Although MCI has proven problematic to define, Petersen [6] has described MCI as a transition between what is considered normal age-associated cognitive decline [7] and clinical dementia. However, he notes that this transitional area is not necessarily a continuum; overlaps at each end of normal cognition, MCI, and dementia are often evident, blurring the distinctions between the three [6]. While it is difficult to accurately pinpoint when someone transitions from one stage into the next, there is consensus that individuals with MCI experience one or more impairments in cognition that are not severe enough to warrant a diagnosis of dementia, and the most common impairment is in memory [6]. When compared to healthy peers, those with MCI are more likely to develop dementia, though the condition may also remain stable or even regress [8].

While a universally accepted set of diagnostic criteria for MCI does not exist, there have been attempts to provide guidance to healthcare professionals and researchers, the most well-known of which is the Petersen’s criteria [6, 9]. Because it precludes recommendation of specific diagnostic tools or cutoff scores, Petersen’s decisional framework relies heavily on clinical judgment, leaving considerable room for interpretation. For example, while the Montreal Cognitive Assessment (MoCA) is increasingly used by clinicians to operationalize general cognition in MCI, the mini-mental state examination (MMSE) is also used for the same purpose. Cutoff scores for MCI range from 23 to 26 on the MMSE [10]. For various other memory tests, cutoff scores of 1.0, 1.5, and 2.0 standard deviations from the mean and age norms [10] have also been reported. And while a cutoff score of 0.5 [11] on the Clinical Dementia Rating (CDR) Scale is regularly used within a research context to delineate MCI (both as a stand-alone tool or within the Petersen framework [11]), most contend that this threshold denotes a broader group of individuals who may have MCI or possible/probable/very mild dementia [12] and does not capture all cases of MCI [13].

These inconsistencies in the clinical definition of MCI have led to variation in the reported prevalence of the condition [10]. In their systematic review on the prevalence and incidence of MCI, Ward et al. [14] identified ten studies that provided prevalence rates for MCI among older adults ranging from 3 to 42 % with a mean of 26.4 %. Similar variations were seen within the three incidence studies, where results ranged from 21.5 to 71.3/1000 person years for those 65 years and older. The statistics for individuals converting from MCI to dementia also demonstrate variation. A meta-analysis by Mitchell and Shiri-Feshki included 41 studies that reported conversion of MCI to dementia and identified progression rates ranging from 1 to 20 % annually [15].

In response to confusion over clinical diagnosis and the resulting inconsistencies within the literature, an updated diagnostic scheme was developed to include amnestic and non-amnestic subtypes of MCI and differentiate between single and multiple domain impairments [6, 16, 17]. There have been interesting findings surrounding prognoses in different groups. For example, those with amnestic types of MCI seem most likely to convert to dementia [18]. Reversion to a non-MCI state may be most likely for individuals with single domain types [18]. It has also been suggested that while most types of MCI progress to Alzheimer’s disease (AD), conversions to non-AD forms of dementia, such as Lewy body and vascular or frontotemporal dementia, are more likely for those with non-amnestic types of MCI [6, 11].

It is apparent that individuals with MCI remain a complex and heterogeneous group with a high level of variability across symptoms, severity, and outcomes. With such diversity, it should be expected that many individuals with MCI would be able to successfully handle a complex task such as driving a vehicle, while others would not. In the next section, we will explore some of the common skills required for safe driving, and take note of the potential for MCI to impact on these areas.

Attention is the most researched cognitive domain associated with driving [19], and researchers have examined a number of types of attention in relation to MCI. For instance, some individuals with MCI have exhibited impairments in focused attention [20]. Focused attention can be conceptualized as basic attentiveness, and the entire process of driving a vehicle is a prime example. As well, selective attention is important to driving; it allows one to focus on certain stimuli while disregarding irrelevant stimuli. For a driver this could mean knowing to focus more attention on the cyclist ahead, while ignoring billboards. Diminished levels of selective attention have been observed in some people with MCI [21, 22]. Another type of attention shown to be impaired with MCI is sustained attention [21]. A driver with diminished sustained attention may not able to focus on the task-at-hand for a long period; in essence, a degraded attention span translates into proneness to distraction. Alternating attention is also important; it involves shifting focus between different types of tasks and could include switching the view of the driving environment using the side- and rearview mirrors. This type of attention has also been shown to be affected by MCI [22]. The ability to multitask while driving (e.g., driving while conversing with a passenger) is considered divided attention, and deficits in this skill have been reported in individuals with MCI as well [21, 23, 24].

