Study of mobility, road crashes and associated factors in students at university of Granada
- Pablo Lardelli Claret Director
- Elena Espigares Rodríguez Director
- Juan de Dios Luna del Castillo Director
Defence university: Universidad de Granada
Fecha de defensa: 01 April 2011
- Aurora Bueno Cavanillas Chair
- José Juan Jiménez Moleón Secretary
- Luis de la Fuente de Hoz Committee member
- María Seguí Gomez Committee member
- Joaquim Jorge Fernandes Soares Committee member
Type: Thesis
Abstract
1. BACKGROUND AND RATIONALE Even today, when talking about - Road Crashes (RC), we tend to consider them as unpredictable events that happen completely by chance and therefore, difficult to prevent. That is the reason why the term "Injuries caused by traffic" (ICT) is preferred, since it allows qualifying that the event or series of events can be analysed from a rational point of view, and preventive measures can be set up (Jiménez, 2007). The ICT are the ninth cause of mortality and of Disability-Adjusted Life Expectancy (DALE) all over the world. Their upward trend will put ICT as the fifth cause of mortality and the third cause of DALE in 2030. (WHO, 2009a; González, 2009). They mainly impact on young population, being one of the 3 first causes of mortality of people between 5 and 44 years, and the first one for people between 15 and 29 years. (WHO, 2009a). According to gender, the frequency of involvement of men in ITC is 3 times higher than for women (Lyles et al., 1991; Factor, 2007). Spain, with 6.8 deceases / 100,000 inhabitants, is in an intermediate position in the ranking of the 25 EU members (DGT, 2008). According to the data provided by the National Institute of Statistics (INE) in 2007, the highest mortality rate due to ICT was registered in young people from 15 to 34 years old, especially in the group of people from 20 to 24 years old, for which this rate reaches the value of 13.4 deaths per 100,000 inhabitants. By gender, mortality rate was higher for men (13.5 per 100,000 inhabitants) than for women (3.5). The annual cost derived from ICT worldwide is estimated in 518 billion dollars (359 million Euros), which represents between 1 to 3% of a country's GDP (Peden et al., 2004). In Spain, during the last four years, the global cost of ICT amount to 144.000 billion Euros (Moclús, 2008). Despite ICT are admitted as a first magnitude public health problem, they are, paradoxically, one of the most forgotten causes of premature deaths in epidemiology (Plasència 2003; Seguí, 2007). However, it is well known that the analytic epidemiology of ICT can be approached in the same way as many other health problems (Haddon, 1969; Bull, 1986; González, 2007). More specifically, a complex group of risk factors may affect a causal chain composed by five links ordered according to the following sequence: total population, intensity of exposure, road crashes, harmfulness, and the final health consequences (mortality, residual or chronic harmfulness, disability, cost, etc.). Most epidemiological research about ICT, worldwide and specially in Spain, is focused on the last links of the abovementioned causal chain, basically due to the ease of collecting data related to harmfulness and mortality, coming from secondary sources (medical assistance and police records) (Ferrando, 1998; Kelley, 2003; Pérez, 2006). This fact has lead to a lower knowledge about risk factors or markers acting on the first links of the causal chain, that is, those affecting intensity of exposure and number of road crashes. Several strategies have been implemented in order to face up this problem. Firs approaches involved decomposition methods based on ecologic designs (Wakefield, 2008; Morgenstern, 1995). However, these methods do not allow the knowledge of the causal associations between risk factors and outcomes at an individual level (Wakefield, 2008; Morgenstern, 1995). Record linkage of data coming from different registers (police + medical assistance) is another option. Nevertheless, independently of the imitations inherent to the use of registers (Jeffrey, 2009; Petridou 2009; Razzak, 2005), such information appears once again focused basically on harmfulness and mortality. Unfortunately, this information turns out to be incomplete in order to know and prevent ICT in a comprehensive way. Even so, we consider that the implementation and following of a concurrent cohort of road users is the best alternative in order to determine which risk factors and/or markers are associated to every link of the causal chain, and how they influence people moving from one link of the chain to the next one. The suitability of such type of studies has been pointed out by several authors (Ivers, 2009; Begg, 2003, 2009). However, there are not many studies of this type up to date. The cohort study of New Zealanders drivers (Begg et al., 1999; 2003), the Gazel cohort (Nabi, 2007) and, in Spain, the Monitoring cohort by the University of Navarra (SUN) (Seguí, 2007b), represent some examples. However, the last two cohorts were designed with the aim of being multipurpose cohorts, so they do not allow a complete analysis of the causal chain. For the time being, cohort studies on young population focused to determine their mobility (exposure), their risk of road crash and their risk of injuries after the crash, as well as the factors affecting each one of the former links, have not been developed in Spain. Taking into account all the above comments, we concluded that it would be highly justified to design a pilot study aimed to assess the feasibility of implement a cohort of young road users in our environment. 