Human factors contribute to traffic accidents,1 but few studies have looked at stressful life events that might predispose toward accidents. Holt et al.2 studied 55 patients admitted to orthopedic wards after accidents (24 at fault and 31 not at fault). The group at fault had experienced considerably more stressful life events than the other group in the previous year. Selzer3 interviewed 532 male drivers from Michigan and found that transitory life changes and subjective stress were correlated with traffic accidents, whereas personality and demographic variables were not. In both studies, however, life events were reported after the accident.
A 1970 study of the driving records of 410 persons involved in divorce proceedings found that in the 6 months before and after filing of the divorce petition, traffic accidents and violation rates were markedly higher than the average population rates.4 Comparison of the marital status of victims of motor vehicle fatalities and of persons who died of natural causes in the United States in 1992 showed that divorced persons were more likely to die in road traffic accidents than married individuals (OR = 1.2; 95% CI = 1.1–1.3).5 One possible explanation is that marital conflict produces stress that may be associated with inattention or with irresponsible behaviors.
We conducted a prospective investigation of the association between potentially stressful life events and the risk of serious traffic accidents over an 8-year period in a large cohort of French workers.
The participants, all employees of or recent retirees from the French national electricity and gas companies (Electricité de France and Gaz de France), volunteered in 1989 to participate in a research cohort. These 2 firms employ approximately 150,000 people of diverse trades and socioeconomic groups throughout France. This GAZEL cohort, studied by the National Institutes of Health and Medical Research (Institut National de la Santé et de la Recherche Médicale, INSERM), initially included 20,625 participants: 15,011 men age 40–50 years in 1989 and 5614 women age 35–50 years. A comprehensive database has been regularly updated since then with data from the companies’ human resources and occupational medicine departments and medical insurance program and from an annual questionnaire mailed to participants. This cohort study has been described in more detail elsewhere6,7; its principal objective is to collect data about the annual prevalence and incidence of chronic health problems.
Data From the GAZEL Cohort
Social and demographic data from the cohort database include sex, year of birth, and occupational category (the latter updated annually). Participants were defined as episodic alcohol consumers when they reported drinking alcohol a maximum of 2 days a week. They were also asked to indicate the maximum amount of alcohol drunk during 1 day; when this amount exceeded 3 glasses of wine, 3 pints of beer, or 2 measures of spirits, respondents were classified as high-quantity users. These 2 indicators were combined to define a composite time-dependent alcohol consumption variable with the 4 following categories: low-quantity regular user, high-quantity regular user, low-quantity episodic user, and high-quantity episodic user.
From January 1993 onward, the annual questionnaire included a set of questions about the experience of potentially stressful life events during the previous 12 months. Those life events included an important purchase, hospitalization, retirement, change in job position, and reorganization in the workplace. Life events related to participants’ partners were death, hospitalization, retirement, and change in job position. Life events related to the family circle were birth, death, and hospitalization. From January 1994 onward, this list also included the participant's separation or divorce, the departure of a child from the household, and separation or divorce in the family circle.
In 1978, the company's health insurance medical department began recording worker health problems in a database8 that contains demographic, socioeconomic, and occupational data, as well as information about causes of sick leaves.
Data From the Driving Behavior and Road Safety Questionnaire
A questionnaire on driving behaviors and traffic safety was pilot-tested in December 2000 on 500 randomly selected participants, 330 of whom responded. We used their answers and comments to finalize this questionnaire, which was sent to the 19,894 living members of the GAZEL cohort in March 2001. It was designed to elicit descriptions of serious traffic accidents over the past 12 years (since joining the cohort). Participants were asked to report all of their serious traffic accidents in the follow-up period (defined as a traffic accident in which someone was injured, ie, required medical care). The medical department provided participants with lists of the dates of any sick leaves for traffic accidents during the period covered by the questionnaire. Every reported serious accident was investigated extensively with a set of 25 questions about the conditions, injuries, vehicles, and reasons for the accident and the responsibility of each driver involved.
Mileage was estimated from the reported number of kilometers driven in a 2- or 4-wheel vehicle in the previous 12 months.
The study protocol was approved by the French National Board for Data Protection (Commission Informatique et Liberté [CNIL]). Before inclusion in both the GAZEL cohort and in the study on traffic safety, all participants received an information letter describing the main objective of the study and the noncompulsory nature of their participation.
Two people separately entered the data from the driving behavior and traffic safety questionnaire into a database. The 2 entries were compared, discrepancies were corrected, and inconsistencies checked for.
The main outcome studied was the first serious accident as a driver of a 2- or 4-wheel vehicle in the 8-year follow-up period for which life events were reported (1993 onward). Life events reported each year covered the previous 12-month period.
