The risk of road traffic crashes for occupational drivers: A responsibility study with comparison to the general population

2.1 Study data

The study data were taken from the VOIESUR (Vehicle-Occupant-Infrastructure Road-User Safety Studies project ANR11-VPTT-0007), which set up an information system based on analysis and meticulous coding of police reports, computerized and centralized by the TransPV agency, which provides insurance companies with police road accident reports [11–13]. When necessary, the data collection services provided important missing elements such as accident diagrams, vehicle photographs or injury assessments. This collection of information was based on injury or fatal accidents collected by the police for the year 2011 in metropolitan France (excluding overseas territories). The database included all fatal accidents recorded in France (accidents with at least one person killed on the spot or within 30 days), all non-fatal injury accidents recorded in the Rhone department (administrative area of 1.6 million inhabitants), and 1/20 (drawn at random) of the non-fatal injury accidents recorded in France (excluding the Rhone department). Consequently, the sampling weights applied were: 1 for all fatal accidents, 1 for non-fatal injury accidents in the Rhone department, 20 for non-fatal injury accidents outside the Rhone department. The database thus comprised 8,541 accidents, described in terms of more than 300 variables.

2.2 Survey plan

A retrospective responsible/ non-responsible retrospective case-control study included all drivers involved in an injury or fatal accident, aged 18-65. Responsible drivers were considered cases, and non-responsible drivers were considered controls. To select a population representative of the working population, drivers who were retired or unemployed, or driving a vehicle on a trial basis or participating in a competition, going to or returning from a party, dance, concert, festival, or discotheque, were excluded.

2.3 Study groups and risk factors

Four populations were distinguished:

drivers on personal journeys (going on or coming back from vacation, shopping, personal matters, leisure, touring, visits to family, friends or personal acquaintances, and journeys to or from school or university);

Study risk factors comprised:

driver’s age, in 7 categories: 18-20, 21-25, 26-30, 31-40, 41-50, 51-60 and 61-65 years;

Vehicles were classified in 10 categories:

cycles: bicycles, electric bicycles, any other pedal vehicle;

2.4 Determination of responsibility and constitution of case and control groups

The underlying principle of responsibility studies as used here is to compare a group of drivers considered to have been responsible for an accident [14], due to directly causal action or inaction, versus a group of drivers involved in an accident for which they did not bear responsibility. The hypothesis is that the latter show characteristics similar to those of drivers not involved in any accident [12]. Road-user responsibility is not being used here in a legal sense. A person causing or contributing to an accident is deemed responsible due to an inappropriate maneuver, such as driving against the traffic, failure to respect a red light, obvious loss of control, etc.) or failure to act (braking too late, etc.). It is essential that responsibility be defined in terms of these actual behaviors, and not of their causes (e.g., fatigue, consumption of drugs or alcohol, etc.), otherwise the impact of such risk factors would be widely overestimated. Responsibility was determined by an expert panel based on all available evidence, including accident diagrams and comments by those involved and by the police. This provided a responsibility criterion that was both reliable (in the sense of “contributive”) and as objective as possible (i.e., based on facts).

The two comparison groups were based on 5 categories: 1- completely responsible; 2 –largely responsible; 3- partially responsible; 4- largely non-responsible; and 5- not at all responsible. The “Responsible” group comprised categories 1, 2 and 3. “Largely” and “partially” responsible drivers were included here because accidents frequently occur due to a combination of factors, the absence of any one of which would often have avoided the accident; in other words, the accident would not have happened if the driver had not done something that led the expert to hold them completely or partially responsible. The Responsible group thus comprised drivers who made a mistake considered necessary (even if not sufficient) for the accident to have occurred. On this approach, several drivers may all be deemed responsible in a single accident. The “Non-responsible” group comprised categories 4 and 5: drivers considered to be involved by bad luck, being in the wrong place at the wrong time. Based on this concept of “responsibility”, the study can be seen epidemiologically as a case-control study. The source population comprised all drivers using public roads or private roads open to the public, and both groups came from this source population, as they were involved in accidents meeting this inclusion criterion.

2.5 Data imputation

Simple imputation of missing data was applied, using the MICE (Multivariate Imputations by Chained Equations) method [15]. This was mainly used to impute the type of journey, when unknown, using all relevant variables. Thus, variables imputed and used in the imputation model comprised: type of journey, driver’s occupational status, gender and socio-occupational category, being the owner of the vehicle or not, vehicle category, day of the week and time of accident, intended travel distance, distance actually traveled, frequency of driving at the accident site, accident occurring in an administrative area (Département) other than the driver’s home area, and vehicle categorized as “Special” (taxi, ambulance, fire-engine, police car, school bus, or dangerous goods transport vehicle).

2.6 Statistical analysis

Statistical analyses used R software, version 3.2.4.

Sampling weighting was applied to each driver for all analyses: 1 for drivers involved in a fatal accident, and 20 for those involved in an injury accident.

We used the svyglm function of the R survey library to take into account the weighting of the data (1 for fatal accidents, 1 for non-fatal injury accidents in the Rhone department, 20 for non-fatal injury accidents outside the Rhone département) and thus obtain valid variances. The goodness of fit was tested using Cox & Snell’s pseudo-R squared (psrsq function).