2. OBJECTIVES GENERAL AIMS 1. To assess the feasibility of implementing and monitoring a dynamic and prospective cohort of users of road transport vehicles, starting from the population of University grade students of the Department of Preventive Medicine and Public Health at the University of Granada. 2. To describe, for the aforementioned population, models for mobility, use of safety devices and number of road crashes, and for the subgroup of private car drivers, their driving styles. 3. To identify, in the subgroup of private car drivers, the factors associated with their amount of exposure, their driving styles, and their number of road crashes as drivers. SPECIFIC OBJECTIVES Regarding General aim 1: 1.1. To determine how many of the university students of the Department of Preventive Medicine and Public Health at the University of Granada, consent to participate in the study and complete the general questionnaire. 1.2. To estimate the volume of drop-outs in the baseline population after one year of monitoring, as well as the factors associated with the drop-outs. Regarding General aim 2: 2.1. To identify the existence of a potential internal structure in the matrix of driving styles designed for this study. 2.2. To describe the mobility patterns of the university students of the Department of Preventive Medicine and Public Health at the University of Granada, as drivers and passengers of two and four wheels vehicles. 2.3. To describe, for the population mentioned above, the frequency of use of safety devices depending on their position in the vehicle. 2.4. To describe, for the same population, the yearly frequency of road crashes, as well as the main features of the crashes reported. 2.5. To describe, for the subgroup of private car drivers, their driving patterns as well as their frequency of involvement in circumstances potentially associated with a higher risk of suffering a road crash. Regarding General aim 3 3.1. To quantify, among the university students of the Department of Preventive Medicine and Public Health at the University of Granada who declared to be car drivers, the association between the amount of their exposure as drivers, their driving patterns and circumstances and the frequency of road crashes reported in the previous year. 3.2. To analyse the association between the aforementioned variables with the gender, age and seniority of the driving license reported by the drivers. 3.3. To assess the changes in the magnitude of the recorded variables occurring in the cohort of drivers during one year of follow-up, as well as the association between such changes and the variables determined at the beginning of the follow-up period. 3. METHODS 3.1. DESIGN A mixed study (cross-sectional + prospective cohort) has been designed. 3.2. GEOGRAPHICAL SCOPE The city of Granada (Spain). 3.3. STUDY POPULATIONS a) Target population All the students of any of the grade subjects offered by the Department of Preventive Medicine and Public Health at the University of Granada, under the degrees of Food Technology and Science, Nursery, Pharmacy, Medicine, Physiotherapy, Human Nutrition and Dietetics, Odontology, Social Work and Occupational Therapy. b) Sampling population A sample of the target population complying with the following inclusion criteria: 1. Being a registered student during one of the following academic years 2007/08; 2008 /09 or 2009/10. 2. Having attended at least one class during the first two academic weeks of their respective subject. 3. Accepting to take part in the study after being informed about it. c) Sample Of the cross-sectional study This project was conceived as a pilot study. One of its purposes was to assess the participation rate, in order to establish the sample size required for the final cohort study. Therefore, we did not define a sample size for the present study; we included all students of the sampling population. The number of students registered on all the grade subjects previously defined during the three academic years was 7418. After applying the pre-specified inclusion criteria, we obtained an initial sample of 1597 students (21.5% of the whole target population in the three academic years). The final sample size was 1595 students, because we excluded two students who committed severe inconsistencies when they filled in the study questionnaire. Of the cohort study From the whole sample described above, we began the cohort study with a sub-sample of 269 students, recruited during the 2007-2008 academic year. These students gave us information about their E-mail, post address and/or phone number in the questionnaire, in order to allow their follow-up one year later. From this sample, we selected all the 239 students who reported to have driven a car in the year before the survey and could be allocated to a specific category of driving exposure (see later). 3.4. SOURCES OF INFORMATION A self-administered questionnaire (the 10th version of a series of questionnaires designed in 2007 by our research group) was used to collect the data. This version was used for the baseline filling out and, one year later, for the filling out by the students followed-up in our cohort. This questionnaire is divided in 5 sections (Annex 1): 1. Personal details. 2. Information about amount of exposure: km/year travelled in the year previous to the survey by the different type of users in routes authorised for road traffic. 3. Information on the use of safety devices in roads and urban areas. 