Crash rate?

This is an area fraught with methodological pitfalls as demonstrated by Hakamies‐Blomqvist 1998 and as such there is mixed evidence that drivers with mild cognitive impairment (MCI) and mild dementia will have an increased crash rate. Few authors have attempted a prospective study on crash rate. Anderson 2005 found a positive correlation between neuro‐psychological tests (Rey Auditory Verbal Learning Test (AVLT) and Complex Figure Test Recall (CFTR)) with prospective crash rate. They also assessed participants on a driving simulation and a composite score from all the neuropsychological tests, which correlated with each other but not with prospective crash rate. Ott 2008 prospectively evaluated driving performance in participants with early Alzheimer's disease (AD) and healthy controls at six‐month intervals. They found that the rate of motor vehicle accidents was not significantly different at baseline and that over the initial 18 months more crashes were observed in the control group but that after correction for miles driven this difference was not significant.

Crash rates have been suggested to be higher in retrospective studies; however, these studies are dependent on the accuracy of recalled information and results have differed considerably when data are collected from participants, carers or police records. Stutts 1998 looked at cognitive performance and found that although the association with retrospective, state‐recorded crashes in older adults was present the effect was small (less than two‐fold increase comparing the top and bottom deciles on cognitive scores) and they were not able to identify a cut‐off point that identified a significantly higher risk of crash. Trobe 1996 and Carr 2000 suggested that there was no increased risk of crash in people with dementia who's Clinical Dementia Rating (CDR) (Morris 1994) score was between 0.5 and 1 when looking at state recorded crash rates rather than those recalled by subjects, carers or relatives. This may represent a difference in the types of crashes reported to local authorities and although we can reasonably infer that more serious crashes are more likely to be recorded and that retrospectively recalled events may be less accurate, we have no comparative data on these systems of reporting. Parker's paper has built on previous evidence to show that driving errors are not well correlated with future crashes but previous violations were (Parker 2000). This may go some way to explain the discrepancy between high failure rates on road tests and low reported accident rates. Retrospective odds ratios (OR) for crash comparing drivers with dementia and with no dementia have been reported from 7.9 by Friedland 1988 to 10.7 by Zuin 2002. These studies show a marked difference with respect to numbers of subjects, which may help explain the variable strength of the associations (Trobe: n = 858, Carr: n = 121, Zuin: n = 87, Friedland: n = 50). A much larger study involving 3238 older drivers applying for re‐licensing had cognitive tests (Trail Making tests A and B, American Association of Retired Persons Reaction Time test and Short Blessed Cognitive screen) compared with State crash records for the previous three years (Stutts 1998). In that time frame there were 411 crashes including 97 crashes by 45 people, an annual average of 0.043 crashes per person. The investigators found an OR of approximately 1.5 for crashes comparing drivers with cognitive performance in the lowest decile to those in the highest decile. Unfortunately this study did not document a history of dementia in any driver or use cognitive assessments that would allow a diagnosis of dementia, so extrapolating these data to drivers with dementia is difficult. We might infer that the lowest decile of cognitive performance group included at least some people with dementia but the magnitude of the effect on crash risk is unknown.