Statistical tests were 2-tailed, with the significance threshold set at p < 0.05 and 95% confidence intervals were established. Logistic regression modeled responsibility according to type of journey; because our goal was to see whether certain risk factors for being responsible for an injury accident were statistically different by type of journey, we tested the (first-order) interactions between these factors and type of journey. The significance of each factor or interaction was tested by comparing the likelihoods of the nested models (with and without each factor or interaction).

3.1 Type of journey

After application of exclusion criteria detailed in the survey plan paragraph, a total of 7,051 police reports were analyzed. After weighting, 69,352 drivers were involved in a fatal or injury accident in France in 2011. 1,631 (2.4%) had died within 30 days, 33,782 (49.2%) were injured, 33,211 (48.4%) were uninjured, and data were missing in 0.6% of cases.

Type of journey was known in 75.2% of cases, and imputed in the other 24.8%. After imputation, a majority of drivers (50.2%) were on personal journeys, 15.3% and 13.3% commuting to or from work respectively, and 21.1% traveling on duty.

3.2 Road accident victims (Table 1)

Drivers involved in an accident while traveling home from work, to work from home or for personal purposes were relatively similar in terms of age, gender and socio-occupational category. Notable differences concerned a lower rate of artisans, shopkeepers and business owners commuting to work and of 18-20 year-old injured drivers traveling home from work, and a slightly higher rate of males on personal journeys.

Drivers on duty differed in some respects from drivers on other types of journey, with higher rates of professional drivers and of males (p < 0.001) (Figure 1), and a lower rate of 18-25 year-olds than for drivers on personal journeys (p < 0.05).

OPEN IN VIEWER

Fig. 1

Occupational categories according to the type of journey, drivers involved in a fatal or injury accident in France in 2011.

The vehicle involved in the accident was most often a private or company car, whatever the type of journey. Logically, on the other hand, occupational vehicles (utility, heavy goods, special or agricultural vehicles, buses/coaches and tramcars) figured more frequently in on-duty accidents than in other types of journey (Figure 2).

OPEN IN VIEWER

Fig. 2

Vehicle driven according to the type of journey, drivers involved in a fatal or injury accident in France in 2011.

Drivers on duty were those least often testing positive for alcohol, followed by those commuting to or from work. Drivers on personal journeys were more than 7 times more likely to test positive for alcohol (at≥0.5 g/l) than on-duty drivers (p < 0.001). On the other hand, blood alcohol level in alcohol-positive injured drivers did not significantly differ according to type of journey.

Responsibility was attributed in 97.3% of police reports, and about 80% of drivers were considered completely responsible (42.3%) or completely non-responsible (37.5%) by the experts. Responsibility was considered partial in only 5.8% of drivers.

Analysis of road accident responsibility risk according to type of journey revealed lower risk in on-duty drivers and drivers commuting home than in drivers on personal journeys (p < 0.001), the lowest risk being for journeys home from work (Table 2). After adjustment for gender and age, the difference remained significant (models 1 and 2). After adjustment for blood alcohol level, on the other hand, the protective effect of commuting home was lower and that of being on duty disappeared (model 3). After further adjustment on the frequency of driving at the accident site, the odds ratio between commuting home and personal journeys no longer differed significantly from 1 (model 4). However, there remained extra risk for commuting to versus from work (OR = 1.38 [1.06; 1.78], p = 0.015). Adjustment on the time of the accident was also tested; multivariate analysis including all the variables in model 3 plus time of accident did not significantly affect the results for commuting to or from work for on-duty driving: changes in OR for significant variables were less than 10%.

Model 4 tested each interaction between each risk factor and the type of journey. There was no significant interaction in the case of age, suggesting that age-linked risk is independent of type of journey. The impact of alcohol, on the other hand, differed according to type of journey, being significantly greater in commuting to work (“alcohol x commuting to work” interaction: OR = 6.77; p = 0.02) and in commuting home (“alcohol x commuting home” interaction: OR = 5.74; p = 0.02) than in personal journeys. Also, although the interaction between driving on duty and gender was not significant, being on duty nevertheless reduced the extra risk of male gender seen in other types of journey (“male x on-duty journey” interaction: OR = 0.61; p = 0.06). Finally, the impact of frequency of driving at the accident site differed according to type of journey. In the case of sites driven through less than once monthly, risk was greater when commuting to work than when on a personal journey (“site driven through less than once monthly x commuting to work” interaction: OR = 13.51 (p < 0.01); “site driven through very rarely or for the first time x commuting to work” interaction: OR > 50 (p < 0.01)).

Analysis by type of journey (Table 3) for on-duty drivers showed lower risk for bus and coach drivers (OR = 0.29 [0.14; 0.59]) and higher risk for light utility vehicle drivers (OR = 3.97 [1.42; 11.05]) than for private and company car drivers. Light utility vehicle drivers’ risk was especially high for professional drivers (OR = 29.83 [5.19; 171.38]). Risk was significantly higher for on-duty drivers on temporary or work/study contracts than for manual workers (OR = 11.64 [2.15; 62.94]). Multivariate analysis including all the variables in Table 3, plus age, did not significantly change results for commuting to or from work or on-duty driving; changes in OR for significant variables were less than 10%.

Finally, on-duty driver risk did not significantly differ between taxi, ambulance, fire engine, police vehicle, school transport and dangerous goods vehicle drivers versus drivers of other vehicles, with or without adjustment on age and gender.