4. Information about the drivers: years holding the driving license, perceived speed and driving quality, and the involvement, during the month previous to the survey, in 28 circumstances that could happen during driving. Some of them are considered by literature as a risk of suffering a road crash. 5. Data on road crashes in the last year. Regarding reliability of the data about the amount of exposure, once compared the collected data with the reviewed version of the Driving Habits Questionnaire by Owsley et al. (1999), an acceptable consistency between both tools was obtained, and nowadays the validation of the driving circumstances matrix is being carried out. 3.5. STUDY PROTOCOL Handing out of the baseline questionnaire was carried out during the two first academic weeks of the respective teacher. The put in practice and monitoring of the cohort started from a population sample made up by 239 drivers in the academic year 2007/2008. A year later of having filled out the baseline questionnaire, they received by post or email, the monitoring questionnaire. The diagram followed to systematise such deliveries is showed in figure 16. 3.6. VARIABLES Different variables have been considered, divided into four groups with the aim of modelling the phenomenon studied in a Directed Acyclic Graph (DAG) based on the ICT causal chain. Therefore, we can consider: a) Amount of exposure (Km /year). b) Intermediary variables: use of seat belt, speed and driving quality perceived by the own driver in comparison to other drivers and the involvement in driving circumstances in the last month. c) Number of road crashes. d) Confounding variables: gender, age and length of the driving license. 3.7. ANALYSIS 3.7.1. Analysis of the internal structure of the driving circumstances matrix: In order to analyse the internal structure of the driving circumstances matrix, for the subgroup of car drivers, circumstances C11 (due to have being formulated in two different questions in the different academic years) and C15 and C16 (referred to have suffered road crashes with and without injuries in the month previous to the survey, due to the low frequency of positive answers) were excluded. With the 25 remaining circumstances, two complementary analysis strategies were applied: 3.7.1.1. Factorial analysis, with factors extraction by means of the principal components method. In order to determine the number of factors, a self-value higher than 1 was considered. Subsequently, a varimax rotation was applied. For every factor extracted, its corresponding index was defined using the sum of the circumstances included in it. The internal consistency of the global scale and of the extracted factors was assessed by means of the Cronbach's alpha coefficient. In a second step, the factorial analysis was repeated, ruling out those circumstances clearly not associated with the frequency of having suffered an accident in the previous year (p>30). 3.7.1.2. Mokken Model: This model is proposed to define the structure of the binary items of a scale that follow a hierarchical structure between them. It is based on the following three hypotheses: 1º) The probability of a behaviour being adopted by a driver depends on the risk inherent to the driver (¿) and on the features of the behaviour making it more or less frequent (¿). Both of them are modelled in terms of the Logit transformation of this difference: (exp (¿-¿)/(1+ exp(¿-¿)). If both components influence the involvement in a certain circumstance with a similar magnitude, the probability of being involved in such circumstance will be 0,5; if the punctuation in the risk scale depending on the driver is higher than the risk scale depending on how much "compulsory" the behaviour is, the probability will be higher than 0,5, and vice versa. 2º) The probability of a driver having a certain behaviour is higher when higher his or her risk punctuation is, and this is valid for all the behaviours. 3º) The higher the number of behaviours that a driver has marked, the higher the probability of he or she having a certain risky behaviour. The model verifying the aforementioned hypothesis is called monotone homogeneity model, and the one which is valid for all the behaviours in a scale, and for all the drivers, is called double monotony model. In order to verify the double monotony of a scale (or sub-scale), obtained from a set of binary choice items (as our driving circumstances), an algorithm is applied which try to preserve the monotony of the scale, that is, to construct scales in which the less daring the drivers are, the less probability of developing a certain behaviour they have and, for a given driver, it would be more likely that he or she marks a frequent behaviour than a non-frequent one. The algorithm is based on the idea of counting, for a given pair of behaviours, how many times the abovementioned hypothesis are violated in the sample (Observed violations: Ov) and how many times we would expect such violation happening, if the monotony hypothesis were verified (Expected violation: Ev). The higher the value: 1-(Ov/Ev), the higher the degree of monotony. This value is called H coefficient and its magnitude measures the monotony degree (for H>0). The algorithm is performed in a step by step sequence. 3.7.2. Descriptive Study The frequency distribution of the sample across categories of each categorical variable recorded in the study has been obtained, including and excluding missing values. Mean, median, range and standard deviation have been obtained for continuous variables. Depending on each variable, this study has been performed for the whole sample or for the following specific subpopulations of road users: a. Cyclists. b. Drivers of motorized vehicles. c. TWMV drivers. d. Car drivers. 