Driving Assessment for Persons with Dementia: How and when?

Dementia is a progressive neurodegenerative disease leading to deterioration in cognitive and physical skills. Driving is an important instrumental activity of daily living, essential for independence. However, this is a complex skill. A moving vehicle can be a dangerous tool in the hand of someone who cannot maneuver it properly. As a result, the assessment of driving capacity should be part of the management of dementia. Moreover, dementia comprises of different etiologies and stages consisting of different presentations. As a result, this study aims to identify driving behaviors common in dementia and compare different assessment methods. A literature search was conducted using the PRISMA checklist as a framework. A total of forty-four observational studies and four meta-analyses were identified. Study characteristics varied greatly with regards to methodology, population, assessments, and outcome measures used. Drivers with dementia performed generally worse than cognitively normal drivers. Poor speed maintenance, lane maintenance, difficulty managing intersections and poor response to traffic stimuli were the most common behaviors in drivers with dementia. Naturalistic driving, standardized road assessments, neuropsychological tests, participant self-rating and caregiver rating were the most common driving assessment methods used. Naturalistic driving and on-road assessments had the highest predictive accuracy. Results on other forms of assessments varied greatly. Both driving behaviors and assessments were influenced by different stages and etiologies of dementia at varying degrees. Methodology and results in available research are varied and inconsistent. As a result, better quality research is required in this field.


1)
To determine whether driving competence is dependent on performance in secondary driving tasks.
2) To assess whether cognitive status affects driving outcomes. Navigation-related secondary task performance (routefollowing and landmark/ sign identification tasks), On-task safety errors, Baseline safety errors.
AD patients were worse than PD patients at identifying stop signs. The diseased group (AD + PD) performed worse than the healthy group in global secondary task performance, global landmark/sign identification, and global on task safety route following, baseline safety, lane observance, turns, visuosensory function, visuospatial construction, memory, processing speed and executive functioning.
Visuospatial construction predicted baseline safety errors per mile. Processing speed predicted on-task These are often linear models depicting progression from one stage to another which underestimate the true complexity of the driving process They explore multiple cognitive domains including perception style, reaction time, attention, vigilance and decision making

Cybernetic models
It describes how the sensory environmental input affects the driver's processing and decision making which are reflected in voluntary motor action directing car movement. Figure 1 is an example of how this model can be applied to driver steering movement Divided into 4 main parts: 1. Anticipatory moduleresponsible for input from visual stimuli of the road's geometry and environment from a far point 2. Compensatory moduleresponsible for input of the road geometry and environment but from a near points 3. Processing time and speed of the input information 4. Neuromuscular system -produces the output based on the stimuli and processing information

Hierarchal control models
Namely focus on the interaction between the driver and their environment.
Two main models: 1. The skill-rule-knowledge model identifies different levels of driving behaviours which can categorise driving errors as shown in figure 2. 2. The three-level hierarchical model divides its three levels strategical, tactical and operational levels.

Motivational Model
These models focus on the driver's performance based on the driver's subjective risk-taking behaviour in relation to the particular environment or situation. safety errors. Visual sensory functioning, memory and executive function predicted safety in navigation-related secondary task performance. Crivelli L, Russo MJ, Farez MF, Bonetto M, Prado C et.al. (2019) To produce a driving assessment protocol for drivers with mild dementia by identifying the neurophysiological tests that best predict driving competence. AD drivers made more mistakes than HC during the on-road driving assessment and the traffic signal recognition score. AD patients also had lower traffic signal reaction scores and brake time reaction scores on simulator test. Difficulty in recognising the following signs was more prominent in the AD group: "No entry for vehicles" "U-turn prohibited" "No overtaking" "Narrow road" "Pedestrian crossing" Croston J, Meuser TM, Berg-Weger M et.al. (2009)  Only 29% of drivers with dementia were active drivers. 53% of caregivers rated the dementia participants' driving ability as fair, poor or unsafe prior to driving cessation.
Behaviours that were more common before cessation included: Drove closer to home, Reduced frequency of driving; 70% reported at least one abnormal driving behaviour, the commonest being failure to monitor traffic followed by speed maintenance, difficulty with intersections or turns, backing up and lane maintenance.
Barriers to driving cessation included lack of insight or personality issues in the driver with dementia (33%) followed by caregiver belief of good driving capacity (17%) and risk of social isolation (14%). Davis JD, Wang S, Festa EK et.al. (2018) To identify unsafe behaviours or distracted driving with a potentially high crash risk in cognitively impaired drivers. 44 possible or probable AD  16 Healthy controls (HC) NINCDS-ADRDA criteria for AD. CDR scale for severity rating.

Rhode Island Road Test (RIRT)
Naturalistic Driving Assessment: CDAS was used.

Automated computerised analysis of naturalistic driving
The Modified Mockingbird Event Scoring System was used.

