Accident Analysis and Prevention 50 (2013) 483– 489
Contents lists available at SciVerse ScienceDirect
Accident Analysis and Prevention
j ourna l h o mepage: www.elsev ier .com/ locate /aap
he Maze Test: A significant predictor of older driver crash risk
oren Staplina,∗, Kenneth W. Gisha, Kathy H. Lococoa, John J. Joycea, Kathy J. Sifritb
TransAnalytics, LLC, 336 West Broad Street, Quakertown, PA 18951, United StatesU.S.DOT/National Highway Traffic Safety Administration, 1200 New Jersey Ave., SE, Washington, DC 20590, United States
r t i c l e i n f o
rticle history:eceived 17 February 2012eceived in revised form 30 April 2012ccepted 22 May 2012
eywords:rash risk
a b s t r a c t
A study sponsored by the National Highway Traffic Safety Administration performed functional assess-ments on approximately 700 drivers age 70 and older who presented for license renewal in urban,suburban, and rural offices of the Maryland Motor Vehicle Administration. This volunteer sample receiveda small compensation for study participation, with an assurance that their license status would not beaffected by the results. A comparison with all older drivers who visited the same sites on the same daysindicated that the study sample was representative of Maryland older drivers with respect to age and
prior driving safety indices. Relationships between drivers’ scores on a computer touchscreen versionof the Maze Test and prospective crash and serious moving violation experience were analyzed. Resultsidentified specific mazes as highly significant predictors of future safety risk for older drivers, with aparticular focus on non-intersection crashes. Study findings indicate that performance on Maze Testswas predictive of prospective crashes and may be useful, as a complement to other, established cognitive
ying
afety screening tools, in identif
. Introduction
Motor vehicle crash statistics show that, relative to their milesriven, older persons are at greater risk of fatal crash involve-ent than any group except newly licensed, teenage drivers; in
art, this reflects their increasing frailty, such that crashes of com-arable severity result in more bodily harm for an older than aounger person (IIHS, 2008). These data also appear to reflectge-related declines in the visual, cognitive, and physical abilitieseeded to safely operate a motor vehicle in everyday traffic condi-ions (Staplin et al., 2003), spurring research into which domains ofunctional ability significantly predict crash involvement by olderrivers, and how best to measure them.
Perhaps the most urgent research need in this regard is to iden-ify reliable measures for detecting drivers at elevated crash riskue to mild cognitive impairment (MCI) or to early stages of demen-ia. Our population is rapidly aging, and it has been estimatedhat 13% of persons over 65 and 45% of persons over 85 will beffected by Alzheimer’s Disease (Alzheimer’s Association, 2011).s reviewed by Carr and Ott (2010), crash studies indicate thatrivers with a dementia have at least a 2-fold greater risk of crasheshan cognitively intact older adults; but it should be noted that
vidence (cf. Fitten et al., 1995) suggests that it is the degree of cog-itive impairment rather than type of dementia (diagnosis) that is theore important determinant of risk. Accordingly, a standardized
measurement technique with strong sensitivity to tap those mostpertinent cognitive abilities and establish a performance thresholdor cutpoint that significantly predicts older driver crash involve-ment would clearly be of value in diverse clinical and, potentially,regulatory settings.
It is not just the degree but the type of cognitive impairmentthat predicts driving difficulties. Reger et al. (2004) performed ameta-analysis of neurological tests and driving that highlightedthe importance of measuring visuospatial skills, compared to othercognitive domains (e.g., memory), to discriminate differences inon-road tests of driving ability in persons with dementia. A promi-nent example of such tests that also draws upon the ‘executivefunctions’ of planning and foresight (Snellgrove, 2005) as wellas judgment and visual attention (Ott et al., 2008) is the MazeTest.