3.7.3. Analytical Study The analytical study has been restricted to the subpopulation of private car drivers. In order to analyse the magnitude of the association between the different variables, the sequence marked by the different arrows contained in the Directed Acyclic Graphs (DAGs), has been followed. Such DAGs have been worked out both for the cross-sectional study and the subsequent cohorts design (Figures 20 and 21). The following bivariate and multivariate analyses have been applied: logistic regression (for dichotomic variables such as number of accidents, always using the seat belt or involvement in risky situations), multinomial logistic regression (for polytomous variables such as levels of exposure, or self perceived speed and driving quality), one-way analysis of variance (ANOVA) (for quantitative variables or approaching to a normal distribution) and, finally, multiple lineal regression (for quantitative variables such as age, number of circumstances of years having the license). The introduction of the variables in the different models has been done attending to one or more of the following three criteria: 1. A possible confounding effect introduced by each variable. 2. Statistical significance of the association starting from previous models. 3. Stepwise procedure with a remove p-value of 0.10. 3.8. COMPUTER SUPPORT All the analysis were done using the statistical software STATA version 11.0 4. RESULTS 4.1. CROSS-SECTIONAL STUDY 4.1.1. Analysis of the internal structure of the risk matrix circumstances 4.1.1.1. Factorial analysis Factor analysis identified six factors, which accounted for 47.9% of the total variance. Factor 1 included the six driving situations most often mentioned by respondents (driving at night, driving alone, driving on highway or motorway, driving under rain, snow or fog, driving listening radio and changing the station and driving over the speed limit). Factor 2 included five driving situations, three of which explored attitudes potentially associated with distractions (talking on the phone, changing the CD and eating) and two were related to fatigue (driving drowsy or more than two hours without rest). Factor 3 grouped three circumstances that explored the traffic code violations (Not respecting a traffic light, a stop sign or a pedestrian crossing), along with two others: I have suffered a distraction at the wheel and a passenger told me that I run much. Factor 4 included two driving circumstances which explored the driving under the influence of alcohol, along with not using seat belt. Factor 5 grouped three driving situations associated with aggressive driving style (To beep to the front driver, to argue with other drivers and to overtake when it is not allowed), along with other associated with a traffic violation (I have received a fine from the police). Finally, Factor 6 included to smoke and other drugs use. The Cronbach alpha obtained for the complete matrix was 0.846. Coefficients of the dimensions identified in the factor analysis ranged from 0.821 to 0.285 for factor 1 and factor 6 (table 22). 4.1.1.2. Mokken method Two subscales were identified which met the condition of double monotony (tables 25 and 26). Three circumstances (To argue with other drivers, to overtake when it is not allowed, and driving without seat belt or helmet use), did not show a hierarchical structure and were omitted from the scales. Cronbach alpha coefficients of both subscales were 0.749 in the first and 0.731 in the second (Table 27). The correlation between scales was high, with a correlation coefficient of 0.6664 (P <0.001) 4.1.2. Descriptive study The ages of the students making up the sample ranged between 18 to 69 years old (mean = 22. 8, SD = 4.5, median = 22). Mean age in the subgroup of car drivers was a bit higher (23.3 years). Regarding gender, a 76.1% (73.2% in the subpopulation of car drivers) were women and the predominant nationality was the Spanish one, with 96.4% of those polled having this nationality (Tables 1 and 28) Regarding the amount of exposure measured in km/year travelled by the different road users, there is a predominance of exposures < 500 km/year for cyclists and users of 2 wheels motorised vehicles (TWMV). Around 70% of car drivers and passengers had exposure levels lower than 5000 km/year (Table 2). All the passive safety devices were more frequently used in roads than in urban areas. The greater differences can be found in the "always" use of the seat belts in the back seats of the car (66.7% in roads vs. 46.8% within the city). In the subgroup of car drivers, 10% of them declared not always using the seat belt when driving in urban areas (Tables 3 and 30). In this subgroup, the most frequent length of the driving license possession period was between 2 and 3 years. Regarding the speed to which they perceived to be driving, 45.3% of them declared to drive at the same speed as the rest of drivers, opposite to a 18.7% who declared to drive faster. The 53.1% of the surveyed people perceived themselves as good or excellent drivers (Tables 31 and 32). Table 33 shows the driving situations in which drivers were most frequently involved in the month previous to the survey. With a frequency higher than 70% they pointed out: driving at night, driving alone, driving in highways or motorways, driving under rain, snow or fog, listening the radio and tune in to a radio station. Regarding the remaining situations, the frequency is always lower than 50%. We can highlight: driving over the allowed speed (47%), changing a CD (37%), not observing the