Despite the fact that it is a study of older drivers and does not diagnose or identify a diagnosis of dementia in any participant or group, the Maryland Prospective Older Driver Study (Staplin 2003) can be useful to demonstrate some of the difficulties encountered in driving research. In this paper, 2508 adults over 55 years of age were screened using cognitive and physical performance measures in Maryland. Subjects were recruited by random invitation from all licensed drivers, by medical referral and at re‐licensing at a senior citizens residential area. This had the advantages of a prospective follow‐up for a mean of 20 months following assessment for crash and moving violation outcomes in a large population‐based sample. They suggest, based on peak ORs for crash in each of four cognitive tests that cut‐off points can be identified for increased risk of crash and high risk requiring intervention. ORs for crash at the following cut‐points were: Motor‐Free Visual Perception Test (Visual Closure subtest) (MVPT/VC): 5 incorrect, OR 4.96; Trail Making Test, Part B: 180 sec, OR 3.5; Cued/Delayed Recall: 2 incorrect, OR 2.92; Useful Field Of View subtest 2: 300 msec, OR 2.48. Using raw data in the paper we may extrapolate some values for sensitivity and specificity based on the strongest predictor, MVPT/VC, as a screening tool. MVPT/VC of 5 or more correctly predicted 18 crashes had 258 false positives, 93 false negatives and 1503 true negatives in the 1872 participants who had valid test results. This gives us a positive predictive value of 93%, negative predictive value of 94% and a sensitivity and specificity of 83% and 85%, respectively. If we look at the test as an intervention to prevent crashes, then we will need to screen 143 older drivers in order to prevent one crash in the following 20 months and an additional 20 drivers will fail testing.

Driving test performance?

There are reports that drivers with MCI and AD will have a high failure rate on driving tests from which we might infer a higher risk of crash; however, rates of driving test failure vary considerably and, with the exception of Anderson's paper quoted above (Anderson 2005), none of these estimates have been validated prospectively. Snellgrove 2005 reported the highest driving test failure rate in a non‐peer reviewed paper for the Australian Transport Safety Bureau. However, this was a highly selected group (consecutive referrals to a memory clinic) with no control group but they did demonstrate an extremely high rate of failure on their road test: 70% of the group of 117 subjects with predominantly early dementia (80%) and MCI (20%). Of particular concern was their finding that 50% of the group required a physical intervention by the assessor to prevent a crash (a finding that has not been reported in any other paper). They also found that 50% of the MCI group failed the road test. Similarly, 41% of 55 participants with dementia were reported to fail the road test in Clark 2005 and 63% of 19 subjects for Fox 1997, 36% of 96 subjects with cognitive impairment in Kay 2009, 65% of 99 subjects with dementia in Carr 2011 but only 18% of 65 drivers with dementia in Lincoln 2010. Comparative data versus controls were found in Duchek 2003, which gives test failure percentages of 30% versus 3% for people with and without dementia, and from Lincoln 2006, 37% versus 0%. Ott 2008 showed an increased hazard ratio for test failure of 3.51 in people with established dementia (CDR 1.0) versus early dementia (CDR 0.5) in a prospective study over 18 months. In addition, the CDR 1.0 group also showed a higher rate of driving cessation: only 31% of the CDR 1.0 presented for re‐test at 18 months compared to 48% of CDR 0.5. Road test failure rates in Berndt 2008 were 34% for CDR 0.5, 58% for CDR 1.0 and 95% for CDR 2.0. Interestingly one person with CR 2.0 passed the road test suggesting that there may be some merit in offering a road test to people who wish to avail of the opportunity irrespective of their cognitive testing.

Driving simulator performance?

Simulators have also demonstrated higher rates of adverse driving behaviours in studies by Rebok 1994, Cox 1998, Frittelli 2009 and Vaux 2010 and also crashes (Rizzo 1997; Rizzo 2001; Uc 2006, Frittelli 2009). However, some authors believe that performance in driving simulators is not strongly related to on‐road driving performance (Bylsma 1997;Rizzo 2001).