In addition to attention, executive function is important to driving in a number of ways. It encompasses the ability to focus on multiple demands at one time (organization) while revising plans as necessary (regulation); in essence, executive function incorporates all of the brain functions that manage, organize, and integrate cognitive skills [25] and generally allows for the performance of tasks required to operate a vehicle. There is little consensus on the processes included under the umbrella of executive function [25], and the effect of MCI on many of these processes is not clear, with outcomes ranging from absence of deficits to global impairments [26]. This may be related to the fact that those with MCI comprise a heterogeneous group and/or the need for the clarification of diagnosis criteria. However, we can look at some of the more important executive functions related to driving that have been identified as being affected by some types of MCI. These include planning [27, 28], problem solving [27], working memory [23, 27, 29], cognitive flexibility [30], task switching [29–31], and response inhibition [23, 29, 30, 32].

Memory is another domain that has the potential to impact on driving. On the most basic level, drivers must be able to remember the rules of the road and how to operate the vehicle. Although not as crucial, drivers should also have the ability to remember the routes to familiar places and the capacity to retain the purpose of a drive. Any of these tasks have the potential to be compromised with MCI (mainly the amnestic types), which can affect familiarity based memory as well as recollection [33, 34]. Difficulties with visuospatial memory are documented [35, 36], as well as diminished episodic memory, delayed recall, and associative memory [37]. It has also been suggested that prospective memory is more affected than recollection [26, 38] and that those with MCI have insight into their memory deficits [39].

Other cognitive, sensory, and physical domains have also been noted in the literature as associated with driving behavior [2]. For example, visuospatial skills have been shown to be important determinants of driving among some clinical populations [40]. While MCI primarily impacts cognitive domains, such as memory, the importance of other changes in functioning should not be neglected by health professionals.

The extent to which the above noted impairments impact actual driving ability in MCI has been assessed in a number of ways. First, some researchers have examined crash incidence. In one of the few studies to look at driving and MCI specifically, Jeong et al. [41] compared the self-reported crashes of MCI drivers (n = 169, operationalized using the Consortium to Establish a Registry for Alzheimer’s Disease neuropsychological assessment battery [CERAD-NP]) to those of healthy controls (n = 623) and could not discern differences between the groups. Prior to the widespread adoption of MCI terminology or classification, a handful of researchers examined the crash rates of individuals with CDR scores of 0.5, which can overlap individuals with MCI and those with very mild dementia. Although the researchers in most of these studies identified their cohorts with CDR scores of 0.5 as having very mild dementia, rather than MCI, these studies may still offer some insight into people on the relatively mild spectrum of cognitive impairment. In 1992, Dubinsky et al. [42] compared self-reported crash rates of controls (n = 98) to participants with CDR scores of 0.5 (“probable AD”; n = 67). According to their results, the CDR = 0.5 participants experienced 26.6 crashes per million miles traveled compared to 8.5 for controls, representing a relative risk of 3.13. In 1996, Trobe et al. looked at the driving records of 143 individuals with a CDR equivalent of 0.5 (“probable AD”) [43] and compared them to matched controls (n = 715) [44]. Although the crash rates for the two groups were similar, the researchers did not include driving exposure in their analysis. Conversely, in 2000, Carr, Duchek, and Morris [45] looked at crash records to compare rates among drivers with CDR scores of 0.0 (“non-demented”), 0.5 (“very mild AD”), and 1 (“mild AD”) and found no statistically significant difference in crash rates between the three groups despite adjusting for driving exposure.

While the limited research on crash rates does not clarify how MCI impacts driving, studies involving on-road evaluations of individuals with MCI provide a better understanding. Wadley et al. [13] performed on-road tests with MCI drivers (clinically diagnosed using Petersen’s criteria; n = 46) and compared their results to a cognitively normal control group (n = 59). Although the drivers with MCI performed statistically significantly worse than controls (in addition to poorer overall scores, those with MCI had more difficulty with lane positioning), the conclusion was that those with MCI did not warrant classification as unfit drivers; they were simply less optimal than controls [13]. In sharp contrast to Wadley’s findings are those of Snellgrove [46] in a report to the Australian Transport Safety Bureau. She compared drivers with MCI (classified using the Alzheimer’s Disease Assessment Scale-Cognitive Subscale [ADAS-Cog]; n = 24) to those with early dementia (n = 92) and found that 50 % of the MCI group and 75 % of the early dementia group failed their on-road test, respectively. According to the researcher, errors were related to poor observation and planning skills, problems controlling the speed and position of the vehicle, pedal confusion, and difficulties with anticipating tasks and defensive maneuvers. However, these results do not indicate how individuals with MCI compare to those with normal cognition and the reported failure rates are out of keeping with the results of Wadley et al. and with the findings of other studies in people with very mild and mild dementia that are presented below.

