Exposure measurement in bicycle safety analysis: A review of the literature

Accident Analysis and Prevention 84 (2015) 9–19

Contents lists available at ScienceDirect

Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap

Exposure measurement in bicycle safety analysis: A review of the literature Jef Vanparijs a , Luc Int Panis b,c , Romain Meeusen a,∗ , Bas de Geus a a b c

Department of Human Physiology, Faculty of Physical Education and Physical Therapy, Vrije Universiteit Brussel, Belgium Flemish Institute for Technological Research (VITO), Mol, Belgium Transportation Research Institute (IMOB), Hasselt University, Diepenbeek, Belgium

a r t i c l e

i n f o

Article history: Received 30 March 2015 Received in revised form 1 August 2015 Accepted 4 August 2015 Keywords: Bicycle Exposure Accident Incidence rate Safety Active mobility

a b s t r a c t Background: Cycling, as an active mode of transportation, has well-established health benefits. However, the safety of cyclists in traffic remains a major concern. In-depth studies of potential risk factors and safety outcomes are needed to ensure the most appropriate actions are taken to improve safety. However, the lack of reliable exposure data hinders meaningful analysis and interpretation. In this paper, we review the bicycle safety literature reporting different methods for measuring cycling exposure and discuss their findings. Methods: A literature search identified studies on bicycle safety that included a description of how cycling exposure was measured, and what exposure units were used (e.g. distance, time, trips). Results were analyzed based on whether retrospective or prospective measurement of exposure was used, and whether safety outcomes controlled for exposure. Results: We analyzed 20 papers. Retrospective studies were dominated by major bicycle accidents, whereas the prospective studies included minor and major bicycle accidents. Retrospective studies indicated higher incidence rates (IR) of accidents for men compared to women, and an increased risk of injury for cyclists aged 50 years or older. There was a lack of data for cyclists younger than 18 years. The risk of cycling accidents increased when riding in the dark. Wearing visible clothing or a helmet, or having more cycling experience did not reduce the risk of being involved in an accident. Better cyclist-driver awareness and more interaction between car driver and cyclists, and well maintained bicycle-specific infrastructure should improve bicycle safety. Conclusion: The need to include exposure in bicycle safety research is increasingly recognized, but good exposure data are often lacking, which makes results hard to interpret and compare. Studies including exposure often use a retrospective research design, without including data on minor bicycle accidents, making it difficult to compare safety levels between age categories or against different types of infrastructure. Future research should focus more on children and adolescents, as this age group is a vulnerable population and is underrepresented in the existing literature. © 2015 Elsevier Ltd. All rights reserved.

1. Introduction Cycling as an active mode of transportation holds the potential to reduce traffic congestion and air pollution, and promotes an active lifestyle which in turn improves public health (Andersen et al., 2000; Higgins, 2005; Mueller et al., 2015). The health benefits of active commuting by bicycle are well established (Mueller et al., 2015; de Geus et al., 2008, 2009; Oja et al., 2011). However, safety concerns may be a drawback, especially for children,

∗ Corresponding author. E-mail address: [email protected] (R. Meeusen). http://dx.doi.org/10.1016/j.aap.2015.08.007 0001-4575/© 2015 Elsevier Ltd. All rights reserved.

adolescents and the elderly (Panter et al., 2010; Davison et al., 2008; Mindell et al., 2012; Maring and van Schagen, 1990; MartinezRuiz et al., 2014), age groups that incur more accidents than in adults (18–65 years) (Martensen, 2014). Cyclists often have to use the same infrastructure as cars, buses and trucks but are more vulnerable than the motorized road users as they are not protected by their vehicle in the case of an accident (Davis, 2001). Therefore, safety for cyclists must be improved if there is to be a modal shift from passive (motorized) transportation to active transportation. To create a safer cycling environment, we need to understand where, when and under what circumstances bicycle accidents occur. When analyzing the literature, different methodological