Neuropsychology

Although studies of older drivers without dementia have shown evidence that office‐based tests correlate with future crashes, this has not been evaluated in drivers with a diagnosed dementia. What can be said is that in studies that included people with diagnosed dementia, office‐based tests correlate, grossly and without helpful cut‐off points, with road test performance and driving simulator performance. Also many studies have shown the strongest correlation between road testing requires a composite of multiple tests (Rizzo 1997; Anderson 2005;Clark 2005; Lincoln 2006), which may limit their use to trained individuals and require lengthy assessments. Clark's composite was more realistic (MMSE, Trail Making test‐A and Wechsler Adult Intelligence Scale‐block design) with a sensitivity and specificity of 82% and 90%, respectively, but it remains to be established that such an assessment is independently adequate to disqualify drivers or merely a screening tool to identify those requiring road testing. There is little evidence that simple tests alone are as effective as road tests. Their most likely role is in screening to identify those in need of more detailed evaluation. Even with composite tests, such as the Stroke Driver Screening Assessment (SDSA) (Nouri 1992), adjustment has been needed for stroke in international populations (Lundberg 2003), and have needed to be used in other combinations of instruments for Parkinson's disease, traumatic brain injury (Radford 2004) and dementia (Lincoln 2006). The Nordic version of the SDSA has been shown to be effective in evaluating stroke drivers but had poor sensitivity and specificity in drivers with dementia (Selander 2010).

Behaviour/driving specific

The model of assessment suggested by Michon 1985 and operationalised by de Raedt 2000 uses a hierarchical and behavioural approach to the driving task. Michon's model of selection, optimisation and compensation describes successful adaptive strategies in individuals with disability. Examining driving behaviour in similar terms may help identify those able to overcome increasing levels of disability ‐ physical or psychological. De Raedt's model examined driving behaviour in three hierarchies: strategic behaviours involved in planning destinations and routes and avoiding hazardous weather or traffic conditions; tactical refers to anticipatory behaviours such as speed adaptation; operational behaviour is the physical performance of the task (examined in most studies). De Raedt showed high levels of strategic behaviours in those performing badly on road tests but who did not have high accident rates. High accident rates were more associated with low tactical performance scores. This adds an extra dimension to assessment and may help explain the apparent discrepancy between high test failure rates and low per capita crash rates as well as some of the variability in the studies quoted. This model was used by Grace 2005 in a small study with 21 in each group comparing participants with AD, Parkinson's disease and healthy controls. In this study, test failure was rare: only two of the AD group and none of the other two groups failed the road test. However, errors were more common in AD and Parkinson's disease groups, with strategic, tactical and operational errors observed in people with AD while only tactical errors were observed in people with Parkinson's disease. This may suggest that a single standardised test will not be suitable in all older drivers or even all neurologically impaired drivers and a tailored approach may be necessary. Another approach that has more congruence with modern concepts of driver behaviour is that of self efficacy, and an early study in stroke (an illness with high levels of cognitive impairment) has shown promise (George 2007).

On‐road tests

Many studies have used on‐road testing as their gold standard (Hunt 1993; Fitten 1995; Fox 1997; Hunt 1997; Dobbs 1998; Snellgrove 2000;Duchek 2003; Wild 2003; Brown 2005;Clark 2005; Grace 2005; Ott 2005; Uc 2005;Lincoln 2006;Berndt 2008;Ott 2008;Dawson 2009;Kay 2009;Okonkwo 2009;Lafont 2010;Lincoln 2010;Patomella 2010;Selander 2010;Barrash 2010Carr 2011). The road test lacks an accepted standard and considerable variation exists in term of vehicles, routes and tasks. More recent studies have begun using standardised and validated road tests such as the Sepulveda Road test (Fitten 1995), the Washington Road Test (Hunt 1997) or the Test‐ride for Investigating Practical fitness to drive or TRIP (Tant 2002) designed specifically for older driver assessment. Prospective studies will be required to define cut‐off scores that predict crash risk.

Driving simulators

Driving simulators have been used in many studies (Rebok 1994; Rizzo 1997; Cox 1998; Rizzo 2001; Anderson 2005; Uc 2006; Frittelli 2009; Vaux 2010). Simulators have obvious advantages, allowing detailed assessment in a safe and controlled environment; however, they have not been shown to predict prospective crashes and some authors feel they are not correlated well with the driving task in older drivers and drivers with dementia (Bylsma 1997; Rizzo 2001). There is also considerable variation in the technology involved ranging from computer‐based simulations (Rebok 1994) to scale‐model vehicles that simulate audiovisual data in addition to momentum (Rizzo 2001). Tolerability may also be problematic as simulators need to have verisimilitude but older people also get more motion sickness (Edwards 2003).