Within this (visuospatial) domain, the Maze Test also standsout with respect to consistency of test administration methodsand scoring protocols, in contrast to such alternatives as the ClockDrawing Test for which at least half a dozen different scoring cri-teria are documented in the literature (Lam et al., 1998; Mendezet al., 1992; Shua-Haim et al., 1996; Shulman, 2000; Sunderlandet al., 1989; Wolf-Klein et al., 1989). In addition, the Clock DrawingTest has been shown to be relatively poor at detecting milder cog-
nitive impairment in older community-dwelling adults (Nishiwakiet al., 2004) while the Maze Test appears to effectively discrimi-nate between persons with mild dementia1 and healthy controls
1 Clinical Dementia Rating (CDR) scale score of 0.5 or 1.0.
Ott et al., 2008). It could be asserted that the Clock Drawing Testnd the Maze Test likely overlap to some degree in the constructshey are tapping; and, researchers have demonstrated significantorrelations between impaired clock drawing performance anderformance on a low-fidelity driving simulator (Freund et al.,005) and on an on-road driving test (Oswanski et al., 2007). Theaze Test remains the tool of choice for the application examined
n this research, however.The research carried out by Snellgrove (2005) was based on the
orteus Maze Test, a timed paper-and-pencil procedure. Perfor-ance, measured in terms of time (seconds) to successfully draw a
ine from the beginning to the end point of the maze, and total num-er of errors (entering a dead-end alley or failing to stay within the
ines) discriminated with high accuracy (77.8% sensitivity and 82.4%pecificity) between participants who failed an on-road driving testnd those who passed it. Participants included 115 community-welling older drivers age 65 and older (mean age 76.9 years) withither mild cognitive impairment (MCI) or probable (early) demen-ia. The author cited anticipatory and defensive driving skills inxplaining the predictive validity of the Maze Test as demonstratedn this research.
Ott et al. (2003) found that Porteus Mazes were the only sig-ificant predictor among a battery of standard neuropsychologicalests of caregivers’ ratings of driving ability for an older sampleith questionable to mild dementia. Next, Ott et al. (2008) exam-
ned the ability of Maze Tests (five separate mazes) to predict roadest performance for 121 drivers ages 40–90, including a ‘possi-le’ Alzheimer’s Disease (AD) group, a ‘probable’ AD group, andealthy controls. Subjects completed an on-road drive test basedn the Washington University Road Test (WURT), with a drivingnstructor blind to the subject’s diagnosis. In both studies, Ott andis colleagues used an innovation: computerized mazes. Subjectsrew lines on a touchscreen instead of using the paper-and-pencilethod. For all groups, based on a logistic regression model that
lassified road test performance as safe versus marginal versusnsafe, total maze completion time accounted for 15% of the vari-nce with a correct classification rate of 68.6%. Considering onlyhe mildly cognitively impaired sample (CDR = 0.5) and the con-rols, total maze completion time accounted for 23% of the variance,ith a correct classification rate of 76.5%.
Finally, Carr et al. (2011), in collaboration with Snellgrove, Ottnd others, found that the Snellgrove Maze Test was a significantredictor of passing or failing the modified WURT, in a sample of9 older people with dementia (63% male, mean age 74.2) referredy community physicians to an occupational therapy driving clinic.easures of visual and motor functioning were not associated with
oad test failure.This article describes an investigation that advances our under-
tanding of the validity of the Maze Test for predicting olderriver crash risk. Researchers used the same stimuli employedy Ott et al. (2008) in a standardized, computer-based test pro-ocol to assess a much larger sample of drivers, age 70 and over,ho were representative of the general older driver population in
heir State in terms of recent driving history. Most importantly,rospective crash experience—rather than a driving performanceeasure serving as a safety surrogate—was the dependent vari-
ble in this study. The central research hypothesis was that driversho required longer times to complete mazes (by tracing a con-
inuous path from the start to the end), or who committed morerrors during maze drawing (evidenced as ‘dead ends’ where aubject was required to discontinue the path s/he was followinghrough the maze and shift to another path), would demonstrate
significantly higher risk of crash involvement and/or of citationor the most hazardous types of moving violations in an 18-
onth observation period keyed to each participant’s assessmentate.
Prevention 50 (2013) 483– 489
2. Research method
Our research team recruited 692 drivers for this study frompersons who visited one of four Maryland Motor Vehicle Admin-istration (MVA) field offices to conduct business (license renewal,title transfer, etc.) between September 2008 and June 2009.All persons age 70 or older with a valid Maryland driver’slicense were eligible to participate. The study sites includedone large city (Baltimore), one small city (Annapolis), one sub-urban location (Loch Raven/Parkville), and one rural location(Easton). Recruitment and assessment activities were discon-tinued at the Annapolis MVA office in November 2008, dueto volumes that were much lower than anticipated; the otherthree sites remained active for the duration of data collec-tion.
