<Back    TITLE PAGE    Forward>

OBJECTIVES

The purpose of the study was to compare the crash and citation rates of drivers with medical conditions to drivers without medical conditions. We sought to analyze each functional ability level for each functional ability category (medical conditions) for the study period 1992 – 1996. Some functional ability categories were excluded from analysis entirely, however, because they contained too few drivers for analysis by each functional ability level. Also, some specific functional ability levels for other functional ability categories were excluded because of very small numbers. Analyses for drivers licensed with multiple medical conditions were performed separately, by restriction status.


METHODOLOGY

Probabilistic Linkage

Data used in this analysis existed in several disparate databases available to Utah CODES. Probabilistic linkage was used to link data elements from these databases in order to combine the relevant data elements needed for such a study. Probabilistic linkage is an iterative tool which can overcome inaccuracies or differences in the separate databases (e.g., incorrect, missing or duplicate data, typographical errors, changes in surnames, etc.) which exact matching cannot; it is amply described elsewhere . Data linkages were performed using Automatch Software and are described below:

Crash to Utah Master Driver License File

Variables from the Utah Department of Transportation Crash Files were linked to variables from the Utah Master Driver License File for the years 1992-1996. Fields used to link these two files included the license state of the crash driver, name (last, first, middle initial), sex, date of birth, and driver license number. The medical condition database was provided in a relational file to the Utah Master Driver License File.

The crash file identified 397,849 Utah licensed drivers as having had a crash during the study period. The Utah Master Driver License File contained 1,750,918 drivers license records. Of the Utah licensed drivers in crashes, 384,311 (97%) drivers were successfully matched to the corresponding driver license records. A copy of the match file is located in Appendix C.

Utah Death Certificate Database to Utah Master Driver License File

Probabilistic linkage was used to identify persons who held valid driver licenses and died either during the study period, or in the five years previous to the study period. This linkage was performed because drivers licensed with medical conditions were thought to have a higher mortality rate than the general population of drivers and deaths would effect the eligible number of driving days. Variables used to link these two files included name (last, first, middle), city, state, residential zip code, sex, date of birth, and social security number.

The death certificate database was subset to include persons ages sixteen years and over (i.e., persons eligible for a driver license). Thus, the resulting data set contained 100,248 death certificates for the years 1986-96 that were linked to 1,750,918 drivers license records from the 1997 Utah Driver License Master File. Of these, 59,709 (59.6%) were successfully matched. [1]. This matching procedure allowed a date of death variable to be created for drivers who died during the study period. By creating this variable, drivers who held valid driver licenses when they died could be excluded at the date of death (since deceased persons cannot drive even though their license is still valid) and allowed the replacement of comparison drivers who died prior to the study start date. [2] This procedure was performed in order to minimize misclassification bias of the number of days a driver was eligible to drive in the study. A copy of the match file is located in Appendix C.

Comparison Driver Selection

Ideally, crash and citation rates should be related to exposure, expressed as events per mile driven and controlled for risk factors that affect the likelihood of the event occurring. For example, if two drivers have the same number of crashes per year but one drives only half as much as the other, the rates are the same per unit time but two-fold higher when comparing driving distances. Additionally, factors such as weather, road surface, traffic conditions and speed limit may affect crash risk. Likewise, local law enforcement patterns in areas where drivers frequently drive affect the risk of citation. Unfortunately, true exposure data is not available. There is no reliable information on the number of miles driven by individuals with medical conditions.

This concept of exposure is important when comparing the crash and citation rates of different populations, particularly in the older persons or persons who have medical conditions that may affect driving. During the study period, drivers who reported medical conditions in Utah were much different than the general population of drivers. Figure 1 illustrates the differences in ages between the medical condition drivers by restriction status and the rest of the driving population. Note that the medical conditions driving population is much older than the general driving population overall, and that restricted drivers tend to be older than unrestricted drivers licensed with medical conditions. In addition to age, other characteristics of drivers with medical conditions may affect their driving habits. It has been shown that chronically ill drivers who are not confident in their driving ability limit the amount they drive, or limit their driving to times or conditions when they feel comfortable to drive . For example, persons who do not see well at night may schedule trips during daylight hours. These factors meant that a direct comparison of medical conditions program drivers with the entire population of Utah drivers would probably not be valid.