Dementia type

AD is specifically evaluated in most studies although many of the studies did not specify which dementia subtypes were within the experimental group or did not use consistent and clinically useful criteria (such as CDR or NINDS) to document disease severity. Vascular dementia (VaD) is rarely tested and then usually as a part of an unselected dementia population. No subgroup analysis or specific studies of driving in VaD were identified. Fitten 1995, Zuin 2002, Clark 2005 and Lincoln 2010 were the only studies identified that specifically mentioned including a number of subjects with VaD but did not look at differential outcomes due to small numbers. Fronto‐temporal dementia (FTD) has only been evaluated in one study (de Simone 2007), which showed a higher rate of crashes and errors in a simulated drive but only enrolled 15 subjects and 15 controls. Zuin 2002 did comment on a particularly high incidence of adverse driving behaviour in FTD but this only accounted for two out of 87 subjects and no further analysis was attempted. Lincoln 2010 also included  a small number of people with VaD, FTD and dementia with Lewy bodies (DLB) but the numbers were too small to yield significant results. Other forms of dementia are almost never evaluated in studies despite reasonable suspicion of higher risk with deficits in perceptual skills and impaired judgement and disinhibition, for example, in DLB. It would seem logical to suggest that deficits of people with VaD, FTD and DLB may not correlate well with the studies on people with AD and would need specific studies to validate driving assessment techniques.

Dementia rating

Use of clinically relevant instruments such as CDR to rate dementia severity have helped to categorise drivers who perform poorly in some studies (Hunt 1993; Duchek 2003; Ott 2008) and by incorporating functional and behavioural measures may have stronger predictive value. As discussed earlier, Ott 2008 showed a significant difference in prospective driving performance according to CDR score reinforcing the value of road testing in earlier stages of AD and in at least one case in a person with a CDR score of 2.

Functional tests

Functional measures, such as delayed recall, Trail Making test, visual perception, useful field of view, lower limb strength, and head and neck mobility, used in the Maryland Pilot Older Driver Study (Staplin 2003a), are potentially reliable predictors of crash risk with clear cut‐off scores thus far only evaluated in an unselected older driver population. It bears mentioning though in that this is the first study that had a component looking at mobility and driving cessation. Newer data from the study, however, indicate that the predictive value of functional tests may decrease with increasing time since assessment, although the predictive value of delayed recall and rapid pace walk was preserved (Staplin 2003).

Collateral history

Collateral history is useful but may be misleading and subject to positive and negative bias. Wild 2003 showed more accurate prediction of driving ability from carers than people with AD but not for healthy older people. However, carers' assessment of functional impairments can be influenced by the burden placed on a carer's time (Zanetti 1999) and carers may have vested interests in a person's continued driving, in the example of a non‐driving spouse. This may explain the higher crash rates estimated by carers as found in Friedland 1988. Despite this, dementia research in other fields lends credence to the collateral history, which can be a more accurate estimate of cognitive and functional decline (Archer 2007).

Other approaches to managing driving and dementia: mandatory reporting and voluntary reporting

A study from Oregon showed that, where physician reporting of impaired drivers was linked to mandatory suspension from driving, only 18.5% of 1664 drivers requested a re‐test or appeal against suspension and that 56% of those re‐tested regained driving privileges (Snyder 2009). The majority (88.8%) of these automatic suspensions were for cognitive impairment. No data have been presented on whether this measure improves safety on Oregon's roads.

Meuser 2009 described a system where introduction of voluntary reporting of impaired drivers in Missouri led to driving cessation in 96.5% of 4100 reported individuals, 50% of whom did not seek any further assessment. This measure did lead to a reduction in crashes in this group but the study did not look at the impact on mortality or morbidity outcomes for this group other than motor vehicle accidents.