pedestrian crossings (34%) and driving using the mobile phone at once (24%) Regarding road crashes, 10.9% of the students declared having suffered any type of road crash in the previous year. In the subgroup of car drivers, 4.9% reported a crash when driving. Most of the students who reported a road crash declared not having suffered injuries as consequence of it (Tables 4, 5, 35 and 36) 4.1.3. Analytic Study A higher amount of exposure is associated to a higher risk of to suffer road crashes, specially for exposures higher than 1000 km/year. However, the adjusted models show that most of this association is due to the association established between the exposure and the involvement in certain risk driving circumstances, especially the following ones: drowsy driving (ORa=2.61), driving during more than two hours without rest (ORa=1.98), driving over the speed limit (ORa= 2.51) and do not observe a STOP sign (ORa=2.10). In our study, when such circumstances are taken into account along with the confounder variables, the association between exposure and road crash did not show a clear pattern (Table 91). The small number of drivers who reported to have suffered injuries in the crash, has prevented us from analyzing the association between the road crashes and the rest of injuries-related variables included in the questionnaire. Now we are commenting the effect that the 3 confounding variables (gender, age and length of the license) had on the associations proposed by the DAGs. Male gender is solidly associated with a greater intensity of exposure (ORa=4.06 for the category of 5000 or more km with regard to the one of < 500), with a longest length of the license and also with higher perceived speed an driving quality (Tables 73 to 75). The involvement of males in all the circumstances is much higher than that for women. The greater differences were observed for the following circumstances: going over the speed limit (38.5% of women vs 69.9% of men), driving for more than two hours without rest (14.9% vs 32.6%), don't observing pedestrian crossings (29.7% vs 45.6%), driving alone (74.3% vs 90.2%) and driving under the influence of alcohol (10.2% vs 26 %). However, once adjusted all the intermediary and confounding variables, we observe that men reported less road crashes than women (Table 90). Considering the adjusted effect of gender and seniority in the license, age does not show a solid association with the intensity of exposure during the year previous to the survey (Table 73). However, it is detected that the greater the age, the lower the number of circumstances in which the driver was involved in the previous month. To be more specific, the older the person is, the lower is the frequency of: getting distracted while driving, listening the radio and tune another station and do not observing the pedestrian crossings, among others (Tables 78, 80 and 84). Nevertheless, an association between age and road crash was not detected (Tables 90 and 91). Regarding length of the driving license, it is observed that the higher the length of the license, the higher the amount of exposure. Lengths of license of 6 or more years were associated to the highest exposures, with ORa=11.05 compared to the newer drivers (Table 73). The drivers with more seniority in the driving license also declared to drive faster than the others and to be involved in a greater number of driving situations in the previous month, especially: driving alone, getting distracted while being at the wheel, listening the radio and changing the station, eating while being at the wheel, and driving more than two hours without rest (Tables 74, 77 to79 and 84). Finally, association between number of years having the license to have had an accident was not observed either (Tables 90 and 91). Table 85 presents the multiple regression models for the first five subscales derived from factor analysis (the last factor, with only two items, had not a good fit to normal distribution). Factor 1 showed a positive association with the intensity of exposure, perceived speed and perceived quality. Factor 2 was inversely associated with age and positively associated with the intensity of exposure and length of the driving license, as well as greater perceived speed and non seat belt use in urban areas. Factor 3 showed, again, an inverse association with age and a positive association with the intensity of exposure, with high level of perceived speed and male gender. Factor 4 was closely associated with the non seat belt use on open roads and urban areas, as well as with male gender, amount of exposure and perceived speed. Finally, factor 5 scores were positively associated only with an increase in perceived speed and the highest exposure category. The results of the regression models constructed for the two Mokken subscales are shown in Table 86. Both were closely associated with the intensity of exposure, length of driving license, male gender and greater perceived speed, although for the last variable, the association was much stronger in subscale 1. The inverse association with age appeared strongly for subscale 2. This last subscale was also positively associated with not using always seat belt in urban areas. Table 92 shows the three groups of logistic regressions developed to evaluate the association between frequency of road crashes and the scores of subscales created from factorial and Mokken analyses. When the amount of exposure and confounding variables were entered into the models, the scores for factors 1 to 4 derived from the factorial analysis were positively associated with the frequency