METHODS

Data sources

Data for this study were from two federal sources: the 1993 National Mortality Followback Survey (NMFS)¹¹ and the Fatality Analysis Reporting System (FARS).¹² The 1993 NMFS included a total of 22 957 decedents sampled from 2 215 000 adults aged 15 or older who died in 1993 in the US, excluding South Dakota, based on a stratified random sampling strategy. Information on decedents was obtained by telephone or personal interview from the informants named on the death certificates. One informant was interviewed for each sampled death. The informants usually were spouses, parents, children, or siblings of the deceased. The overall response rate was 83%. NMFS questions included use of alcohol.

FARS collects data on all traffic crashes within the US that involve a motor vehicle traveling on a public road and result in a death within 30 days of the crash. Data on BACs and previous convictions for driving under the influence or while intoxicated were obtained from FARS for the sample of driver fatalities included in the 1993 NMFS.

The National Highway Traffic Safety Administration provided a file of the 1456 fatally injured drivers included in NMFS, matched by death certificate number with their records in FARS. Of the 1456 cases coded as drivers in NMFS, 1121 also were coded as drivers in FARS. The remaining 335, primarily coded in FARS as pedestrians, were excluded from analysis. BACs were available for 818 of the 1121 drivers (73%), and relatives answered questions related to drinking for 84%–93% of the 818, depending on the individual question.

Imputed BACs¹³ for deceased drivers with missing data were not used because using imputed values would add uncertainty to any observed associations between drinking histories and BACs. The 818 drivers with FARS recorded BACs were similar in distribution by age and gender to all 23 043 fatally injured drivers in 1993. In addition, the BAC distributions of the 818 drivers resembled those of the 16 503 fatally injured drivers in 1993 for whom BAC data were available: 31% of NMFS drivers in the present study and 30% of driver fatalities in FARS had BACs of 0.15 g/dl or greater.

Analyses

Reported drinking and other behavioral characteristics relevant to traffic safety were analyzed in relation to BACs recorded in FARS. Drinking behavior was measured by a number of variables including reported quantity and frequency of drinking and various indicators of problem drinking. Time frames for the questions ranged from the decedents’ last month to last year of life.

Using drinking level categories modified from those of Cahalan et al, decedents were classified as abstainers or light, moderate, heavy, or very heavy drinkers, as illustrated in table 1.¹⁴ In this paper we use both of the following terms, which have different but related meanings: problem drinker and hard core drinking driver. The former refers to drinking histories suggestive of alcohol dependence and life problems related to drinking, and the latter is used by some organizations to refer to people who chronically drink and drive and who are resistant to changing their behavior. Although there is a standard operational definition for a hard core drinking driver (repeated alcohol related convictions or a BAC of 0.15+ g/dl), there is no standard definition of a problem drinker. We chose four potential indicators of problem drinking, namely whether the relatives described the decedent as:

A problem drinker during the last month of life (the NMFS interviewer explained: “By alcoholic or problem drinker, I mean a person who has physical or emotional problems because of drinking, problems with a spouse, family or friends because of drinking, problems at work or school because of drinking, problems with money because of drinking, or problems with the police because of drinking, such as drunk driving”).

Driving within two hours after having five or more drinks at least once a month during the last year of life.

A heavy or very heavy drinker during the last year of life (table 1 categories).

Having five or more drinks at one time at least once a month in the year preceding death.

The first two indicators were chosen because they appeared to be direct measures of problem drinking; however, the second measure is not definitive because alcohol impairment is dependent upon the period during which drinks are consumed. The latter two indicators referred to heavy drinking, which is not synonymous with problem drinking; however, most problem drinkers are heavy drinkers.^15–17

Prior convictions for driving while under the influence or while intoxicated were not used as an indicator of problem drinking because most alcohol impaired driving does not result in an arrest and many arrests do not lead to convictions.^18,19 In addition, FARS data are limited to convictions occurring within the three years before the crash so some offenders escape identification as repeat offenders.²⁰

The drinking histories reported by different types of informants (spouses, children, siblings, parents) were compared to see if some informants gave more credible information concerning driving after drinking relative to the known BACs of drivers killed in crashes. Separate estimates of problem drinking prevalence were calculated from data obtained from the most reliable informants.

The primary statistical analyses were comparisons of proportions of drivers in various BAC categories (zero, 0.01–0.09, 0.10–0.14, and 0.15+ g/dl) with respect to problem drinking indicators and other variables of interest. Dying with a BAC of 0.15+ g/dl was the primary outcome of interest because such drivers meet the standard operational definition of a hard core drinking driver.

As a supplement to these analyses, risk ratios (with 95% confidence intervals) for drivers having BACs of 0.15+ g/dl in comparison with drivers having no detectable blood alcohol were calculated for problem drinking indicators. Risk ratios for other BAC categories were not calculated because the numbers in other BAC positive categories were too small for robust statistical interpretations.

Two statistical software programs (SAS and SPSS) were used to compute risk ratios and χ² tests comparing proportions of variables by BAC category. See appendix at end of paper for formulas for risk ratios and their 95% confidence intervals.