10

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

approaches are used. Some studies focus on injuries reported in hospital or police data (Dhillon, 2001; Boufous et al., 2011), some analyze the type of accidents (Meuleners et al., 2007; Keller et al., 2006; Boufous et al., 2013) and others analyze the relation between the built environment and infrastructure, and the occurrence of bicycle accidents (Dumbaugh, 2012; Harris et al., 2013). Regardless of methodological approaches, exposure data are needed for a better interpretation of the risk of being involved in a bicycle accident (Christie et al., 2007; Howarth, 1982; Roberts et al., 1997). An illustration of the need to take exposure into account comes from the study by Daniels et al. (2009). They concluded that “Roundabouts that are replacing signal-controlled intersections seem to have had a worse evolution on the number of bicycle accidents compared to roundabouts on other types of intersection” (Daniels et al., 2009). As the authors indicate, no correction for specific variations in traffic volume (exposure) was possible. If more people started using those roundabouts (increasing exposure) compared to signal-controlled intersections, the study conclusions may have been that the risk of being involved in a bicycle accident is statistically lower at those roundabouts. It is thus critical to take into account exposure when making decisions or producing bicycle safety guidelines based on such observations. When evaluating bicycle safety, several independent factors should be considered. First, there is the ‘demographic factor’. People of different ages, gender or socioeconomic status may need different traffic safety guidelines. The traffic situation and safety guidelines should be adapted to the users. For example, children cycling to school do not perceive the infrastructure in the same way as adults (Ghekiere et al., 2014; Timperio et al., 2004). Therefore, children may need different cycling infrastructure and safety guidelines compared to adults. A second factor that has an impact on cycling safety is the ‘built environment’ (Ewing and Dumbaugh, 2009). This includes ‘infrastructure’ and ‘other traffic flows’. The level of safety of an infrastructural design depends not only on its users (a demographic factor) but also on traffic speed and density. A third factor is the ‘the weather’ and ‘lighting conditions’. It is possible that winter and summer conditions produce different cycling outcomes. Traffic flow and lighting conditions also vary diurnally, as does the weather. Safety factors differ between a dark snowy winter day and a bright summer day. The final factor is the ‘behavioral’ factor. This factor describes behaviors in traffic such as wearing a helmet, the speed of cycling, listening to music or using a cell-phone. The transport mode and purpose of the trip are also variables. Since all these factors interact with each other, analyzing the safety of one factor should include controls for the other factors. The aims of each study will determine the factors to be investigated. In addition to controlling for confounding factors, each factor should be controlled for at least one exposure unit in order to allow meaningful comparisons. Well-chosen exposure units should ensure that findings are robust, can be challenged and are consistent regardless of the scientific research paradigm applied (Stevenson, 2014). Comparisons can then be made between different studies using the same exposure units. In particular, the comparison of different infrastructural designs between countries may help to provide insights into improving bicycle facilities. Different countries show differences in cycling activity, built environments, infrastructure, travel behavior and age distribution, highlighting the importance of exposure reporting (Christie et al., 2007). Another advantage of consistently including exposure parameters in bicycle safety research is that it allows trends in exposure over time to be observed. This can be used to understand temporal trends in accident data, the consequences of policy interventions or the effect of new infrastructure on cycling safety. Several methodologies – each with advantages and disadvantages – have been developed to