Miles driven per week and crash history.
AD patients drove fewer miles/day when compared to HC.
When controlling for mileage driven, AD participants had higher error scores than HC participants.
AD patients made more severe errors for lane maintenance and not looking far enough ahead to anticipate traffic situations and manoeuvres. HC participants made more severe errors for behaviors related to distraction, failing to keep an out, and generally riskier behaviors. Davis JD, Babulal GM, Papandonatos GD et.al. (2020) To identify driving errors in drivers with early AD.
To compare driving behaviors in drivers with early symptomatic AD to cognitively normal (CN) drivers with preclinical AD and healthy adults without evidence of AD. When controlled for miles driven, speeding was the most common driving behaviour in the HC and AD dementia groups while hard braking was the most common event in the preclinical AD group. Dawson JD, Anderson SW, Uc EY et.al. (2009)  Failing to proceed through intersection even though the light had turned green. AD drivers make more total safety errors, lane observance errors, and serious safety errors than elderly drivers without AD.
A neurophysiological test score combining performances across multiple cognitive domains, was a better predictor of driving safety in AD when compared to single domains. Visuospatial and motor response test were also good predictors of driving safety in AD. de Simone V, Kaplan L, Patronas N et.al. (2007) To examine the driving behaviour of FTD patients, especially the impact of personality changes and deficits of social cognition on driving safety in FTD patients. FTD patients had greater speed variability, with significantly higher number of speeding tickets. FTD patients remembered fewer words on both the free recall and cued recognition tests.
Amongst FTD patients, 60% had collisions, 47% had off-road accidents, and 33% ran stop signs. In contrast, none of the control subjects ran a stop sign or had a collision, and only one had one off-road accident. Dementia severity was not correlated with any driving measure.
Duchek JM, Carr DB, Hunt L et.al. (2003) To investigated longitudinal driving performance in healthy aging and early-stage Alzheimer's Dementia at a 6 monthly interval over 2 years. HC drivers took a significantly longer time to receive a rating of not safe compared to Mild AD patients (CDR 1). The time for the very mild AD drivers to become unsafe fell between HC and th mild AD groups but the difference was not statistically significant.
The driving behaviours that were statistically correlated with increasing dementia severity (higher CDR scores), were impairment in lane change and signals. The driving behaviours that showed a statistically significant decline over time were qualitative judgement and speed control but were not affected by CDR scores.
There was a longitudinal decline in driving skills across all three groups of drivers.
Three driving behaviors showed a significant decline for all groups: qualitative judgments, reaction to others, and speed control. Baseline age was a significant risk factor for receiving a rating of not safe. Economou A, Pavlou D, Beratis I, et al. (2020) To identify driving variables that predict crash risk in drivers with mild AD.

Near-infrared spectroscopy (NIRS):
Used to measure brain activity during the simulator test by detecting changes in cerebral blood flow. Bilateral frontal and temporal regions were examined.
The reaction time and force of breaking was significantly lower in the AD group when breaking was needed to avoid collision.
There was no significant difference in brain activity between AD and HC participants during routine driving. There was also no significant activation in the prefrontal cortices during baseline driving. During the collision avoidance scenario, there was a significant difference in frontal activation between HC and AD groups. Prefrontal activation was evident in all four scenarios in the HC groups but was diminished in the AD participants.

Standardised road test using ARGOS instrumented vehicle
The following measures included: 1) Incorrect turns; 2) Times lost (incorrect turns which the driver did not recognise and correct); 3) At-fault safety errors.
AD drivers made significantly higher driving errors when compared to HC drivers. RFTs increased the likelihood of committing safety errors in AD drivers. The HC group performed significantly better than both AD group on the onroad driving test but there was no statistically significant difference between the very mild (CDR 0.5) and mild (CDR 1.0) AD groups.
When compared to on road test scores, physician rating was most accurate followed by informant rating, with participant rating being the least accurate (in both very mild and mild AD groups). Burns T, Lawler K, Lawler D et.al. (2018) 1) To assess the predictive value of the CPT for driving competence.
2) To assess the accuracy of cognitive screening tools in classifying cognitive impairment as mild versus major.

Caregiver and Participant Perception of driving ability via AANQ
Simulator driving assessment: traffic signal reaction tasks and brake reaction tasks.

Traffic signal recognition task
AD patients performed worse on cognitive assessment when compared to HC. The driving questionnaires revealed no difference in driving ability between AD and HC group in both participant and caregiver answers.
The TMT-B, Verbal semantic Fluency and FDS were identified as the most important predictors of driving performance which statistically significant correlation with all driving scores. The TMT-A had a statistically significant correlation with both simulator tests and on road test but not the traffic signal recognition task.The MMSE, Logical Memory, DSMT, BNT, RAVLT, ROCFT, FAB, NPI-Q and FAQ had a statistically significant correlation with the on-road testing, traffic signal recognition task and the simulator traffic signal reaction task but not brake reaction task.
Age, DSMT and BNT were the variables that best predicted performance on the driving test. No single variable was enough or better than the other to obtain a cut off score but a good performance on two of these test was an indicator of good driving performance. Davis

Automated computerised analysis of naturalistic driving
The Modified Mockingbird Event Scoring System was used.

