Predictors of Automobile Crashes and Moving Violations among Elderly Drivers

Methods

Participants

We selected study participants from the Project Safety cohort, a probability sample of noninstitutionalized persons aged 72 years and older living in New Haven, Connecticut, in 1989. Project Safety studied the risk factors for falls and fall-related injuries in an older community-living population. The sampling technique for this cohort was similar to that used to establish the New Haven Established Populations for Epidemiologic Studies of the Elderly (EPESE) cohort. Both sampling techniques have been described in detail elsewhere [25-27]. A census was taken of all 2483 age-restricted elderly housing units not occupied by persons enrolled in the EPESE cohort. We identified non–age-restricted community houses and apartment buildings through utilities listings. Every 62d non–age-restricted unit was sampled, and the next 12 units were identified as a cluster included in the study. Eligibility criteria included the ability to speak English, Spanish, or Italian; to follow simple commands; and to walk across a room without human assistance. Of 1392 eligible persons, 1103 (79%) agreed to participate and were enrolled in the cohort.

All cohort members had a baseline interview and a 1-year follow-up interview. There were 1103 respondents to the baseline interview and 915 respondents to the first-year follow-up interview. Of the remaining baseline respondents, 111 (10%) refused to participate further, 59 (5%) died, and 18 (2%) were proxy respondents. Study participants included all respondents who reported driving (n = 283) in the period between the baseline and follow-up interviews.

Data Collection

All participants had a baseline interview and physical performance assessment in their homes by a trained research nurse. Independent variables in this study included items potentially related to driving from the following five domains: demographic, health, psychosocial, activity, and physical performance.

Demographic features included age, sex, number of years of education, race, marital status, and type of housing. Data in the health domain were obtained on self-rated health, alcohol use, several chronic conditions, dizziness, loss of consciousness, and urinary incontinence. We used a Rosenbaum card [28, 29] to measure corrected static near visual acuity and used the Whisper test [30] to assess hearing. Medication use was determined by asking patients about prescription and over-the-counter medications and by examining pill bottles. Assessed medications potentially related to driving ability included opioid analgesics, tricyclic and tetracyclic antidepressants, antipsychotic agents, benzodiazepines, insulin, and oral hypoglycemic agents. The psychosocial domain included cognitive impairment assessed by the Folstein Mini-Mental State Examination [31], depressive symptoms assessed by the Center for Epidemiologic Studies-Depression scale (CES-D) [32], and availability of help with chores.

The activity domain consisted of self-reported independence in basic and instrumental activities of daily living derived from Katz and colleagues [33], Branch and colleagues [34], and the Older Americans Resource Services Instrument [35]. We also ascertained the number of flights of stairs and blocks walked on an average day. We assessed higher-level physical activity using a modification of the Yale Physical Activity Survey [36], in which participation in several activities was converted into a scale based on the estimated kilocalorie expenditure for each activity and the frequency of participation.

A battery of physical performance items included balance (side-to-side stand, tandem stand, single-leg stand, and withstanding a sternal nudge) and was scored on a 4-point scale, with 1 point given for each item done without instability [37, 38]. We determined strength and range of motion using manual muscle testing of shoulder abduction, grasp, hip flexion, knee flexion, and knee extension and categorized them as good (full resistance and full range of motion) versus fair or poor (less than full resistance or range of motion) [39]. We assessed ability to stand on toes and heels and the number of foot abnormalities present—toenail irregularities, calluses, bunions, and toe deformities such as hammer toes. Timed performance measures included hand signature, three chair stands, usual-pace walk (10 feet each up and back, including a turn, at usual pace), rapid-pace walk (10 feet each up and back as fast as the participant felt safe and comfortable), and foot tap (10 taps alternating between two circles on a mat).

Outcomes

At the follow-up interview, participants were asked if and how often they had driven in the past year (daily, every other day, once or twice a week, or less often), and if they had been stopped by the police or been cited for a moving violation. They were also asked if they had been involved in an accident while driving and, if so, if they had been injured or hospitalized as a result. The term “accident” was chosen because it is familiar to older drivers and because alternatives such as “crash” may connote a more severe event so that persons might not report minor incidents.