estimate or calculate bicycle exposure (Vandenbulcke et al., 2014). The purpose of this review is to firstly examine the different methodological aspects related to measuring cycling exposure, and discuss their advantages and disadvantages. Secondly, the findings from the selected papers dealing with bicycling exposure will be discussed, with suggested approaches for future research. 2. Search strategy We searched for papers related to bicycle accidents and exposure in four different databases: Pubmed, Web of Knowledge, ScienceDirect and Transport Research International Documentation (TRID). Only English language peer reviewed papers were included in this review. To identify relevant studies, the search terms used were bicycle, crash, accident, exposure, safety and infrastructure, including wild cards and in all possible combinations. Papers reporting any form of exposure measurement related to bicycle safety were included. Papers that did not measure exposure directly but estimated it using mathematical models were excluded. Only papers dealing with ‘utilitarian cycling’ (defined as ‘commuting to or from work’ or ‘cycling to other destinations’) were considered. ‘Leisure’ and ‘sports-related’ cycling (e.g. road cycling for competition or mountain biking) papers were excluded. The initial literature search was conducted in spring 2014 and had an outcome of 1578 hits. After the first screening for relevant titles and/or abstracts, 27 papers remained. The bibliographies of these 27 papers were scrutinized, resulting in the inclusion of 8 additional papers. The full texts of the 35 included papers were reviewed for relevance. Twenty papers (2% of the initial search) were retained for further analysis (Table 1). 3. Methodology and study design 3.1. Methods used to collect accident data Studies of bicycle safety that report exposure data require two data categories, one for the exposure parameters (denominator) and the other for accident data (numerator). Several approaches may be used to collect accident data. First, both data categories (numerator and denominator) may come from the same cohort. Selecting cyclists involved in an accident through official accident registrations (e.g. hospital or police data) and asking for their exposure parameters is a possible approach. However, this would not necessarily represent bicycle exposure of the actual population or infrastructure, because people not involved in accidents are not taken into account. This would miss cyclists who demonstrate factors leading to safe bicycling, for example the choice of safer routes or taking better safety precautions than accident victims. This limitation can be overcome by including matched controls from a population not involved in an accident, but this increases the scope of a study. Another possibility is combining data from official accident registrations with exposure data from national statistics. This includes cyclists who have not been involved in an accident. However, the exposure and accident cohorts are not matched. Another drawback is the incompleteness of official accident registrations. Only 7.1% of the incidents observed by de Geus et al. (2012) were also registered by police. A third approach overcomes these weaknesses by including a single, more general cycling population for accident and exposure data collection, even though respondents may not (yet) have been involved in an accident. The follow-up study design includes a general cycling population and uses the same cohort for exposure and accident data. This includes cyclists that are not involved in bicycle accidents and thus

Table 1 Studies investigating bicycle safety with reported exposure data. Design

Accident data and severity

Methodology

Participants

Exposure type

Definition of incidence, rate and risk

Aim and factors influencing accidents

Results and conclusions

Aultman et al. 1998 (30) Ottawa-Carlton, Canada

Retrospective 1 year and 3 years previously

Questionnaire asking about collisions in the past 3 years

Questionnaire and map, distributed on parked bicycles

1445 responses; 26% women 74% men 4 participants under 18 years

Distance cycled on-road, off-road and on side-walks

Safety analysis of different infrastructures; built-environment

Cycling on-road is safest followed by off-road cycling; Highest IR was found for side-walk cycling

Aultman et al. 1999 (31) Toronto, Canada

Retrospective Up to 1 year and 3 years previously

Questionnaire asking about collisions in the past 3 years; Self-reported accidents

Questionnaire and Map, distributed on parked bicycles

1169 full responses; 40% women 60% men

Distance cycled on-road, off-road and on side-walks

Safety analysis of different infrastructures; built environment

IR off-road cycling was lowest; IR/km for bicycle was 26–68 times higher compared to automobile; Exposure and attitude of cyclists affect safety

Bacchieri et al. 2010 (32) Brazil

Retrospective Up to 1 year previously

Questionnaire asking about accidents in the past 12 months

Face to face interview asking about cycling aspects and accidents over the past 12 months

1133 participants >20 years 100% men

Average time/day and days/week over the past 12 months

ER: N incidents on a particular facility type reported by a particular group by the total cycle km traveled on that facility type by that group; RR: ER of one type of infrastructure/ER of another type of infrastructure ER: N incidents on a particular facility type reported by a particular group by the total cycle km traveled on that facility type by that group; RR: ER of one type of infrastructure/ER of another type of infrastructure % accidents (IR): N of accidents for each variable (e.g. age, exposure, risk-taking behavior); RR: IRV1 /IRV2 , where ‘V’ is a variable (e.g. exposure) and ‘1’ is variable component (e.g. cycling 6 days/wk)

Analysis of safety equipment use and risk behaviors and relation to accidents; Behavioral factors

Blaizot et al. 2013 (33) Rhône county, France

Retrospective

Regional household travel survey; Interview about one specific days travel behavior