of crashes, with adjusted ORs ranging from 1.80 to 1.34 for F1 and F3, respectively. On other hand, both Mokkens' subscales also showed a positive association with the frequency of crashes, with almost identical adjusted ORs (1.3). When all factors were included in the models only the scores of Factors 1 and 2 maintained their association with the crash, although their adjusted ORs were lower than the values obtained in previous models. Regarding to Mokken subscales, only the first one continued showing an association with the frequency of road crashes. Finally, the models with all factors, including confounders and the amount of exposure, showed similar results to those above, although in this case the adjusted OR for F2 was not statistically significant. The same occurred when the equivalent model was constructed for Mokken subscales: only the first one maintained its association with the frequency of road crashes. 4.2. COHORT STUDY Only 80 out of the 239 students who started the follow-up, answered the questionnaire one year after, once finished the deadline established in the figure 16 diagram. The high amount of drop-outs (66% of the initial sample) limits the validity of our estimates. However, it was possible to verify that the exposure one year later was strongly associated with the baseline exposure, with driving at the same speed as the other drivers and with the circumstances "not observing a red traffic lights" and "driving more than two hours without rest" (Tables 126 and 127). We also detected a sound association between the perceived speed and driving quality in the baseline questionnaire and one year later. Finally, the number of driving situations in which a driver was involved in the previous month in the follow-up questionnaire, was only associated with the same variable in the baseline questionnaire (Table 156). 5. DISCUSSION The choice of a mixed design (cross-sectional and cohort), is due to both the stated objectives and the conviction that, by means of this type of design, it was possible to highlight the effect that the different risk factors and/or makers have over the links of the epidemiological chain of ICT, verifying the causal connection, if applicable. The advantages of cohort design in the study of analytical epidemiology of ICT has been argued in several studies (Ivers, 2009; Begg, 2003 y 2009). Our study population was made up, as we have already explained, by students of different subjects of our Department at the University, who comply with the proposed inclusion criteria. The main reason for the election of such population was the accessibility and the feasibility a priori to obtain a sample size large enough to implement the piloting of a cohort of road users. Once the deadline established in the diagram proposed in figure 16 run out, our percentage of drop-outs, close to 67%, prevented us from reaching the goal. Some hypothesis may explain this high rate of drop-outs. First, a great number of students in the baseline sample were in the last course of their degrees, so they have left the university one year after the first survey. Second, we did not offer any compensation for continue the study, and this factor is well documented (Bosnjak, 2001). Regarding the questionnaire used to collect the data, it was developed by the research team of our Department. Up to date, different questionnaires have been used for the study of ICT; perhaps the most widely used has been the classic Driving Behaviour Questionnaire (DBQ), developed by Reason et al, focused exclusively on the human factor (Lajunen, 2004). Other self-completed questionnaires as the Young Driver Attitude Scale (YDAS), proposed by Malfetti et al. (1989), the Safety Attitude Questionnaire by Rundmo et al. (1996), or more recently, the Safety Behaviour Questionnaire (SBQ) by Ehring et al. (2006), are focused on the measurement of risky driving behaviour by the drivers. The majority of these questionnaires has been designed, applied and validated in contexts different to our country and under a cross-sectional perspective (Clapp, 2010; Ullerberg, 2002 y Lajunen, 2004). Our questionnaire brought, as added value, the collection of data not only about driving styles but also about intensity of exposure, road crashes and injuries suffered, by means of simple questions for which, unlike the previous questionnaires, no more than 7 minutes are required to fill in. The matrix of driving circumstances was not originally conceived as a unitary questionnaire to measure dimensions of risk in young drivers but only to identify behaviors associated theoretically to risk of suffering a road crash. Such circumstances should be easy to understand and remember to be included in a questionnaire aimed to recruit and follow-up road users. However, in order to design a definite version of this matrix, it should advisable to analyze its internal structure. Classical factorial analysis clearly detected five factors: The first one, included high prevalence circumstances linked to the driving activity. The inclusion of driving over speed limits in this factor suggests that this high-risk driving style is being incorporated into the normal driving patterns among young people (INTRAS, 2007 and 2008). The second factor included circumstances related with fatigue and distractions. This association has been also previously reported (Bergasa, 2008; Sandin, 2007; Bunn, 2005; Lawrence, 2003). Factor 3 grouped some traffic code violations circumstances as well as a driving distraction. Although some authors have noted the association between infractions and distraction (Wickends, 