Statistical Analysis

We initially compared potential predictors of adverse events from the five domains of independent variables in bivariate analysis using chi-square tests for categorical variables (the Pearson chi-square for dichotomous variables and the Mantel-Haenszel chi-square for ordinal variables) [40]. For infrequently occurring factors with expected cell counts of 5 or less, we used the Fisher exact test instead of the chi-square test. We also grouped continuous variables by quartiles or at the median unless accepted cut-off scores were available (for example, 16 for the Center for Epidemiologic Studies-Depression scale) and analyzed them using chi-square tests. If no gradient was apparent for quartiles, the quartiles were collapsed into two levels. We entered variables significantly associated (P < 0.05) with the occurrence of adverse events in bivariate analysis into relative-risk binomial regression models using Generalized Linear Interactive Modeling [41] and adjusted for driving frequency to account for exposure and type of housing (the original sampling variable).

Results

The mean age of the 283 participants was 77.8 years (range, 72 to 92 years); 57% of participants were male. Forty-eight percent of participants lived in community dwellings, 45% lived in private age-restricted housing complexes, and 6% lived in public age- and income-restricted housing complexes. Fifty-five percent of participants reported driving daily, 24% every other day, and 21% 2 times a week or less.

Thirty-eight of the 283 drivers (13%) reported an adverse event in the first year of follow-up. Thirty-one of these persons reported a crash, 4 of whom were also cited for a moving violation; 4 participants were only cited for moving violations, and 3 participants were only stopped by the police. Of the 31 persons reporting a crash, 6 sustained an injury, and 1 was hospitalized.

The factors associated with adverse events in bivariate analysis are shown in Table 1. The only factor in the health and demographic domains that was significantly associated with adverse events was the number of chronic conditions, although this was based on the five participants with four conditions. The only types of medication marginally associated with adverse events were antidepressant agents, although only five persons were receiving them, and the association was not statistically significant. The occurrence of adverse events did not substantially differ between persons with better than or those with worse than 20/40 near static visual acuity; most states require 20/40 acuity for an unrestricted license [13]. Alcohol consumption was not associated with adverse events.

Persons with borderline cognitive impairment (Mini-Mental State Examination score of 23 to 25) were more likely to have adverse events than were those with higher or lower scores (relative risk, 2.0; 95% CI, 1.1 to 3.7). To investigate this relation further, we examined the components of the Mini-Mental State Examination individually and by cognitive domain (orientation, memory, attention, language, and visuospatial ability). The item most closely associated with adverse events was impaired design copying (24% of persons who could not correctly copy the intersecting pentagons had events compared with 8% of those who could [relative risk, 3.0; CI, 1.6 to 5.6]).

In the activity domain, walking less than one block per day was associated with adverse events (relative risk, 1.9; CI, 1.1 to 3.5), whereas driving frequency was not significantly associated. Several physical examination and performance measures were associated with adverse events. Persons with three or more foot abnormalities were more likely to have adverse events (relative risk, 2.2; CI, 1.2 to 3.9). Impaired left-knee flexion was also associated with adverse events (relative risk, 2.9; CI, 1.2 to 6.7). The timed performance test most strongly associated with adverse events was rapid-pace walk (>7 seconds versus ≤ 7 seconds [relative risk, 2.0; CI, 1.0 to 3.8]).

We entered four of the factors that were significantly associated in bivariate analysis (design copying, number of blocks walked, number of foot abnormalities, and time of rapid-pace walk) into binomial relative risk models and adjusted for driving frequency and housing type; we did not enter chronic conditions and left-knee flexion because of the few participants in the group at high risk for these factors. The factors that remained significantly associated were impaired design copying, fewer blocks walked, and more foot abnormalities (Table 2).

To determine how well the final model predicted the outcome, we formed risk groups on the basis of how many of the three significantly associated risk factors were present. If no factors were present, only 6% of drivers had an adverse event; if 1 factor was present, 12% had events; if 2 factors were present, 26% had events; and if 3 factors were present, 47% had events (Table 3).

Discussion

In this representative sample of community-living elderly persons aged 72 years and older, 13% of drivers had had a motor vehicle crash, were cited for a moving violation, or were stopped by police in a 1-year period. Factors measured at baseline that were most predictive of adverse driving events were poor design copying, fewer blocks walked, and more foot abnormalities.

Our prospective design is unique among studies of driving and allowed us to determine predictors of adverse events in this population. Participants were drawn from a representative sample of elderly persons in an urban community and were not over-represented by a particular disease or impairment (for example, cognitive or neurologic). The age of the participants (72 years and older) was another advantage because older persons are at higher risk for adverse driving events [1, 2, 7] and because relatively little is known about the occurrence of these events in older age groups.