Weighted for all ages

Distance; time; trips

IR: N injuries per exposure measurement, and scaled per 1 million trips, km or h

Injury risk of cycling compared with car, walking and PTW; Behavioral factors

Mindell et al. 2012 (9) England, UK

Retrospective

Police-reported crash data Hospital-based crash data; All injury severity for the periods 1996–1997 and 2005–2006 Hospital data; All severity, including fatal 2007–2009 data

National travel survey 2007–2009 data

Weighted for all ages

Time; distance

Risk (IR): observed N casualties per unit of exposure (distance traveled (billion km) and time spent traveling (million h))

Characteristics of traffic injuries for different travel modes; Demographic & behavioral factors

Most items of bicycle safety equipment and the risk behaviors of bicyclists showed no significant association with accidents; Only commuting by bicycle seven days per week, as opposed to five or six, and a combination of extremely imprudent behaviors such as zigzagging through traffic, riding after drinking alcohol, and high-speed riding were found to be risk factors for accidents. RR for PTW was the highest; RR for cyclists was lower in 1996–1997 than in 2005–2006; IRs are higher in less dense built environments Risk of fatal accidents and hospital admissions for cyclists increases with age, especially in males

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

Author and location

11

12

Table 1 (Continued) Design

Accident data and severity

Methodology

Participants

Exposure type

Definition of incidence, rate and risk

Aim and factors influencing accidents

Results and conclusions

Lusk et al. 2011 (34) Montreal, Canada

Retrospective

Hospital data and police-recorded accident data. All severities

Data from automated traffic counts; Analysis of cycle tracks versus reference roads

All ages

Distance (calculated from number of passages)

Analysis of RR of cycle tracks (built environment)

Two-way cycle tracks have lower or similar RR compared to cycling on the street

Rodgers et al. 1995 (35) 48 states of US and Colombia District

Retrospective Up to 1 year previously

National Center for Health Statistics; Highway Traffic Safety System; Fatal accidents

Telephone questionnaire about travel behavior

6076 calls, 4346 qualified responses and 1254 bicycle riders

Time

RR for each bicycle track: (N injuriestrack /bikestrack )/(N injuriesref /bikesref ), where injuriestrack and injuriesref are the count of injuries on the cycle track and reference street(s), respectively; bikestrack and bikesref were the corresponding cyclist counts RR: ¨ ¨ (P(X/D)/P(X))/(P(X/D)/P( X)), ¨ where P(X/D) and P(X/D)

Higher RR for males, for bicyclists over the age of 44 and for riding after dark

No injury; self-treated injuries; injuries for which medical treatment from a general practitioner or hospital (but not overnight stay) was sought or injuries that required at least one night in hospital Self-reported crashes; Only if crashes disrupted daily activities for at least 24 h

6 survey weeks in 1 year via the web.

2038 participants >18 years 72 (4%) male

Distance; time

Incident Crash (=Crash Rate) and Injury Rates: total number of crashes or injuries reported per 1000 h or per 1000 km traveled

Determine RR for bicyclists according to age, gender and lighting conditions; Demographic and weatherrelated factors Measuring exposure based crashes and incident rates; demographic, behavioral factors & built environment

Survey distributed to participants of mass bicycle event (2006 Wattyl Lake Taupo Cycle Challenge)

2469 responses >16 years

Distance (estimated by participants)

Exposure (km cycled/year) was used to adjust the regression models

Low visibility may increase the risk of crash-related injuries

National travel survey

All ages

Time

IR: (total N cycling injuries/year)/(total time spent cycling (million h)/year)

Described methods and characteristics of participants of a longitudinal study of cyclists; demographic & behavioral factors Assessing regional variations in traffic injuries in pedal cyclists; built environment

Poulos et al. 2015 (36) New South Wales, Australia

Retrospective (baseline) and Prospective 1-year follow-up

Thornley et al. 2008 (37) New Zealand

Retrospective Up to 1 year previously

Tin Tin et al. 2011 (38) New Zealand

Retrospective

National dataset of all deaths registered in New Zealand; National dataset about patients discharged from hospitals

are estimated from the fatality databases, and P(X) ¨ are estimated and P(X) from the exposure survey database