Initial contact to recruit study participants took place in oneof two ways: a counter staff member at the MVA told poten-tial participants about the study and provided a research flyer;or, the MVA mailed a letter to older drivers in the geograph-ical catchment area of each field office whose license renewaldate was approaching in the next month, advising them of thisresearch opportunity. Both methods directed interested personsto project research assistants (RAs) on-site at each MVA officefor more information. These RAs were trained by the lead authorto administer the data collection protocol, then practiced inpairs under his supervision, prior to their interactions with olderdrivers.
The RAs enrolled potential subjects who received informationabout the research opportunity and indicated an interest in par-ticipating. Recruitment procedures, including informed consentprocedures, were carried out according to protocols approved bythe Institutional Review Board at Chesapeake Research Review.Those seeking more information were informed that this was a fed-erally sponsored research study in which (a) all data are reportedat the “group” level and no individuals would be identified, and (b)study participation would “not affect your driver’s license in anyway.” They received a description of the research project, includingthe IRB-approved consent form, and learned that compensation (inthe form of a $25 gift card for use at local convenience stores) wasoffered for their participation. Those who assented to participatein the research were guided to a nearby, private office, where theRA completed computer-based functional assessments, includingthe Maze Test, using a Windows® 2000 PC with a capacitance-based touchscreen display (Synaps Model S15TSM 15-in. LCD TFT,1024 × 768).
The maze navigation test, described as a “route planning task”to subjects, replicated the stimuli used by Ott et al. (2008). Subjectstraced a path, using their fingers or a stylus, through each of 5 mazespresented one after another on the touchscreen (see Fig. 1).
Subjects received the following instructions:
You will see five pages. Each contains a maze. Trace a path througheach maze from the left side to the right side as quickly as possible.
If you make a mistake, you can backtrack along the path you havetraced, until you reach the point where you wish to head in a newdirection.
When you complete each maze, a new one will appear. Your scoreon this test will be the time to complete all five mazes.
If a subject lifted his/her hand/finger from the screen whiledrawing a path through a maze, the line remained in place, and thesubject then continued forward on the same path (or backtrackedif s/he determined the path to be incorrect) when re-engaging
the maze at the point s/he left off. At an RA’s discretion, s/hecould prompt a subject to “Please continue drawing from where youstopped.”
L. Staplin et al. / Accident Analysis and Prevention 50 (2013) 483– 489 485
sed in ‘Route Planning’ task.
fefjfittods
3
slrtttpaossso
3
wiprpofpg
vtesd(Macsdba
Table 1Gender distribution of drivers in the study sample and comparison group.
Gender Study sample Comparison group
N % of sample N % of group
Male 365 52.7% 3292 40.9%Female 327 47.3% 4765 59.1%
Total 692 8057
Table 2Observed (expected) frequencies of drivers with and without retrospective at-faultcrashes, 2006–2008.
Sample Number of drivers Total
With at-faultcrashes
Without at-faultcrashes
Study sample 23 (20) 669 (672) 692Comparison group 229 (232) 7828 (7825) 8057
sample and 200 drivers among the comparison sample receivedconvictions carrying points from 2006 to 2008. A chi-square
Table 3Observed (expected) frequencies of drivers with and without retrospective pointviolations, 2006–2008.
Sample Number of drivers Total
With convictionscarrying points
Without convictionscarrying points
Fig. 1. Maze stimuli u
The PC recorded data that defined three potentially use-ul dependent variables: completion time, planning time, andrrors/dead ends. Completion time was the total amount of timerom the instant a maze appeared on the touchscreen until the sub-ect completed tracing a line through the maze and arrived at thenish point. The construct ‘planning time’ could be derived by sub-racting the time when the subject’s finger was actually drawinghe line through the maze from total completion time. The countf errors/dead ends revealed how often a subject was required toiscontinue the path he/she was following through the maze, andhift to another path.
