those participants who rated their driving ability as unchanged or better also being diagnosed with MCI (final model: Hosmer and Lemeshow’s R² = 0.16, Cox and Snell’s R² = 0.20, Nagelkerke’s R² = 0.27, model χ² = 5.455).

The results of this study replicated earlier findings showing that drivers with MCI performed significantly more poorly on a sign recall task across varying levels of driving task demand than a cognitively intact comparison group. However, in the earlier study the MCI group was, on average, more than a decade older than the control participants, whereas in this study the control participants and MCI drivers were of comparable age.

This study found that message recall scores in a simulated driving scenario with elevated working memory demands, self-reported frequency of driving avoidance, and driver age did not predict a clinical diagnosis of MCI. Only self-reported changes in global driving ability were significant in this regard.

This research also examined the utility of this measure of cognitive status obtained in a driving simulator in a broader context—that is, early identification (screening) and assessment of drivers at risk because of cognitive impairment. This utility will likely become more pronounced as the population ages (Dickerson, 2014). Although clear evidence exists that dementia affects driving (Wheatley et al., 2014), it is not clear when or even whether a person with MCI will become an at-risk driver. Licensing jurisdictions are struggling to determine appropriate licensing renewal policies and practices while attempting to be sensitive to the risks associated with normal aging. Considering that medical conditions are more prevalent among older drivers, it has been recommended that older drivers receive individual driving evaluations (Bédard & Dickerson, 2014; Lane et al., 2014; Wheatley et al., 2014). The cost of such evaluations may be prohibitive, however, so it is important to consider other methods of testing, such as on a driving simulator, to determine fitness to drive. Can early identification of drivers who are at risk because of MCI be accomplished with more cost-effective methods? Although our results are tentative, given the small sample size, they suggest that this question may be answered in the affirmative. Future research with a larger sample is needed to explore the issue further.

Although people with MCI and those in the earliest stages of a progressive, dementing illness may be able to continue to drive safely for some time, a proper diagnosis is important not only for planning treatment but also when considering appropriate driving or licensing restrictions and requirements for periodic review to determine license status. Our results suggest that screening for age-related cognitive impairment should incorporate a subjective perception of changes in driving proficiency (i.e., using one’s earlier self as the baseline) to complement clinical test results for early identification of drivers who merit medical review.

When more in-depth assessment is needed, simulators that provide objective measures of driver performance remain an essential tool to better understand the interaction between individual differences and varying situational demands in safe and effective vehicle control. Even with a small sample, a reliable main effect of group membership indicates that older drivers with MCI are at a disadvantage when new information is presented (e.g., on a variable message sign) that must be retained in working memory and applied after some additional period of driving. In addition, differences shown in this study, though not reliable because of the small sample size, suggest that this effect is exaggerated as driving task demands increase. As a platform for research and assessment, driving simulators may become even more important as assistive technologies and automated control systems are introduced to delineate the conditions under which both healthy and cognitively impaired drivers may retain control of their vehicles.

The results of this study have implications for practitioners. First, self-reported global changes in driving was the only significant factor in predicting a clinical diagnosis of MCI. If this finding is replicated in future research, clinicians will have evidence-based support to ask a simple question about a client’s driving for use as an indicator to explore the issue further. This support is important because current evidence suggests that practitioners (e.g., occupational therapists, physicians) are not asking their clients about driving (Argintar et al., 2013; Dickerson, Schold Davis, & Chew, 2011). Practitioners interacting with clients at all levels of service (e.g., acute care, subacute care, rehabilitation) and with certain primarily physical diagnoses (e.g., hip replacement, amputation) could screen for potential safety risk in driving by asking clients about their perception of their driving. If a client indicates that his or her driving abilities have decreased, further screening for MCI would be warranted because few of the healthy control participants indicated this decrease.

It is to be expected that some people will not answer the question negatively because cognitively impaired people generally do not have insight into their driving abilities (Adler & Kuskowski, 2003) and do not cease driving even when given the recommendation (Dobbs, Carr, & Morris, 2002). The straightforward question, “How do you rate your driving ability now compared with 5 years ago?” offers an appropriate and nonthreatening lead to identifying potential safety risk.

The second contribution of this study is support for the use of a driving simulator to assess driving ability. Although the sample size was small, this study’s results show a clear trend toward poorer performance in the MCI group, a finding that needs to be explored further. The increasing number of errors drivers made as the tasks demands increased suggests that increasing task demands may differentiate people with cognitive impairment from those without impairment. Although the on-road assessment is clearly the most ecologically valid evaluation for fitness to drive (Classen et al., 2012), safety concerns for the evaluator and client limit its use. Safety concerns are mitigated on the simulator. Therefore, using a driving simulator to examine number of driving errors may be an effective method of identifying clients with cognitive impairment before an on-road evaluation. Future studies are needed to determine specific protocols and evaluation criteria.

For clinicians who already use interactive driving simulators, this study also offers some practical guidance in use. Simulator adaptation syndrome is more prevalent in the older population (Brooks et al., 2010), and it may be a barrier to use. Because making left- or right-hand turns in the simulator appears to be a major factor in increasing the symptoms, the use of memory tasks and increasing demands on relatively straight roads may allow clinicians to use the simulator with more clients.

The practice implications of this study can be summarized as follows:

• Self-reported global changes in driving, the only significant factor in predicting a diagnosis of MCI, is important for practitioners to consider and can be assessed by asking the client an easy, informal question.

• The trend toward poorer performance for the MCI group in simulator measures supports the use of interactive driving simulators in clinical settings with clients as a screening or assessment tool for MCI or for driving abilities.

• The concept of grading driving simulator activity protocols by increasing task demands through a series of driving scenarios may increase the viability of using driving simulation as a critical intervention strategy.

• The use of memory tasks and increased demands on relatively straight roads (rather than turns) allows clinicians to use the simulator with more clients by avoiding movements that increase the symptoms of simulator adaptation syndrome, more prevalent among older adults.

As with any study, this one has several limitations. Self-reported measures may become less useful the older the target population and the greater the extent of cognitive impairment (Zimmerman & Magaziner, 1994). An investigation with a substantially larger sample may reveal additional, significant predictors of an MCI diagnosis among a broader array of simulator measures, exposure data, and self- or informant reports on diverse driving behaviors. In addition, larger samples with appropriate measurement techniques could analytically better account for the influence on driving behaviors and performance of confounding variables such as age and other characteristics (e.g., driving experience and driving exposure) associated with driving competence.

The results of this study add to the growing body of knowledge about the use of driving simulators as a clinical tool in occupational therapy practice. Because interactive driving simulators are a complex technological tool, their use requires training and knowledge by the clinician before they can be an effective tool for evaluation or intervention. The other outcome of the study is that an inquiry to a client about a change in the IADL of driving could be relatively useful in identifying possible undiagnosed cognitive impairment or driving risk, both critical concerns for occupational therapy practitioners working with older adults.

This article is based on two research projects implemented within the framework of the operational program “Education and Lifelong Learning” of the National Strategic Reference Framework, namely the THALES research funding program (Investing in knowledge society through the European Social Fund) and the Action: ARISTEIA (Action’s Beneficiary: General Secretariat for Research and Technology), cofinanced by the European Union (European Social Fund) and Greek national funds.

We thank Nikolaos Andronas, MD, PhD candidate, Attikon General University Hospital, for examining the participants of this study; Constantinos Antoniou, for his suggestions during the data analysis; and Dimosthenis Pavlou, PhD candidate, for his work on the simulator experiment.