Crash rate (CR) 0.29/1000 km or 6,06/1000 h; Injury Rates 0.148/1000 km or 3.09/1000 h; CR and IR are greater for females, less experienced and commuter cyclists

Indication of risk in scarcity for New Zealand cyclists

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

Author and location

Retrospective Up to 22–2 years previously

Teschke et al. 2012 (40) Toronto and Vancouver, Canada

Retrospective

Twisk et al. 2013 (41) The Netherlands

Retrospective

Hospital data for seriously injured cyclists (MAIS 2 or more)

Vandenbulcke et al. 2009 (42) Belgium

Retrospective; 2002–2005

Hospital data for bicycle accidents including all severities

Yiannakoulias et al. 2012 (43) Hamilton, Canada

Retrospective

Cyclist-motor vehicle collision database for Hamilton, based on hospital and police data

National dataset of all deaths registered in New Zealand; National dataset about patients discharged from hospitals Participants recruited through hospital data if they were involved in a bicycle accident

National travel survey

>5 years

Time

IR: (total N cycling injuries/year)/(total time spent cycling (million h)/year)

Face to face interview asking about accident information and control site information Dutch National Travel Statistics

≥19 years

Risk of injury (adjusted odds ratio) for different types of infrastructure

RR: estimated via OR = (N injury sites route type n/N control sites route type n)/(N injury sites reference route type/N control sites reference route type)

All ages

Time; distance

Census data from National Institute for Statistics

18–65 years

Time; distance

IR: N seriously injured cyclists/distance traveled (millions of km); RIR: ratio between IR in darkness (two conditions) and the IR for that particular age group during daylight Risk: N average annual injuries/total cycling time (return trip)

Canadian Census: start and endpoint of commuting and mode of transport

>15 years

Distance

RR: yi /ei where yi is observed and ei is expected N collisions in census tract i, and where expected N collisions is equal to the total cyclist km traveled in tract i multiplied by the study area rate of collisions (total collisions divided by total km traveled)

Exposure based analysis of bicycling accidents and injuries; demographic & behavioral factors Analysis of injury risk of 14 route types; built environment

Increase of fatal and hospitalized injuries over last decade

A lower IR on quiet streets and streets with bike-specific infrastructure

Analysis of risk of cycling in the dark; weather

Increased IR at dusk compared to daylight; Higher IR at dawn linked to alcohol use

Exploring the variation of bicycle use when commuting with the level of urban hierarchy; built environment Mapping risk of collision between cyclists and motor vehicles; built environment

Big discrepancies between different communes/regions suggests a spatially differentiated policy for cycling safety

Map of Hamilton with IR

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

Tin Tin et al. 2010 (39) New Zealand

13

14

Table 1 (Continued) Design

Accident data and severity

Methodology

Participants

Exposure type

Definition of incidence, rate and risk

Aim and factors influencing accidents

Results and conclusions

Beck et al., 2007 (44) USA

Retrospective

Travel exposure data (person-trips): 2001 NHTS (nationally representative sample of households)

All ages

Person-trip: defined as a one-way journey between two points

Risk of having traffic injury relative to passenger vehicle occupants

Rates of fatal and non-fatal traffic injury by travel mode; demographic

de Geus et al. 2012 (45) Belgium

Prospective 1 year follow-up

Fatal injuries: Fatality Analysis Reporting System (FARS; national census); Nonfatal injuries; General Estimates System (GES; police-reported crashes); GES and FARS contain only events with a motor vehicle involved Self-reported accidents; Minor accidents

Online Travel Diary

1087 participants; >18 years With job. 32% women 68% men

Trips; time; distance

Incidence: N injuries during 1-year follow-up period; IR: N per (i) 1000 trips; (ii) 1000 h; and (iii) 1000 km of exposure