. Analysis and results
Descriptive analyses characterized the demographics of the testample and the extent to which the sample represented the popu-ation of all older drivers in the State. Statistical tests examined theelationship between performance on the Maze Test and prospec-ive (18 months) crash and serious moving violation experience ofhe drivers assessed as described above. Qualifying crashes werehose reported by the Maryland State Police, or a by county or localolice agency; these typically, but not always, involve an injurynd/or require an involved vehicle to be towed from the scene. Seri-us moving violations were those that highlight the most seriousafety concerns; these included failure to stop at a red light or stopign, driving on a sidewalk, driving the wrong way on a one waytreet, failure to yield right of way, and driving on the wrong sidef the road.
.1. Sample demographics and driving history comparison
The composition of the study sample and of a comparison group,hich includes all other age-matched licensed drivers who vis-
ted the participating MVA offices during the term of this researchroject but did not agree to undergo functional assessment, areeported below. In addition, the driving histories of the study sam-le and comparison group were examined in terms of two safetyutcomes connoting culpability for at-risk driving behavior: at-ault crashes, and convictions for point violations in the prior 3-yeareriod (2006–2008). This comparison was undertaken to gauge theeneralizability of the study’s findings.
Apart from the study sample, 8057 drivers age 70 and olderisited the same MVA offices during the period of data collec-ion. These individuals form the comparison group; they werexposed to the recruitment flyers, but did not participate in thetudy. It is unknown if any drivers in the comparison sample hadirect interaction with the project’s RAs. Comparison group dataaggregated) were provided to the research team by the Maryland
otor Vehicle Administration, Office of Driver Safety Research.Thege range for the study sample was 70–93. The age range of theomparison group was 70–99. The mean driver age in the study
ample was 77.41 (s.d. = 5.29); for the comparison group, the meanriver age was 77.47 (s.d. = 5.80). Fig. 2 presents a more detailedreakdown of the study sample and comparison groups by 5-yearge cohort. As indicated, the driver age group 70–74 is slightly
Total 252 8497 8749
under-represented in the study sample and the age group 75–79 isslightly over-represented, with respect to the comparison group.
Table 1 describes the makeup of these groups by drivers’ gen-der. As indicated, a greater percentage of older males than olderfemales participated in the study, in contrast to the gender distri-bution among all older drivers visiting the study sites during thesame period. The driving history of the study sample versus thecomparison group was examined with respect to at-fault crashesand point violations, to test for a potential bias as evidenced by anunder- (or over-) representation in such events by those individualsconsenting to participate in this research.
These driving history data are reported in Tables 2 and 3, whichcontrasts the observed versus expected (in parentheses) number ofevents for the study and comparison samples, using the traditionalchi-square contingency table format.
As shown, 23 drivers among the study sample and 229 driversamong the comparison group were involved in at-fault crashesfrom 2006 through 2008 (Table 2). A chi-square calculation forthese observed versus expected counts yielded a test statistic valueof �2 = 0.503 (n.s.). Table 3 shows that 20 drivers among the study
Study sample 20 (17) 672 (675) 692Comparison group 200 (203) 7857 (7854) 8057
Total 220 8529 8749
486 L. Staplin et al. / Accident Analysis and Prevention 50 (2013) 483– 489
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
70-74 75-79 80-84 85+
Age
Perc
ent o
f Sam
ple
Study SampleComparison Sample
e grou
aa
3
Mneotewaweadabwst‘
Fig. 2. Percentage of drivers by ag
nalysis comparing these observed versus expected counts yielded test statistic value of �2 = 0.587 (n.s.).
.2. Maze Test scores and safety outcomes
Data summarizing performance by the study sample on theaze Test initially were tabulated for total completion time and
umber of errors/dead ends, for all five mazes. This data setxhibited an attrition of approximately 5%, relative to the numberriginally recruited. The reason for this attrition was unspecified:he informed consent agreement stipulated that subjects couldxcuse themselves from the assessment at any time, for any reason,ithout forfeiting compensation. Also, data were not tabulated or
nalyzed for the derived variable, ‘planning time,’ for which thereas a substantially greater amount of missing data due to an appar-
nt measurement error (this value was not greater than zero for subject on one or more mazes). This may have resulted fromifficulty in calibrating the sensitivity of the touchscreen to reli-bly discriminate between a ‘touch’ and a ‘near-touch’ (or ‘hover’)etween subjects. Further, the research assistants observed that
hen a subject periodically relinquished contact with the touch-
creen, his/her hand often remained close to the display, blockinghe view of part of the maze. Without a fully visible stimulus, theplanning’ construct becomes tenuous.