2008; Rakauskas, 2008) the low consistency of this subscale makes us be cautious about this kind of exploratory hypothesis. The two circumstances included in the following factor (driving under influence of alcohol and not using seat belt) have also been found associated among young drivers in previous studies (Lucidi et al., 2010; Hatfield, 2009; Bendak, 2005). Finally, factor 5 included circumstances clearly associated with aggressiveness. With the application of Mokken hierarchical method we obtain two subscales with a clear hierarchical structure within each of them. The first one starts by the circumstances associated with factor 1 and ascend hierarchically to the circumstances of the factors 2, 4 and 5. It resembles a way of progressive acquisition of risky driving styles classically associated with young drivers using cars for their leisure time activities (Tomas-Dols, 2010; Boufous, 2010; Fernandes 2010; Lucidi 2010; Shope, 2008; Laapotti, 2006. On the contrary, the path depicted by the second subscale seemed related with other uses of car (i.e., commuting). This subscale started by the circumstances associated with factor 1 and ascended hierarchically to those included in factor 3. The high correlation between both subscales speaks in favor of an interdependence of them in the same drivers: they could be interpreted as two "ways" which could be done independently, but many drivers in our sample do "together". Here below, we will comment the main results of our study that worth to be discussed. Women represent 76% of the students in our sample, according to the gender distribution existing nowadays in the health-related degrees at our University. According to the data provided by the National Institute of Statistics (INE), during the academic year 2008-2009, the percentage of women registered in Health Sciences degrees at the University of Granada was 69,1% (INE, 2010). Regarding the intensities of exposure, low exposure levels (<5000 km/year) predominate, accordingly to the data provided by literature for this population (Lardelli, 2011; Ministry of Health and Consumption, 2008). The percentage of students travelling as passengers is higher than that observed for driving, which is in accordance with the youth of our population (there are very few ones older than 30). After the private car, bus is the mean of transport more frequently used. Such find was also discovered by the Report on Mobility of the University of Cordoba in 2008 (UCO, 2008). Despite being a young population, the use of private car is predominant over the use of mopeds and/or motorcycles, both as drivers or passengers. The fact that our study is limited to university students can explain it, since university students may have a socioeconomic level higher than the average of population in this age stratum. There are multiple pieces of work which back the underuse of passive safety devices in urban areas opposite to the use in roads (DGT, 2008a; DGT 2008b; Williams et al., 2003b; Fernandes 2010; Nallet, 2010). In our sample, the devices less frequently used are seat belt in the back seats and the helmet when riding on the back. The frequencies of use that we got for them are lower to the ones reported in different studies by the DGT (DGT, 2009; 2008a). With regard to the results of the analytical study restricted to car drivers, we detect a solid association between amount of exposure and a greater number of road crashes. Such foreseeable association has been documented in previous studies (Laapotti, 2006; Lucidi, 2009; Labergue, 1992). However, when we consider the adjusted effect on road crashes of the so-called intermediary variables (perceived speed and driving quality, use of seat belt and driving situations), we observe that such association is fairly reduced, which highlight the importance of such variables for the road crashes, especially those related to driving circumstances. The variables more frequently associated to road crashes were those related to exceeding the speed limit, tiredness, distractions and driving offences. All of them are recognized as risky situations in many studies (Lucidi, 2010; Shope, 2008; Pérez-Díaz, 2004; Technical Research Centre of Finland VVT, 1998). Considering the adjusted effect of gender, age and years having the license over the different links of the epidemiological chain of ICT, we observe that male gender is associated with a higher intensity of exposure, a lower use of all safety devices, a higher perceived speed and driving quality and a greater involvement in all the driving situations. Such associations have been also detected in other studies (Giacomo, 2010; Sangowawa, 2010; Özkan, 2006; Laapotti, 2006). However, we get a lower number of road crashes for men than for women. The greater involvement of women in road crash with regard to men has been reported for older ages (65 or more years old) (Baker, 2003; NHTASA, 2007). Nevertheless, there are few studies in which a greater number of accidents of young women has been detected (Awadzi, 2008). A greater age was associated with a lower frequency of involvement in driving situations like: getting distracted at the wheel, listening to the radio and tuning the station, not observing pedestrian crossings, probably because of a higher perception of risk by the older drivers (Borowsky, 2010; Ivers, 2009). This fewer number of risky driving styles in older drivers opposite to the younger ones has been a common find in the literature about injuries caused by road crashes (Ivers, 2009; Mccartt, 2009, Laflamme, 2006). Regarding seniority of the driving license, a higher number of years having the license was associated with higher intensities of exposure, higher perceived speed and a greater involvement in risky driving situations. Probably it is due to a higher self confidence when driving generated by the experience (Mayhew, 2003b; Simpson and Mayhew, 1992; Maycock, 1991). Similar to what happen with age, we do not find association between either of these two variables, with the number of road crashes. However, other studies like the one by Wallet et al. (2000), demonstrated that both the involvement in risky driving styles and the number of road crashes underwent a decrease of 8% and 6% respectively for every additional year of seniority in the driving license. 