RESULTS

Demographic characteristics

Compared with drivers who had not been drinking, fatally injured drivers with very high BACs (0.15+ g/dl) were more likely to be male (86% v 67%, aged 30–49 (50% v 30%), and divorced or separated (27% v 14%). Differences by urban or rural residence were not remarkable. Drivers with high BACs (0.10–0.14 g/dl) were similar demographically to those with very high BACs except that they were more likely to be younger than age 20.

BACs and drinking patterns

Of the 818 driver fatalities, 254 (31%) had very high BACs of 0.15+ g/dl, 57 (7%) had high BACs of 0.10–0.14 g/dl, 58 (7%) had low to moderate BACs ranging from 0.01–0.09 g/dl, and 449 (55%) had BACs of zero. Forty six (19%) of the drivers with very high BACs had been convicted of driving under the influence or while intoxicated during the three years before their fatal crashes (table 2). Almost one third of the fatally injured drivers—70% of drinking drivers—met the usual operational definition of being a hard core drinking driver (254 with BACs of 0.15+ g/dl and an additional four with BACs of 0.10–0.14 g/dl who had alcohol related driving convictions in the previous three years).

Fatally injured drivers who had BACs of 0.10 g/dl or greater were often characterized by signs of potential problem drinking (tables 2 and 3). Depending on the indicator variable for problem drinking, the estimated percentages of fatally injured drivers with very high BACs (0.15+ g/dl) who were problem drinkers ranged from 21% to 61% (table 2). In contrast, estimates of problem drinking prevalence among driver fatalities with no detectable blood alcohol ranged from 1% to 7%. At least one of the indicators of problem drinking, usually heavy drinking, was reported for 68% of drivers with BACs of 0.15+ g/dl, 41% with BACs of 0.10–0.14 g/dl, 32% with BACs of 0.01–0.09 g/dl, and 7% with BACs of zero. Specific results for individual indicator variables were as follows, all of which had significantly raised risk ratios for having BACs of 0.15+ g/dl (tables 2 and 3):

31% of drivers with very high BACs compared with 1% with zero BACs were described as problem drinkers during their last month of life (risk ratio 3.3).

21% of drivers with very high BACs compared with 2% with zero BACs were described as driving within two hours after having five or more drinks at least once a month during their last year of life (risk ratio 6.8).

61%of drivers with very high BACs compared with 7% with zero BACs were classified as heavy or very heavy drinkers during their last year (risk ratio 8.1).

54% of drivers with very high BACs compared with 7% with zero BACs were reported as having five or more drinks at a time at least once a month during their last year (risk ratio 4.0).

Comparisons among drivers with very high BACs of 0.15+ g/dl and other drinking drivers were hampered by the small numbers with lower BACs. Fatally injured drivers with high BACs of 0.10–0.14 g/dl were similar to drivers with very high BACs with regard to the percentages having five or more drinks on days when they drank and having five or more drinks at a time at least once a week (table 2).

A total of 59 of the 818 drivers (7%) had been convicted of alcohol related driving offenses during the three years before their fatal crashes; eight of the 59 had no detectable blood alcohol at the time of their fatal crashes (table 2). Such a conviction was significantly associated with having a BAC of 0.15+ g/dl (risk ratio 2.7) (table 3). The occurrence of problem drinking indicators was similar among drivers with very high BACs of 0.15+ g/dl whether or not they had histories of alcohol impaired driving convictions, although the small numbers of drivers with convictions made such comparisons less than conclusive.

A variable with implications for alcohol enforcement practices is the place where the fatal trip originated (table 2). Twenty six per cent of drivers with very high BACs of 0.15+ g/dl and 10% of those with high BACs of 0.10–0.14 g/dl had left a bar.

Comparing drinking histories by relationship of informants to drivers killed in crashes

Spouses appeared to provide more credible responses than other relatives to questions about the drinking habits of drivers dying in crashes because they were more likely to report that decedents with very high BACs (0.15+ g/dl) drove after drinking. For drivers with very high BACs, 70% of spouses reported that decedents had driven within two hours after drinking at least once a month during the last year of life compared with 65% of adult children, 57% of siblings, and 44% of parents (table 4). Similarly, spouses were far less likely to report that decedents with very high BACs did not drink and drive at all during the last year of life: 12% of spouses compared with 35% of parents.

When analyses were restricted to decedents for whom spouses were the respondents, problem drinking indicators increased among drivers with very high BACs, with percentages almost doubling compared with reports from other respondents. For the most direct signs of problem drinking (described as a problem drinker during the last month of life or frequently driving after having five or more drinks), spousal reports suggested that the prevalence of problem drinking among drivers with very high BACs could be 22% (based upon having both indicators), 32% (frequently driving after having five or more drinks), 44% (described as problem drinker), or 57% (having either indicator).