Fig. 3. Types of crashes experience
Group
p in study and comparison group.
Relationships between maze scores and prospective motor vehi-cle crashes and serious moving violations were examined using theR statistical computing environment (R Development Core Team,2011). The EpiTools package, which was loaded into R (Aragon,2010), supported significance testing. One-tailed tests were appliedbecause the hypotheses concerned a directional effect on safetyindices of a decline in functional abilities—i.e., poorer functionshould result in higher risk of unsafe outcomes.
Specifically, one-tailed significance tests using the mid-P (shortfor median or mid-probability) method, a variant of the Fisher’sexact test, were used. The mid-P variant compensates for overlyconservative significance testing of 2 × 2 contingency tables causedby discreteness of the data (for details see Berry and Armitage,1995). In the current analyses, the discreteness is due to the smallnumber of crashes (20) and citations (16) among the study sam-ple during the prospective observation period. As such, the Fisher’sexact test with mid-P adjustment was preferred for the currentanalyses over either the Pearson’s chi-square test or the Fisher’sexact test (Lydersen et al., 2009). An odds ratio (OR) was calculatedfor each statistical test that reached significance (p < .05). Cutpoints
were identified separately for crashes and for moving violations,signifying maze scores that resulted in the peak odds ratio with aminimum of 5 drivers per cell; this criterion for a valid odds ratiocalculation was rigidly applied.
d by drivers in study sample.
L. Staplin et al. / Accident Analysis and Prevention 50 (2013) 483– 489 487
F
rfopdaatmaitcatp
patdtMeirmt
nictt
aaamr
significantly predicted intersection crash involvement.Next, the analyses of maze scores in relation to moving vio-
ig. 4. Prospective crash experience of study sample by Maze 1 completion time.
These analyses, which examined the crash and violation expe-ience of drivers in the study sample for a period of 18 monthsollowing assessment, were made possible through the assistancef the Maryland Motor Vehicle Administration (MVA). The MVArovided files to our research team that keyed the prospectiveriving history data for each individual to the date of his/herssessment. All crashes that police reports indicated were drug orlcohol-related were excluded from analysis. After applying this fil-er, there were twenty (20) drivers in the study sample with one or
ore crashes during the observation period; these were classifieds shown in Fig. 3. Based on the distribution of crash types shownn this figure, it was possible to carry out additional analyses forwo subsets of crashes: intersection crashes and non-intersectionrashes. Because each of these subsets included ten crashes, thisllowed for a valid odds ratio calculation if a cutpoint could be iden-ified that resulted in 5 drivers at or above and 5 drivers below aarticular maze score.
The planned analyses of the relationship between maze com-letion time and crash involvement initially aggregated these datacross all mazes. This analysis did not demonstrate a reliable rela-ionship (p < .14). Next, because these five stimuli were of varyingifficulty, relationships involving individual mazes and combina-ions of mazes were analyzed separately. The analysis results for
aze 1 and Maze 2 follow. Completion time for Maze 1 (Fig. 4), theasiest stimulus (fewest turns to solve), exhibited a highly signif-cant relationship with crash involvement (p < .004). Drivers whoequired 19.1 s or longer to complete this maze were 3.55 timesore likely to be crash involved during the observation interval
han drivers scoring below this cutpoint.Analyses for Maze 2, the most difficult, also demonstrated a sig-
ificant (p < .027) relationship between completion time and crashnvolvement (see Fig. 5). Drivers who required 31.2 s or longer toomplete this more challenging maze were 2.54 times more likelyo be involved in one or more crashes than drivers scoring belowhis cutpoint.