Bicycle accident analysis; built environment and weather

Hoffman et al. 2010 (46) Portland, USA

Prospective 1yr follow-up

Self-reported accidents; Traumatic and serious traumatic events

Follow up monthly through online survey sent via e-mail

962 participants; 22–70 years 48% women 52% men

Distance

IR: incidence of traumatic events and serious traumatic events/100,000 miles

Accident characteristics; demographic and behavioral factors

Johnson et al. 2010 (47) Melbourne, Australia

Prospective 4 weeks; 12 h

Detected collisions, near-collisions and incidents through analysis of the video

Video camera attached to participants bicycle helmet

13 participants; >18 years 15% women 85% men

Time

Time of footage analyzed

Sayed et al. 2013 (48) Burrard Bridge ramps at pacific street, Canada

Prospective

Automated camera-based traffic conflicts; Severity was ranked using the Time-To-Collision safety indicator

Automated camera-based traffic count

All ages

N passages and N conflicts

Conflicts/time; conflicts by exposure (defined as the square root of the product of vehicle volumes and conflicts)

Identifying risk factors for collisions involving cyclists and drivers; behavioral Evaluation of automated safety diagnosis for vehicle-bicycle conflicts; built environment

Risk of fatal injury for motorcyclists, bicyclists and pedestrians relative to passenger vehicle occupants was 58.3; 2.3; 1.5 respectively; Higher fatality rates for male than female bicyclists; higher fatality rates for adolescents and adult than children (5–14 years) Accident causation: badly maintained bicycle infrastructure; Inattentive bicycle-car interaction; Underreporting of minor bicycle accidents No effect of safety clothing, helmet use, demography or experience on bicycle accidents: authors concluded that focus should be on the environment Driver awareness of on-road cyclists and early indication when turning may benefit the cyclists

High exposure of cyclists to traffic conflicts and a significant driver noncompliance rate

ER: event rate; IR: incidence rate or injury rate; RIR: relative injury rate; RR: relative risk; OR: odds ratio; N: number (incidence); km: kilometer; ‘/’: divided by; MAIS: maximum abbreviated injury score; PTW: powered two wheeler.

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

Author and location

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

15

overcomes the issue of using only hospital or police data. However, due to the design and small cohort size inherent to follow-up studies, severe accidents are unlikely to be included.

to incidents/1000 km on a bicycle lane, one can make a statement about the safety of one situation relative to the other. This outcome is called the relative risk (RR). Consequently, we define:

3.2. Prospective and retrospective study design

IR =

Prospective study designs are used to evaluate outcomes such as the incidence of bicycle accidents and/or exposure during the study period in relation to other factors such as the built environment or weather conditions. Prospective studies usually involve taking a cohort of subjects and monitoring them over a longer period (“longitudinally”). Prospective investigation is required to make precise estimates of either the incidence of an outcome or the relative risk of an outcome based on exposure. A retrospective study looks at past events and examines exposures to suspect risk or protection factors in relation to an outcome that is defined at the start of the study. In the case of bicycle safety research, we could state that the accident and exposure already occurred at the moment of the study. 3.3. The need to include exposure The need to include exposure in bicycle safety research is illustrated with two examples. Aultman-Hall and Hall (1998) found 139 events on road, 43 events off-road, and 7 events on a sidewalk. Without taking exposure into account, one would think that cycling on the road is the most risky, since there are the most events registered. However, after taking exposure (distance) into account, cycling on the road seems to be the safest and had the lowest events per distance traveled. A study by de Geus et al. (2012) recorded 8 accidents in the Wallonia region, and 34 in the Flanders region in Belgium. Based on this, one might conclude that cycling in Wallonia is safer than in Flanders. Taking exposure into account gave an IR of 0.037/1000 km for Flanders and 0.05/1000 km for Wallonia. These studies support the inclusion of exposure and possible confounding factors in studies of cycling accidents. 3.4. Incidence, incidence rate, relative risk and other variations There was no consensus found on the designation of incidence rate (IR) and relative risk (RR) in the published literature. Therefore we identified the definition of IR and RR that was used by each author (Table 1, 7th column). Based on the most frequent and clear definitions used in the literature, we defined IR as the ratio between incidences (absolute numbers) and exposure. IR can thus serve as a unit of measurement in the assessment of road safety. An incident can be defined as a collision, a single bicycle accident or a near collision (a conflict). A collision is defined as a bicycle accident where there has been contact between a bicyclist and another road user. A single bicycle accident is an accident where no other road users are involved. A near collision is defined as a situation where at least one of the road users has to make a maneuver to avoid an imminent collision, but where no impact occurred. A near collision could be an important indication of a potentially hazardous situation. Exposure can be expressed in different measuring units (distance, hours, passages or trips) and consequently so can IR. The most frequently used IR is based on the distance traveled or time spent cycling and is expressed in “occurrence of incidents per km driven” or “occurrence of incidents per hour on the road”, respectively. A third method to express IR uses a fixed point on the road and counts the incidents per number of passages for each vehicle type. Finally, there is the category of bicycle accident analysis per trip or per bicycle to be considered. This category uses accident databases and expresses them relative to bicycle trips estimates or bicycle volume estimates. By comparing the IR between two situations, e.g. incidents/1000 km bicycling on the road compared