We compared the risk of a serious accident among participants who did and did not report specific life events during the 12 months preceding the accident fitting Cox's proportional hazard regression models with time-dependent covariates.9 The time variable was the year of the follow-up period (ranging from 1993–2000), and the event was the first serious accident in the period. A separate model was computed for each life event. The report of a given life event was included as a time-dependent variable; it could be different each year. All models were adjusted for annual mileage as a driver (as reported in the 2001 driving behavior and traffic safety questionnaire), occupational category each year (a time-dependent covariate in 3 categories: unskilled workers, skilled workers, and managers), age (a time-dependent covariate), and alcohol consumption (a 4-category time-dependent covariate describing quantity and frequency). After checking the linearity of covariates, we applied a logarithm transformation to annual mileage as recommended elsewhere.10 Adjusted proportional hazard ratios were used to estimate the relative risk of a serious accident associated with each of the listed life events. Unmarried participants were excluded from analyses involving events concerning a partner. Separate analyses were also performed to consider only at-fault accidents as events. The proportionality of the Cox model was checked graphically, and interactions between exposure and potential confounders were tested. All hazard ratios (HRs) were computed with 95% confidence intervals (CIs).
The questionnaire on driving behavior and traffic safety was returned by 14,226 participants; 26 participants were excluded because of data discrepancies with the general cohort database. Data on occupational category or mileage were missing for another 345 participants. Consequently, 13,915 questionnaires (10,542 from men and 3373 from women) could be analyzed. The response rate was higher among men (71%) than women (63%), and respondents were slightly younger than nonrespondents (mean age was 56.7 vs. 57.1 for men and 53.8 vs. 54.4 for women). We also observed an upward trend in response rates with occupational category; among men, the response rate was 58% among unskilled workers but reached 71% among skilled workers and 78% among managers; the comparable figures for women were 55%, 66%, and 75%.
For the 1993–2000 period, 4.9% (684) of participants reported one serious accident while driving a 2- or 4-wheel vehicle, 0.2% (25) two serious accidents, and 0.03% (4) three. We analyzed only the first accident for each participant, for a total of 713 accidents.
Table 1 shows the characteristics of the 13,915 participants included in the analysis. Sociodemographic characteristics of those who did not report a serious road traffic accident in this period did not differ greatly from those who did, although more men and younger people reported accidents. Those who reported an accident also reported higher annual mileage.
Table 2 presents the adjusted hazard ratios for serious accidents overall and for at-fault serious accidents according to each reported life event. Marital separation or divorce was strongly associated with a higher serious accident risk (hazard ratio [HR] = 2.9) and even more strongly with an accident for which the participant admitted fault (HR = 4.4). A child leaving home and an important purchase were also associated with a higher risk of at-fault serious accidents (HR = 1.5 and 1.6, respectively). Unit restructuring and hospitalization of respondent's partner were also associated with all serious accidents, but the association was weaker when considering at-fault accidents only. We performed the same analyses without the covariate indicating quantity and frequency of alcohol consumption; estimates of hazard ratios remained unchanged and are consequently not tabulated.
Recent divorce or separation was associated with a 4-fold risk of a serious traffic accident that was at least partly the fault of the driver, compared with drivers who did not report divorce or separation. These hazard ratios are similar to those associated with driving with a blood alcohol level at the legal limit.11 We believe that our result corresponds to a genuine impact of the stress specifically induced by divorce or separation, which were events that emerged from a list of various stressful life events we evaluated for a possible association with at-fault accidents.
The annual divorce or separation rate in our sample was 7 per 1000 (821 reported events over an 8-year period among 13,915 participants). This is consistent with the divorce rate of 2 per 1000 in France, a figure that does not include separations and includes in its denominator the entire French population, including children. Thus, roughly 1% of the entire driving population experienced divorce or separation every year. Applying a relative risk of 4 to this population, we can estimate that approximately 3% of all serious traffic accidents are attributable to the stress induced by marital breakups; this would account for roughly 170 deaths and 4000 injuries each year.
Changes in mileage driven the year of a separation or a divorce might have contributed to our results. That is, this personal event might lead to the participant moving possibly further from work (at least temporarily) or driving further to pick up or visit the children. Mileage was available only for 2001. Ideally, in the Cox's proportional hazard regression, we would like to have adjusted for this time-dependent variable, because mileage is susceptible to variations. Nonetheless, if the increased mileage induced the association between separation or divorce and the risk of traffic accidents, it would have done so for both types: those for which the participant was at fault and the others. Because the association increased substantially for at-fault accidents, we believe that it reflects a real impact of separation or divorce on the risk of accidents.
To confirm this comparison between participants, we also performed a case–time-control analysis (results not shown) with each person serving as his or her own control in a different time period.12,13 Participants who had no accidents were used to adjust for time trend in exposure, and participants who did have accidents were used to estimate the association between life events and the risk of accident. The sample size for this analysis, however, was very small. According to this method, separation or divorce was slightly associated with a risk of at-fault serious accidents among men, although with a very wide 95% confidence interval (HR = 5.3; CI = 0.74—infinity). No other life event showed any association with at-fault accidents in this analysis.