But the result for the combined completion times on Maze 1nd Maze 2 was the most striking outcome of any of the presentnalyses. The relationship between this measure of performance
nd crash involvement was significant at p < .001. Fig. 6 exhibits aarked and sustained increase in the proportion of crash-involved
elative to crash-free drivers, as completion times became longer.
Fig. 5. Prospective crash experience of study sample by Maze 2 completion time.
Drivers who required 42.2 s or longer to complete both Maze 1 andMaze 2 were 4.58 times more likely to be involved in one or morecrashes during the 18 months following assessment, than driversscoring below this cutpoint.
It is interesting to note that a significant relationship with non-intersection crash involvement was found for Maze 1 (p < .01), Maze2 (p < .02), and Maze 1 + Maze 2 (p < .05) completion times. How-ever, the 10 crash-involved drivers in these analyses split 6 (fail)versus 4 (pass), which violates the requirement of at least 5 obser-vations per cell in the OR table established as the criterion for avalid odds ratio calculation, so no OR values are reported for theseresults. Neither individual maze scores nor combinations of scores
Fig. 6. Prospective crash experience of study sample by combined Maze 1 + Maze 2completion time.
488 L. Staplin et al. / Accident Analysis and
F
Dpvii1mp
antbcrt4a
sd
4
vmtlMltrtnar
oee2
ig. 7. Prospective violation experience of study sample by Maze 1 completion time.
UI), as well as violations related to occupant restraint use,arking, license and vehicle registration, and other non-movingiolations. It should be noted that citations were documentednstead of convictions as the dependent variable, to mitigate thenevitable loss of data through administrative actions. A total of6 drivers in the study sample were cited for one or more seriousoving violations during the 18-month prospective observation
eriod.The analysis including all five mazes, together, failed to
pproach significance (p < .104). But again, Maze 1 itself was a sig-ificant predictor, at p < .018. Drivers who required 12.7 s or longero complete this simplest maze were 3.05 times more likely toe cited for a serious moving violation, compared to drivers withompletion times below this cutpoint (see Fig. 7). The analysisesults for the combined Maze 1 + Maze 2 completion time failedo reached significance (p < .09); but, interestingly, the cutpoint of2 s for this OR calculation was the same as found in the crashnalysis.
Finally, none of the analyses including individual or aggregatecores for the measure of errors/dead ends during maze navigationemonstrated a reliable relationship with crash involvement.
. Conclusions and discussion
One conclusion can be drawn from analyses of the study sampleersus comparison group data. These analyses suggest that, whileen seemed disproportionately more likely than women to agree
o study participation, relative to their prevalence among all Mary-anders in their cohort who were concurrently doing business at the
VA, the recruited sample appears to reasonably represent this popu-ation of older drivers in the State with respect to age and with respecto key safety indicators. The present study sample is assumed to rep-esent the “normally aging” older driver population; in contrast tohe research of Ott et al. (2008), however, no neurological, diag-ostic procedures were performed in this study so the extent—ifny—to which clinically significant subpopulations may have beenepresented in these data is unknown.
The reason(s) for the observed difference in the relative rates
f study participation by men and women is unknown. The inter-sted reader may explore potential explanations from researchxamining gender differences in online transactions (Fallows,005), e-commerce adoption (cf. Lee, 2010), and willingness to
Prevention 50 (2013) 483– 489
participate in clinical trials (Ding et al., 2007). It should also benoted that the present data do not allow for any comparisons ofthe exposure of the study sample and comparison group in termsof miles driven or driving context.
Additional caveats to the present results reflect the sample sizeand the associated number of events that entered into each analysisof safety outcomes. A sample large enough—ideally an exhaustive,statewide data collection effort—to yield hundreds instead of tensof crashes would greatly refine our understanding of populationnorms for test performance, and lend greater precision as well asconfidence to the cutpoints used for classifying drivers into riskgroups.