Figure 1 -- Percentage of Drivers Reporting Medical Conditions by Restriction Status Compared to Drivers Not Reporting Medical Conditions By Age, 1992 – 1996

While ideally a study to evaluate the medical conditions program would consider exposure and these other factors, such data were not available. Because collecting these data would have been cost prohibitive, we determined that the best method to approximate these factors would be to match drivers with medical conditions to a representative comparison group. We reasoned that drivers of similar age and location are likely to drive similar amounts, so that age- and location- matching would be a reasonable proxy for true exposure rates.

A case-control methodology was therefore chosen for this study. For each driver with a medical condition, two drivers without medical conditions from the same age group, gender and county of residence were selected for comparison. For this study, only drivers without medical conditions were eligible to be chosen as a comparison driver. This category excludes all drivers with medical conditions and drivers with incomplete information in the master driver license file.

Drivers in the medical conditions program were subdivided by functional ability category (that is, medical conditions categories) (Table 1). The category “hearing” was excluded from analysis since this category was only used for commercial drivers. Drivers in each functional ability category were further subset by functional ability level (Table 2). If a driver with a medical condition fluctuated between functional ability levels over time, he or she was counted in each level for the appropriate time period. The same comparison drivers were used for each medical condition driver who fluctuated for functional ability level. Comparison drivers were followed for the duration if they held a valid driver license during the study period. Drivers listed in multiple functional ability categories were analyzed separately by restriction status. These groupings were further separated into categories by age group, county of residence and gender. Age groups included years 10-14, 15-19, 20-24, 25-29, 30-39, 40-49, 50-59, 60-69, 70-79, 80 and older. Driver's age was calculated at the midpoint of the study period using the date of birth available in the Master Driver License File. We included the age group 10-14 in order to capture new drivers entering the study near the endpoint (1995-1996).

Comparison drivers were selected randomly from all licensed drivers not in the medical conditions program from the 1997 master file. Commercial drivers who were licensed at functional ability level 1 (no history of disease/condition) were not included in the population from which a comparison driver could be selected. Similarly, drivers licensed at functional ability level 2 (past history of disease/condition but licenses are issued the same as the general driving population) were excluded. For each driver with a medical condition, two comparison drivers fitting the grouping criteria (age group, county of residence and gender) were chosen at random from the Utah Master Driver License File. Sampling for comparison drivers was performed with replacement, meaning that each possible comparison driver was eligible to be selected even if that driver was chosen previously to be a comparison. This method was used because there were too few drivers in some groupings to select two unique comparison drivers for each medical condition driver from the same age group, sex and county of residence.

The results of probabilistic linkage were used to determine the number of eligible licensed driving days by functional ability level; and the number of crashes, at-fault crashes and citations occurring at that functional ability level for each driver. Drivers with medical conditions who fluctuated between functional ability levels had the corresponding number of days at each level assigned. As mentioned previously, drivers who died during the study period had their corresponding number of eligible driving days adjusted so that the date of death was included but the following days excluded. Similarly, if a chosen driver had his or her driving privileges suspended because of citations or crashes, they were not excluded from the study; the number of eligible license days was adjusted to reflect the suspension. [3]

Comparison drivers were followed for the duration of the study (1992 – 1996) by their eligible number of driving days (the number of days they held a valid driver license) during the study period. The number of days used for these groups was higher than the number of days for drivers with medical conditions because the licensing periods are much shorter for drivers who have medical conditions (shorter licensing periods are built into the medical conditions driver program). For example, if a driver with a medical condition was in the database for 1 year of the study period, he or she would be counted for 365 days. However, his or her corresponding driver would have been followed from 1992 – 1996, or 1,825 days. This was done in order to simplify the matching process and minimize the computer time used to generate the comparison drivers. The eligible number of driving days for both drivers with medical conditions and their comparisons reflects the data of the Utah Driver License Division. The same two comparison drivers were used for drivers whose functional ability level fluctuated. Events and eligible license days for these comparison drivers were counted at each functional ability level status.