RR =

#incidences exposure IRa IRb

4. Results 4.1. Different units of exposure For each paper included, we looked at the unit of exposure that was used. The main units of exposure used were distance (n = 12), followed by time (n = 11) and number of trips (n = 2). One study (Sayed et al., 2013) used automated traffic counts to evaluate the safety of an intersection. This method is ideal for evaluating the safety of a specific road segment (Schepers et al., 2011). Mindell et al. (2012) suggested the most appropriate way to compare transport modes with different average speeds is the use of RR based on time. For certain comparisons, this exposure unit may be sufficient. However, male and female cyclists do not ride at the same average speeds (de Geus et al., 2014). Therefore, in order to make a detailed comparison between and within transport modes, Mindell et al. (2012) included distance and time in their measure of exposure. This is beneficial, since it will allow other researchers to compare their results. 4.2. Exposure Exposure data were collected using a retrospective approach (Mindell et al., 2012; Aultman-Hall and Hall, 1998; Aultman-Hall and Kaltenecker, 1999; Bacchieri et al., 2010; Blaizot et al., 2013; Lusk et al., 2011; Rodgers, 1995; Thornley et al., 2008; Tin Tin et al., 2010, 2011; Teschke et al., 2012; Twisk and Reurings, 2013; Vandenbulcke et al., 2009; Yiannakoulias et al., 2012) or a prospective approach (Poulos et al., 2015; de Geus et al., 2012; Hoffman et al., 2010; Johnson et al., 2010; Sayed et al., 2013). 4.2.1. Retrospective studies For the retrospective approach, a questionnaire asking a cyclist about transport behavior over the past 12 months is commonly used (Aultman-Hall and Hall, 1998; Aultman-Hall and Kaltenecker, 1999; Bacchieri et al., 2010; Rodgers, 1995; Thornley et al., 2008; Tin Tin et al., 2010). These questionnaires are administered by telephone, e-mail or by pen and paper. Alternatively, retrospective studies may involve asking cyclists about their travel behavior on just one day, or their travel behavior on the day of the accident. These exposure data can be extrapolated to calculate average exposure. Three retrospective studies (Mindell et al., 2012; Blaizot et al., 2013; Tin Tin et al., 2011) included all ages and another (Tin Tin et al., 2010) included participants aged 5 years or older. These studies used a telephone questionnaire administered by a national institution to obtain information about traveling data. Usually, sample households were selected to be representative for all the households in the area studied. Household members were asked about their travel behavior on one week day, the day before the interview. These retrospective studies focused on differences in traffic safety between transport modes. To obtain a representative outcome for each age and gender-specific group, results were multiplied by the estimated population for that group. The use of a questionnaire asking the participants about their past travel behavior (e.g. over the previous 12 months) (Aultman-Hall and