Miles driven per week and crash history.
Total error score, corrected for mileage, was significantly correlated with the road test error score. Mockingbird error scores from automated analysis, was correlated with clinic measures of cognitive functioning with a modest correlations between error scores and MMSE measures, time to perform mazes, and number of errors on mazes. There was no relationship between error scores and CDT, TMT-A, TMT-B. The Mockingbird scoring of discrete events in automated analysis, achieved the highest sensitivity and specificity in predicting AD diagnosis.
The automated, eventbased method is a valid method for driving assessment in AD, but it is more clinically informative when examining multiple relevant behaviors simultaneously rather than relying on global error scores. In-office cognitive measures had significant, but modest, associations with driving errors. Davis

Motor skills measures: Functional
Reach for balance impairment. The timed Get-Up-and-Go test. The Grooved Pegboard task was used to measure dexterity and motor speed.
A higher overall cognitive function was associated with a better safety error score in the AD group. The following cognitive tests were significant predictors of driving errors in AD: BVRT (working memory) TMT-A (visual search and visual motor speed) CFT-copy (visuoconstructional ability) Functional Reach (motor function) Lane observance errors were the most common type of errors in AD group, so cognitive test scores were compared to this type of error separately and the following cognitive tests showed significant correlation: CFT-COPY, UFOV-total and Functional Reach. de Simone V, Kaplan L, Patronas N et.al. (2007) To examine the driving behaviour of FTD patients especially the impact of personality changes and deficits of social cognition on driving safety in FTD patients.

Standardised road test used.
Driving evaluation was divided in three parts: 1) The Driving Researcher Score (DRS); 2) The Driving Instructor Intervention Score (DIIS); 3) The Driving Instructor Judgment (safe, borderline or unsafe).
30% of AD drivers were deemed unsafe compared to 1.8% of HCs.

Caregiver-rated driving ability scale
In part A of the study, the time to complete the Proteus Maze Drawing was the only significant correlation to Caregiverrated driver ability.
In part B of the study, the total score each of the 10 Maze tasks, dementia severity as measured by the CDR score, education and number of crashes, were significantly correlated to Caregiverrated driving ability.
O'Brien HL, Tetewsky SJ, Avery LM et.al. (2001) To analyse the perceptual mechanisms of visuospatial disorientation in AD. AD subjects had impairment in left/right radial motion perception. 85% of AD and 33% of EHC exhibited impaired use of global radial patterns as opposed to YHC who did not exhibit any significant impairment.
Global pattern recognition is impaired first before clinically evident AD which forces all AD subjects to rely on local rather than global cues. Eventually, local motion processing also becomes impaired in clinically evident AD. Performance in the naturalistic setting was mostly correlated with two distinct aspects of driving namely, behaviors related to proper land keeping and responding to traffic and manoeuvring the vehicle. Performance on road test setting was correlated with one rather than the two driving behaviours, namely driving awareness items. Results were adjusted according to route difficulty, and these showed that similarity in the driving environment rather than measurement method (RIRT vs CDAS), is more significant to ensure a strong association between naturalistic and on road test performance. Driving skills assessed during onroad testing are not fully reflective of the full driving skills used during naturalistic driving. Divided in two parts: 'On-task' segment to assess route following tasks (RFT) and the 'no-task' segment typical of a standard road test. The following measures included: 1) Incorrect turns; 2) Times lost; 3) At-fault safety errors. AVLT-RECALL and UFOVTOT were significant predictors of incorrect turns and times lost.
AVLT-RECALL, CFT-COPY, and CS were significant predictors of at-fault safety errors. Uc EY, Rizzo M, Anderson SW et.al. (2006) To assess the behaviour of drivers with AD when faced with a potential risk of rear-end collision. Divided in two parts: 'On-task' segment to assess landmark and traffic sign identification tasks (LTIT); and the 'notask' segment typical of a standard road test. AVLT-Recall and SFM were predictors of at-fault safety errors. Vaux LM, Ni R, Rizzo M et.al. (2010) To assess the ability of AD and PD patients to detect impending collisions. The correlation between global cognitive ratings and rater score, total errors, and number of crashes was not statistically significant.
As a result, measures of global cognitive functions are non-specific and should not be used in isolation to predict fitness to drive in dementia. On the other hand, VOSPobject scores and TEA can give important information on driving safety in drivers with AD dementia. All cognitive domains except language, showed a moderate significant correlation with on-road performance.

Supplementary
Language showed a moderate but nonsignificant correlation.
As a result, measures of memory, attention, visuospatial skills and executive function, are associated with on-road driving performance in drivers with dementia.

Supplmenetary Table 5. Quality Appraisal and Risk of Bias of Observational Studies.
(*Red = High risk, Green = Low risk, Yellow = Uncertain level of risk due to insufficient information) Author  (1) Park, G. D., Cook, M. L. and Fiorentino, D., 2007 (1) T et.al. (2015), 'Investigation into the safety of driving by individuals with higher brain dysfunction' Kawasaki Medical Journal 41 (2) (2017)