DISCUSSION

This study provides detailed information on drinking histories of fatally injured drivers. The results shed light on the question regarding the extent to which drivers who meet the operational definition of hard core drinking drivers (BAC of 0.15+ g/dl or repeated offenses for alcohol impaired driving) have characteristics indicative of problem drinking. Drivers with very high BACs (0.15+ g/dl) were far more likely than other fatally injured drivers to be described as having histories suggestive of problem drinking, but a substantial proportion were not so described. The findings among drivers with high BACs of 0.10–0.14 g/dl and low to moderate BACs of 0.01–0.09 g/dl are consistent with an increase in drinking problems as BACs increase among drivers. A small percentage of drivers without detectable blood alcohol had markers for problem drinking.

Spouses appeared to be the most credible respondents because they were least likely to state that fatally injured drivers with very high BACs had not ever driven a motor vehicle within two hours of drinking during the last year of life. Any informants for deceased drivers with very high BACs who said their relative never drove a motor vehicle within two hours of having a drink during the last year of life clearly were incorrect. Based on spousal reports of the two strongest markers for problem drinking (being described as a problem drinker in the last month of life or frequently driving after drinking five or more drinks), reasonable estimates of problem drinking among drivers with very high BACs range from 22% to 57%, depending on whether both or just one of these indicators were present. The drivers with both indicators would be most likely to engage in alcohol impaired driving on a regular basis and have alcohol related problems in other realms of their lives, although the two markers probably failed to identify some problem drinkers.¹⁵ Conversely, persons who consume five or more drinks before driving are not necessarily impaired or problem drinkers because the length of time during which the alcohol is imbibed is a critical factor.

Small numbers of drivers with BACs of 0.10–0.14 g/dl made findings among them less than definitive. None the less, their similarity to drivers with very high BACs with regard to most drinking and behavioral variables has important implications. BACs of 0.10–0.14 g/dl are more than three times as common as higher BACs among drivers on the road on weekend nights,²¹ and probably contribute to a substantial number of crashes resulting in non-fatal injuries.

One limitation of the NMFS data is that the information was based on proxy reports from the next of kin. It is possible that some proxies were ignorant of habitual problem drinking by the decedents or unwilling to report the drinking problems of a close family member. Also, spouses constituted only 30% of all respondents; decedents with spouses may have had drinking characteristics different from those for whom other relatives acted as informants. The research literature comparing self reports with proxy reports of drinking behavior is contradictory, with some studies finding spouses reporting more problem drinking, other studies finding spouses reporting lower quantities of alcohol consumption than the drinkers, and still others reporting high concordance for drinking descriptions between drinkers and next of kin.^22–26 Based on affirmative answers to driving after drinking for drivers with very high BACs, the majority of NMFS informants seemed to be credible sources.

Another limitation of using NMFS data is that the findings among fatally injured drivers likely differ considerably from those in the general population of drivers; however, there is no reason to think that fatally injured drivers with high BACs would have a lower prevalence of problem drinking indicators than drivers overall. Our study might provide a maximum bound for drinking problems among drivers on the road with high BACs, assuming proxy reports were accurate. Although the data were from 1993, substantial temporal changes since then in the relationship between problem drinking and dying in a crash seem unlikely because the proportion of fatally injured passenger vehicle drivers in the United States with BACs of 0.10 g/dl or higher changed little between 1993 and 2000.²⁷

Drinking history data and BACs were available for only 60%–70% of the fatally injured drivers in the NMFS sample, which was acceptable although less than optimal with respect to avoiding potential biases from missing data. A minor limitation of this study was that linking the data from NMFS to individual drivers in FARS precluded the use of sampling weights. Despite the lack of weights, NMFS drivers whose records were linked to FARS were similar demographically to all drivers killed in 1993.

Results of this research suggest a need for prevention efforts broader than those targeting repeat offenders and problem drinkers. Many fatally injured drivers with illegally high BACs of 0.15+ g/dl or 0.10–0.14 g/dl were not described as problem drinkers or as typically driving after having five or more drinks. Moreover, the distinction often made by researchers and public health practitioners between drinking drivers with BACs of 0.10–0.14 g/dl and 0.15+ g/dl appears to be arbitrary for the purpose of identifying persons who chronically drive while impaired by alcohol.

Tough sanctions to deter repeat offenders, such as vehicle impoundment or ignition interlocks, are an important component of programs to prevent alcohol impaired driving,^28,29 but other measures directed at the general population of drinking drivers also are needed. Some have argued against measures such as reducing the legal BAC threshold to 0.08 g/dl based on the premise that hard core drinking drivers are the major contributors to alcohol related crashes and deserve primary attention.^9,30 The evidence is growing, however, that 0.08 g/dl thresholds dissuade people from driving after excessive drinking across the entire range of BACs, including high BACs.^31–34 Other community-wide measures to deter alcohol impaired driving, including sobriety checkpoints and administrative license suspension, are effective but are underused.^28,35–40 The substantial number of drivers with illegally high BACs whose fatal trips originated from bars suggests that some sobriety checkpoints should be located in areas with bars and nightclubs.

FUNDING

This work was supported by the Insurance Institute for Highway Safety. Partial funding was provided by the Centers for Disease Promotion and Prevention, grant R49CCR302484. The opinions, findings, and conclusions expressed in the publication are those of the authors and do not necessarily reflect the views of the Insurance Institute for Highway Safety.

Appendix