Crashes were considered to be “at fault” if a driver received a citation for the crash or was marked as having contributed to the crash. Only crashes and citations that occurred during the period of time the driver was licensed were considered. Events (citation or crash) were corresponded to the driver's record, and restriction status. Citation, crash and at fault crash rates per eligible licensed driving day were calculated separately for restricted and unrestricted drivers with medical conditions and their corresponding drivers for each functional ability category. These data were then used to estimate the relative risk for each medical condition category, allowing a comparison of the crash or citation risk of drivers licensed with medical conditions to similar drivers licensed without medical conditions from the general driving population. The relative risk approximates a Chi-Square distribution with one degree of freedom. Using this distribution, we calculated a 95% confidence interval for the estimate of relative risk . Relative risk describes the influence of a particular variable on the likelihood of an outcome. For instance, drivers in the visual acuity group at functional ability level 3 have a relative risk for crashes of 1.407; this means that they were 1.407 times as likely to be in a crash as were members of the control group.

The second part of the project was an analysis of drivers with more than one medical diagnosis or medical condition. These drivers were therefore put into more than one functional ability category, and were assigned a functional ability level for each. Only a minority of drivers was listed in more than one functional ability category, and only a few of the possible combinations of functional ability categories occurred often enough to allow for meaningful analysis. Only two-way combinations of functional ability categories (drivers reporting two medical conditions) occurred with sufficient frequency to allow for meaningful analysis, although there were small numbers of drivers with 3 or more medical conditions (as many as 7). Also, the numbers of drivers with two medical conditions was too small to allow for analysis by individual functional ability level. Therefore, the functional ability levels were collapsed into unrestricted (functional ability levels 3-5) and restricted (levels 6-11) groups.


Footnotes

[1] The following checks confirmed these results:

1. Check of Linkage Strategy and Other State's ExperiencesLinkage strategy was reviewed internally by Utah CODES staff and externally by Mike McGlincy of Matchware, Inc. A similar linkage using Los Angeles County drivers and voters, matched at around 60%.

2. Manual Check of SubsetDeath certificates contain a code for cause of death. One such code is driver in a crash (E-codes 8100, 8110, ... 8190). 628 individuals were so identified, and 593 were found to be successfully linked (94%). The remaining 35 individuals were looked up in the crash files. 17 of these drivers were from out of state and therefore, did not have a Utah license. 12 drivers did not have a license number in the crash file, although the state was identified as Utah. This would leave a linkage rate of 593/599 (99%). 4 drivers had license numbers that did not match to the DMV file (e.g., possible data entry errors). The remaining 2 individuals had a Utah drivers license and were found in the DMV data. For these two drivers, social security numbers did not match on 7 and 9 digits respectively.

[2]Overall, drivers with medical conditions did not have a higher mortality rate during the study period than those selected comparisons. Of the 68,769 drivers with medical conditions who renewed their licenses after 1/1/92, 3,810 (5.5%) matched to the death certificate file. Two comparison driver records were selected for each medical condition driver. Of those records, 10,372 (7.5%) of the selected comparison records linked to the death certificate file. However, it is important to note that comparison drivers did not have to renew after 1/1/92 to be included in the study. This was because their driver licenses are valid for 4 or 5 years depending upon the date of issue, as opposed to the shorter periods for drivers licensed with medical conditions. When limiting the linkage results to those comparison drivers who renewed their driver licenses after 1/1/92, the percentage of drivers linking to the death certificate file was 3.2% (3,975/122,863).

[3] The inclusion criteria of selected comparison drivers were chosen similarly to drivers with medical conditions in order to minimize bias. These drivers reflect a random sample of drivers from the general driving population with similar characteristics (age group, gender and county of residence) as those drivers with medical conditions. We did not select comparison drivers who were licensed for the whole study period as this would bias the sample towards those with “good driving records.” Similarly, we did not exclude drivers who died during the study period because drivers with medical conditions were not chosen this way. Both condition and comparison drivers who died were counted for the time of the study period they were alive and licensed as drivers.

<Back    TITLE PAGE    Forward>