The most important conclusions that can be drawn from thisstudy relate to the demonstrated relationships between Maze Testperformance and crash and serious point violation experience: itappears that the simplest and the most complex mazes in this stim-ulus set, both individually and in combination, can identify olderdrivers who, as a group, are at significantly higher risk of prospectivecrash involvement. The analysis including both of these test stimuliyielded an odds ratio of 4.58. These data are compelling—especiallygiven the dearth of alternatives. The Clock Test, for which thestrongest evidence has largely been derived from driving simula-tor studies, is fraught with methodological challenges; and no otherbrief test of ‘executive function’ offers even that mixed record. Thatthe Maze Test can be reliably self-administered using a standard-ized protocol in under 3 min also merits attention.
This does not suggest that the Maze Test is preeminent inthe domain of measures that screen for cognitive impairmentsthat predict older driver crash risk. Indeed, the present findingsargue that this measure may be superior in predicting crash riskat non-intersection locations. While this might characterize a largemajority of an individual’s driving exposure, there is a continuingand understandable emphasis on intersections as being particu-larly problematic for older drivers (cf. Stutts et al., 2009); bothinjury severity and the likelihood of fatalities increase for the anglecrashes that occur in these driving situations.
Instead, the principal utility of the Maze Test may be as one,additional crash predictor that complements other measures ofcognitive decline associated with safe driving ability. For example,there are well-documented tools—most notably the Useful Field ofView and the Trail-making Test—that appear to discriminate moreeffectively between (older) drivers who are and are not involvedin intersection crashes (Rizzo et al., 2001; Owsley et al., 1991).Integrating the Maze Test into a protocol with these establishedprocedures could result in a brief cognitive screen that offers uniqueadvantages where the goal is to prioritize older drivers for a morein-depth assessment of medical fitness to drive.
Acknowledgments
The authors first wish to acknowledge the office for Driver SafetyResearch, Maryland Motor Vehicle Administration, without whoseassistance neither the functional assessments nor the extractionof driver history data needed to complete the present analyseswould have been possible. Ms. Danielle Betkey, in the Driver SafetyDivision of the Maryland Motor Vehicle Administration, and Ms.Cynthia Burch, in the National Study Center at the University ofMaryland School of Medicine, supported this study by compilingcrash and violation data files; their efforts are also greatly appreci-ated.
This research was supported under U.S.DOT/NHTSA Contract
DTNH22-05-D-05043, Task Order 10, “Older Drivers: Relation-ship Between Assessment Tool Scores and Safety Outcomes,” andU.S.DOT/NHTSA Contract DTHN22-09-D-00135, Task Order 01,“Older Driver Assessment Scores, Citations, and Crashes.”
is and
R
A
A
B
C
C
D
F
F
F
I
L
L
L
M
N
L. Staplin et al. / Accident Analys
eferences
lzheimer’s Association, 2011. Alzheimer’s disease facts and figures.Alzheimer’s & Dementia 7 (2), Retrieved from http://www.alz.org/downloads/Facts Figures 2011.pdf.
ragon, T., 2010. EpiTools: R package for epidemiologic data and graphics. Retrievedfrom http://cran.r-project.org/web/packages/epitools/epitools.pdf.
erry, G., Armitage, P., 1995. Mid-P confidence intervals: a brief review. The Statis-tician 44 (4), 417–423.
arr, D.B., Barco, P.B., Wallendorf, M.J., Snellgrove, C.A., Ott, B.R., 2011. Predict-ing road test performance in drivers with dementia. Journal of the AmericanGeriatric Society 59, 2112–2117.
arr, D.B., Ott, B.R., 2010. The older adult driver with cognitive impairment. Journalof the American Medical Association 303 (16), 1632–1641.
ing, E.L., Powe, N.R., Manson, J.E., Sherber, N.S., Braunstein, J.B., 2007. Sex differ-ences in perceived risks, distrust, and willingness to participate in clinical trials:a randomized study of cardiovascular prevention trials. Archives of InternalMedicine 167, 905–912.
allows, D., 2005. How women and men use the internet. Pew Internet and Amer-ican Life Project. Washington, DC. Retrieved from http://www.pewinternet.org/Reports/2005/How-Women-and-Men-Use-the-Internet.aspx.
itten, L.J., Perryman, K.M., Wilkinson, C.J., Little, R.J., Burns, M.M., Pachana, N.,Mervis, J.R., Malmgren, R., Siembieda, D.W., Ganzell, S., 1995. Alzheimer and vas-cular dementias and driving: a prospective road and laboratory study. Journalof the American Medical Association 273 (17), 1360–1365.