Like the crash analyses described above, a few older studies looked at the on-road performance of drivers with CDR scores of 0.0, 0.5, and 1.0, designated, respectively, as control, “very mild AD,” and “mild AD” groups. Duchek and colleagues [47] found that those with CDR = 0.5 (n = 21) had performances that fell between the CDR = 1.0 (n = 29) and control (n = 58) groups. Three percent of controls, 14 % of CDR = 0.5, and 41 % of CDR = 1.0 participants were deemed “unsafe” (a global rating assigned to drivers whose behavior resulted in an increased risk of collision). They reported that lane change and signal behaviors were most affected by increasing levels of cognitive impairment. Using a similar design, Ott et al. [48] found that none of their controls (n = 44), 12 % of their CDR = 0.5 group (n = 52), and 22 % of their CDR = 1.0 group (n = 32) were “unsafe” drivers. In Hunt et al.’s study [49] comparing those with CDR scores of 0.5 (n = 36) and 1.0 (n = 29) to healthy controls (n = 58), results were similar to those of Duchek et al. and Ott et al.; three percent of the controls failed the on-road test compared to 19 % with CDR = 0.5 and 41 % with CDR = 1.0.

Research based on driving simulators also contributes to the literature. A study by Frittelli et al. [50] examined simulated driving performance in mild AD (CDR = 1.0; n = 20), MCI (CDR = 0.5; n = 20), and healthy controls (n = 19). Much like some of the results from on-road evaluation research, the performance of MCI participants in this study fell in between that of AD participants and healthy controls. Similar to the findings of Wadley et al., Frittelli and colleagues reiterated the position that while the MCI participants did not perform as well as healthy controls, their impairments had limited impact on driving performance [50] and appeared to be limited to shorter mean times to collision (TTC, a safety indicator which measures the time span before a collision occurs, if no evasive action is taken [51]). In a study by Devlin, McGillivray, Charlton, Lowndes, and Etienne [52], a simulator was used to compare braking patterns between MCI (based on a score of 24–26 on the MMSE; n = 14) and control participants (a score of >26 on MMSE; n = 14). While the MCI group did not perform as well as the controls across most measures, these results were not statistically significant, possibly because of a small sample size. In another small study, Kawano et al. [53] looked at performance on a car-following task, a road-tracking task, and a harsh-braking task using a simulator. Although the amnestic MCI group (CDR = 0.5; n = 12) performed worse than healthy controls (n = 26) on all of the tasks, performance was only statistically significantly different for the car-following task.

In addition to crash risk, on-road, and simulator evaluations, self-reported behaviors such as restriction and driving frequency offer unique insight. Such behaviors may be considered compensatory and act as moderators between driving capacity and actual behavior within some of the previously discussed models of safe driving; drivers who have lost confidence in their skills may elect to refrain from or avoid certain challenging scenarios (e.g., driving at night or in rush hour) and/or maneuvers (e.g., left-hand turns, parallel parking), as well as reduce driving frequency. O’Connor and colleagues investigated these issues over a series of two studies [54, 55]. They found that MCI participants (clinically diagnosed using Petersen’s criteria), when compared to controls, had reduced driving space (i.e., they did not drive as far beyond their home) and frequency, and reported increased driving difficulty in a number of driving situations (e.g., merging, lane changes). Further, those with MCI were more likely to avoid unfamiliar and high-traffic areas. These results are in line with the observation that cognitive impairment, including MCI, may be accompanied by difficulties with orientation and way finding which can lead to a reduction in out-of-home activities [11].

In summary, while the body of research examining MCI and driving behavior is somewhat inconsistent, it suggests that those with MCI are, on average, “less optimal” drivers than those with normal cognition, but a large proportion, if not most, may remain safe to drive. Given the lack of clarity about how these findings translate into real-world driving and given the heterogeneity of individuals who are labeled with an MCI diagnosis, it is obvious that driving recommendations provided to those with MCI cannot be one size fits all. Rather, there is a need to assess driving competence on a case-by-case basis.

While the majority of drivers with MCI may remain safe to drive, there are still instances where further evaluation is necessary. Yet, there is limited guidance for healthcare professionals on how these fitness-to-drive assessments should be done [56, 57]. Guidelines designed to assist healthcare professionals in the assessment of fitness to drive have been identified to have weaknesses in areas of applicability and lack rigor in their development [58]. In fact, there is disagreement in the field about what an assessment of fitness to drive should comprise [59], forcing specialists to rely on a common sense approach rather than on a well-standardized process. A survey of occupational therapists (OTs) in the United States found that there is a wide range of tests utilized to assess older adult drivers [60]. These included measures of vision (e.g., acuity), cognition (e.g., Trail Making Test Parts A and B, Clock Drawing Test), and physical fitness (e.g., range of motion) along with on-road testing. A small number of respondents indicated that they made use of a driving simulator to assess skills. Interestingly, the researchers found that fitness-to-drive assessments often included tools (e.g., test of rules of the road, color perception) that have no evidence indicating they are related to driving skills. It has been argued that one of the ways for older drivers to remain safely on the road is through proper driver screening and assessment [59, 61–63] which could include off-road testing, a simulator assessment, and a behind the wheel evaluation.