6. CONCLUSIONS In response to the objectives established for this Doctoral Thesis, and in the light of the results obtained, the following conclusions emerge: 1. The implementation and monitoring of a cohort of road users, following the guidelines of the design piloted in this study, is not feasible. This is partially due to the low rate of recruitment from the target population but, particularly, to the high percentage of drop-outs during the follow-up, about 66% of the baseline sample. The dropped out drivers were younger and with less seniority of the driving license than those who remained in the cohort. 2. The amount of exposure as drivers of motorized vehicles was not high in our sample of university students. The non-negligible frequency of reporting not always using the seat belt when travelling in the back seats of the cars, particularly in urban areas, is worrisome. The reported frequency of road crashes in the previous year was 10%, although most of them did not result in injuries. 3. For the car drivers, the 25 driving circumstances matrix designed revealed a clear internal structure: the factorial analysis identified at least 5 underlying dimensions (high-frequency circumstances, tiredness and distractions, aggressiveness, violations of the driving law and consumption of alcohol). Furthermore, a hierarchical structure in such matrix could also be identified, with two subscales reflecting two ways in which drivers acquire a higher road crash risk: one linked to "leisure - high risk behaviours" and another one linked to "non leisure - violation of the law". These results support the usefulness of the circumstances matrix proposed as a basis for the design of subsequent questionnaires. 4. In the subgroup of car drivers, young age and few years of possession of a driving license explain a somewhat low amount of exposure: approximately 70% of them drive less than 5,000 km/year. One out of ten drivers reports not always use the seat belt in urban areas. With regard to the driving styles, half of them consider themselves as good or excellent drivers. In contrast, it has to be remarked how frequently they refer to have been involved, during the previous month, in situations potentially associated to a higher number of road crashes, like driving over the speed limits (nearly 50%), using the mobile phone while driving (25%), driving more than two hours without a break, or getting distracted while driving (approximately 20% each). 5. Longer exposure is associated with an increasing frequency of road crashes. However, this association is mostly mediated by that observed between the amount of exposure and the involvement in several high-risk driving styles: excessive speed, tiredness, distractions and committing driving offences, among others. When such circumstances are taken into account, a remaining association between exposure and number of crashes is not supported in this study. 6. Male gender is soundly associated with a greater amount of exposure, a higher seniority of the license, and a higher perceived speed and driving quality. Men are also more frequently involved than women in all the driving situations taken into account, as well as regarding the total number of circumstances. However, we have not detected an increased frequency of crashes among males. 7. When the joint effect of gender and seniority of the driving license is considered, driver's age is generally associated with a lower involvement in risky driving patterns: lower perceived speed, a fewer number of risky driving situations and a lower frequency of involvement in situations like driving with no seat belt or getting distracted while driving. Nevertheless, the narrow age range of the sample did not allow detecting an association between age and the frequency of road crashes. After adjusting for age, seniority of the driving license is associated with a higher exposure, a higher perceived speed and a greater number of driving circumstances referred in the previous month. However, no association between this variable and road crash was found. 8. Regarding the cohort study, several factors (the high proportion of drop-outs, the differences found in their characteristics when compared with monitored drivers, the short duration of the follow-up and the low number of crashes reported in this period), limit the quantity and quality of the results obtained from this design. Nevertheless, there are two facts worth to be highlighted: - The soundness of the answers given by those students who answered both questionnaires (at baseline and one year later) is quite high, and this indirectly supports the reliability of the questionnaire. - The association patterns obtained with the variables measured at the baseline and after one year are consistent with those described in the cross-sectional study, which supports a causal interpretation of the associations observed in the later design.