reund, B.D., Gravenstein, S., Ferris, R., Burke, B.L., Shaheen, E., 2005. Draw-ing clocks and driving cars. Journal of General Internal Medicine 20 (3),240–244.
nsurance Institute for Highway Safety (IIHS), 2008. Fatality facts 2008:older people. Retrieved from http://www.iihs.org/research/fatality facts2008/olderpeople.html#sec2.
am, L.C., Chiu, H.F., Ng, K.O., Chan, C., Chan, W.F., Li, S.W., Wong, M., 1998. Clock-face drawing, reading and setting tests in the screening of dementia in Chineseelderly adults. Journal of Gerontology 53 (6), 353–357.
ee, J., 2010. The roles of demographics on the perceptions of electronic com-merce adoption. Academy of Marketing Studies Journal 14 (1), Retrieved fromhttp://www.thefreelibrary.com/ /print/PrintArticle.aspx?id=235631650.
ydersen, S., Fagerland, M.W., Laake, P., 2009. Recommended tests for association in2 × 2 tables. Statistics in Medicine 28, 1159–1175.
endez, M.F., Ala, T., Underwood, K.L., 1992. Development of scoring criteria for
the clock drawing task in Alzheimer’s disease. Journal of the American GeriatricSociety 40 (11), 1095–1099.
ishiwaki, Y., Breeze, E., Smeeth, L., Bulpitt, C.J., Peters, R., Fletcher, A.E., 2004. Valid-ity of the clock-drawing test as a screening tool for cognitive impairment in theelderly. American Journal of Epidemiology 160 (8), 797–807.
Prevention 50 (2013) 483– 489 489
Ott, B.R., Festa, E.K., Amick, M.M., Grace, J., Davis, J.D., Heindel, W.C., 2008. Com-puterized maze navigation and on-road performance by drivers with dementia.Journal of Geriatric Psychiatry and Neurology 21 (1), 18–25.
Ott, B.R., Heindel, W.C., Whelihan, W.M., Caron, M.D., Piatt, A.L., DiCarlo, M.A., 2003.Mazes test performance and reported driving ability in early dementia. Journalof Geriatric Psychiatry and Neurology 16, 151–155.
Oswanski, M.F., Sharma, O.P., Shekar, S., Vassar, L.A., Woods, K.L., Sargent, W.M.,Pitock, R.J., 2007. Evaluation of two assessment tools in predicting driving abilityof senior drivers. American Journal of Physical Medicine and Rehabilitation 86(2), 190–199.
Owsley, C., Ball, K., Sloane, M., Roenker, D., Bruni, J.R., 1991. Visual/cognitivecorrelates of vehicle accidents in older drivers. Psychology and Aging 6 (3),403–415.
R Development Core Team, 2011. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0, http://www.R-project.org/.
Reger, M., Welsh, R., Watson, G., Cholerton, B., Baker, L., Craft, S., 2004. Relationshipbetween neurological functioning and driving ability in dementia: a meta-analysis. Neuropsychology 18 (1), 85–93.
Rizzo, M., McGehee, D., Dawson, J.D., Anderson, S.N., 2001. Simulated car crashes atintersections in drivers with Alzheimer’s disease. Alzheimer’s Disease & RelatedDisorders 15 (1), 10–20.
Shua-Haim, J., Koppuzha, G., Gross, J., 1996. A simple scoring system for clockdrawing in patients with Alzheimer’s disease. Journal of the American GeriatricSociety 44 (3), 335.
Shulman, K.I., 2000. Clock-drawing: is it the ideal cognitive screening test? Interna-tional Journal of Geriatric Psychiatry 15, 548–561.
Snellgrove, C.A., 2005. Cognitive Screening for the Safe Driving Competence of OlderPeople with Mild Cognitive Impairment or Early Dementia. Australian TransportSafety Bureau.
Staplin, L., Gish, K., Wagner, E., 2003. MaryPODS revisited: updated crash analy-sis and implications for screening program implementation. Journal of SafetyResearch 34 (4), 389–397.