The completeness of serious accident reporting was enhanced by the medical department's provision of reminder lists of sick leaves for traffic accidents during the period covered by the questionnaire. Participants described the accidents extensively, answering a set of 25 questions about the conditions, injuries, vehicles, and fault; examination of the information reported found no inconsistencies. The proportion of participants who reported serious accidents increased slightly over time (from 2.6% in the 1993–1996 period to 3.3% in the 1997–2000 period); this finding suggests a moderate recall bias. The validity of self-reported fault for traffic accidents was also a matter of concern. Nonetheless, 331 of 713 (46%) participants reporting accidents admitted fault, a rate hardly indicative of strong concealment. Finally, it should be noted that fatal accidents were not included in the study. Causes of death for the cohort participants are available for the period 1993–1999. During that period, 8 participants died of traffic accidents (not necessarily accidents in which they were the driver).
Divorce is associated with increased alcohol consumption. Using time-series analysis of aggregate data, Caces et al.14 found a bidirectional association of divorce rates and alcohol consumption between 1934 and 1987 in the United States. A 6-year longitudinal study of men aged 51 to 61 years in the U.S. Health and Retirement Study found alcohol consumption increased dramatically among those getting divorced.15 A study of a random sample of 27,000 noninstitutionalized Dutch citizens age 15–74 years also found an increased risk of heavy drinking among those who reported separation or divorce.16 We therefore suspect that at least part of the association we found could be explained by increased alcohol consumption during the period surrounding divorce or separation. However, we found no association between the risk of accidents and overall alcohol consumption, although high episodic alcohol consumption was slightly associated with this risk. Hazard ratio estimates were the same regardless of whether we adjusted for high-quantity and episodic alcohol consumption.
Another possible explanation is that life changes such as divorce and separation may trigger depression.17 This disorder is associated with the use of psychoactive drugs, which in turn may increase accident risk. It is also plausible that the social isolation and sense of disorientation and stress that result from divorce may inhibit safe decision-making. This problem is illustrated in studies on the high risk of suicide among divorced individuals.18,19
A 1970 study in the United States4 suggested that rates of accidents and moving violations rise among drivers involved in divorce proceedings. Another study of U.S. mortality files for 1992 found divorced persons more likely to die of road traffic accidents than married, widowed, or single individuals.5 Our study attempted to assess the potential role of stressful life events on the risk of traffic accidents among a large cohort of French workers and found separation and divorce to be of particular importance. Although application of these results in any regulatory context is difficult to imagine, increased awareness of this issue, especially by those involved in separation or divorce, may have some preventive effect.
We thank all those who supported this work and in particular the staff of unit 88 of INSERM. Very special thanks go to Sébastien Bonenfant, Nadine Kaniewsky, Nathalie Lopes, and Marie Zins for their valuable help throughout the study and/or the preparation of the manuscript.
1. McFarland RA, Moore RC. Human factors in highway safety: a review and evaluation. N Engl J Med
2. Holt PL. Stressful life events preceding road traffic accidents. Injury
3. Selzer ML, Vinokur A. Life events, subjective stress, and traffic accidents. Am J Psychiatry
4. McMurray L. Emotional stress and driving performance: the effect of divorce. Behav Res Highway Safety
5. Kposowa A, Adams M. Motor vehicle crash fatalities: the effects of race and marital status. Appl Behav Sci Rev
6. Goldberg M, Leclerc A, Chastang JF, et al. Mise en place d'une cohorte épidémiologique à Electricité de France—Gaz de France. Recrutement des volontaires. Principales caractéristiques de l'échantillon. Revue d'Epidémiologie et de Santé Publique
7. Goldberg M, Leclerc A, Bugel I, et al. La cohorte GAZEL, laboratoire épidémiologique. Bilan des cinq première annés (1989–1993) de suivi des 20 000 volontaires d'Electricité de France—Gaz de France
[in French]. Paris: INSERM—Collection Grandes Enquêtes; 1994.
8. Goldberg M, Blanc M, Chastang JF, et al. The health data base of a nationwide company—its use in epidemiological studies. J Occup Med
9. Klein J, Moeschberger M. Survival Analysis: Techniques for Censored and Truncated Data.
Statitics for Biology and Health; 1997.
10. Janke MK. Accidents, mileage, and the exaggeration of risk. Accid Anal Prev
11. Perneger T, Smith GS. The driver's role in fatal two-car crashes: a paired ’case–control’ study. Am J Epidemiol
12. Maclure M. The case-crossover design: a method for studying transient effects on the risk of acute events. Am J Epidemiol
13. Suissa S. The case-time-control design. Epidemiology
14. Caces MF, Harford TC, Williams GD, et al. Alcohol consumption and divorce rates in the United States. J Stud Alcohol
15. Perreira KM, Sloan FA. Life events and alcohol consumption among mature adults: a longitudinal analysis. J Stud Alcohol
16. Jose BS, van Oers HA, van de Mheen HD, et al. Stressors and alcohol consumption. Alcohol Alcohol
17. Paykel ES, Myers JK, Dienelt MN, et al. Life events and depression. A controlled study. Arch Gen Psychiatry
18. Stack S. The effect of divorce on suicide in Denmark, 1951–1980. The Sociological Quarterly
19. Kposowa AJ. Marital status and suicide in the National Longitudinal Mortality Study. J Epidemiol Community Health