Comparing the frequency of adverse events with other studies is difficult because of differences in the populations studied and in the outcomes reported. Crancer and McMurray [20] reviewed the state records of all licensed drivers in Washington over a 69-month period from 1961 to 1967. They found that 12.4% of licensed (as opposed to active) drivers older than 65 years had crashes or moving violations each year. This Figure is similar to our 13%, although Crancer and McMurray used state records to determine outcome. McFarland and colleagues [42] also used state records for outcomes and reported that 9% of licensed drivers in Connecticut aged 73 years and older had a traffic accident in 1959-1960. Liddell [43] used both self-report and government records and found a 7% occurrence of crashes per year among drivers in Montreal who were 55 years and older.

The factors that we found predictive of the occurrence of adverse events are logical and clinically relevant. We found that participants with borderline cognitive impairment (Mini-Mental State Examination scores of 23 to 25) had more adverse events than did participants with more impaired or less impaired cognitive function. Older persons with borderline impairment may encounter difficulty driving because they continue to drive as often as they have in the past and in complex situations but cannot adequately adjust in emergency situations. More impaired persons may be forced to reduce their amount of driving, and less impaired persons may be better able to compensate in emergency situations. The absence of an association between severe cognitive impairment and adverse events may also be due to the exclusion of these persons from the original cohort or because persons with greater impairment might not report adverse events. Corroborating data from state records or proxy respondents were not available. Further studies are needed to explore the relation of cognitive impairment to adverse events.

The element of the Mini-Mental State Examination most strongly associated with adverse events was the inability to accurately copy intersecting pentagons. This finding is consistent with the recent report by Hunt and colleagues [44], in which another measure of visuoperceptual ability was associated with driving performance in a small group of healthy elderly persons with mild cognitive impairment.

Other driving studies seldom ascertain foot abnormalities, although the association with adverse events is logical because such abnormalities may affect ability to maneuver between accelerator and brake pedals. Fewer blocks walked, slower rapid-pace walk, and impaired left-knee flexion reflect limitations in physical function. Rapid-pace walk was not significantly associated with adverse events in multivariable analysis, and left-knee flexion was not entered in multivariable models because few participants had this impairment. The bivariate association of these two measures and the trends toward association with adverse events that were exhibited by left-knee extension, left-foot tap time, chair stands, and toe stands suggest that lower-extremity dysfunction may contribute to adverse events. Further studies evaluating more participants with these impairments will be needed to determine the importance of this relation.

Near visual acuity was not significantly associated with the occurrence of adverse events. Previous studies have shown a small but statistically significant association between poor static visual acuity and automobile crashes [9, 10]. However, a more recent study evaluating a range of measures of visual function found that visual attention was a much better predictor of crash occurrence than were other measures of visual function in 53 volunteers from an optometry clinic [14]. It may be helpful to assess more detailed measures of visual function than near static acuity in future studies.

Our study has several potential limitations. We did not know the number of miles each person drove; thus, it was not possible to adjust the occurrence of adverse events for mileage. We did have information on frequency of driving, which may provide a more accurate measure of “exposure” in an urban setting than the number of miles driven; persons may make more frequent short trips and in more congested areas than they would in a rural setting. However, driving frequency was not significantly associated with adverse events in bivariate analysis, although there was a trend toward more adverse events with greater driving frequency. We did not collect information on the nature of the accident other than the occurrence of an injury. There was no clear way to distinguish major from minor events and no way to assess fault. In addition, we could not corroborate the self-report of adverse events because of changes in the state recording of crashes and moving violations after October 1990. State records would have been particularly helpful in corroborating the self-report of adverse events among cognitively impaired persons. Self-reporting of adverse events may underestimate the occurrence if persons under-report events because they forget the event or because of concern about how the information will be used (for example, increased insurance rates or loss of license) [14]. However, self-report offers some potential advantages over state records because minor incidents may not be reported to the state or because state record-keeping may be inefficient [45]. Self-report also allows detection of persons who were stopped by police but were not cited for a moving violation.

We have shown that in this sample of elderly persons, several simple clinical measures correlated with the risk for adverse driving events. However, we would not necessarily expect the specific algorithm used here to do as well in a different population. These results are preliminary and need to be further developed and validated. When fully developed, such instruments will help identify a high-risk group that will need to undergo a more detailed assessment of driving ability and medical factors that may adversely affect driving ability, as well as potential interventions to remedy these problems. These instruments should be used to supplement existing recommendations about specific medical conditions and driving [46-48].

Our study is an important step in developing clinical criteria to assess the risk for adverse driving events among elderly drivers. We believe that clinicians, family members, and state regulators need guidance in identifying older drivers who are at particular risk for adverse driving events. The ultimate development of simple clinical instruments that can accurately predict driving problems and help target interventions to avoid these problems will help families, physicians, and society to manage difficult questions about continued driving by older persons.