16

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

Hall, 1998; Aultman-Hall and Kaltenecker, 1999; Bacchieri et al., 2010; Rodgers, 1995; Thornley et al., 2008) may offer a more accurate assessment of exposure since respondents only have to estimate their own travel behavior. Indeed, in the questionnaires administered by a national institution, sometimes one household member must estimate the travel behavior for all household members. However, estimating a year of travel, taking into account weekly and seasonal fluctuations, can lead to a high recall bias leading to possible over- or underestimations. None of our included studies demonstrated this recall bias. Bacchieri et al. found that 8.3% of subjects who completed a follow-up survey reported at least one accident (data from the author) and since 10.8% of the subjects reported at least one accident in the retrospective study, no recall bias was assumed. Additionally, the intellectual differences between children, adolescents, adults and the elderly require different questioning approaches. Studies that do not include children or adolescents (younger than 18 years), cannot be used as the sole basis to develop guidelines for a general infrastructural design since the study cohort is not a correct representation of the total cycling population. Moreover, it has been shown that traffic behavior, such as the decision to cross an intersection, differs between children and adults (Plumert et al., 2004; Hoffrage et al., 2003; Briem et al., 2004). Therefore, guidelines to improve bicycle safety should be based on studies that include the most appropriate cohort of cyclists. 4.2.2. Prospective studies Three prospective studies used a questionnaire together with a follow-up program (Poulos et al., 2015; de Geus et al., 2012; Hoffman et al., 2010), one study (Sayed et al., 2013) used automatic traffic counts and another (Johnson et al., 2010) followed 13 participants equipped with a camera over 4 weeks. The four prospective studies (Poulos et al., 2015; de Geus et al., 2012; Hoffman et al., 2010; Sayed et al., 2013) included only cyclists aged 18 years or older. Their follow-up method allowed them to accurately assess bicycle exposure (distance cycled, and time). They all concluded that infrastructure and environment have important implications on bicycle safety. The follow-up design has smaller cohorts but allows the researcher to measure accurately the exposure of a specific population, and allows self-reported accidents to be included. de Geus et al. (2012), Hoffman et al. (2010) and Poulos et al. (2015) used a 1-year follow-up design with 1087, 962 and 2038 participants respectively. There might be a loss to follow-up in a prospective cohort study which highlights the importance of recruiting enough participants at the start of a prospective followup study (Tin Tin et al., 2014). Due to the prospective design, severe bicycle accidents are less likely to be observed during the follow-up period. Prospective studies that did register severe bicycle accidents could not go into further detail because of the limited sample size. One study (de Geus et al., 2012) registered two major bicycle accidents, defined by hospitalization of more than 24 h, and 68 minor accidents. Another study (Hoffman et al., 2010) registered 50 serious traumatic events (5% of all cases), defined as a traumatic event for which any sort of medical attention was sought, without any other specification about the severity or hospitalization. In both studies accidents could not be compared, or interpreted as severe or not. A uniform code to indicate the severity of an accident such as the Maximum Abbreviated Injury Scale (MAIS) should be included in such studies. Taking this into account, there was no prospective study that successfully analyzed severe bicycle accidents. Thus, prospective study designs do not to be appropriate to analyze severe accidents. To summarize, both methodological approaches have their strengths and weaknesses. In prospective study designs, exposure can accurately be recorded as the subjects are closely followed. Additionally, bicycle accidents that are not officially recorded can

be included. The disadvantage of using a prospective design is that they are time-consuming, expensive, and data on severe accidents are limited. The prospective study is ideal for detailed analysis of bicycle safety factors. Retrospective data collection has the advantage of being easier and less costly to administer. It also allows a larger scale of data collection. There are several weaknesses, including selection and recall bias, risking suboptimal accident self-reporting. This method is ideal for studying bicycle safety factors on a large scale. Of course, combining a retrospective study with a prospective one, by asking cyclists for accident and travel history before a follow-up study to asses exposure and then correlating with official accident data, gives the best and most accurate outcome. 4.3. Factors influencing bicycle accidents All the included papers reported basic demographic data such as gender and age distribution. It is important to include this, since bicycling safety guidelines need to be adapted to their users. Seven studies (Bacchieri et al., 2010; Teschke et al., 2012; Vandenbulcke et al., 2009; de Geus et al., 2012; Hoffman et al., 2010; Johnson et al., 2010; Poulos et al., 2012) had a study population aged between 18 years and 70 years of age. Despite children, adolescents and the elderly (65+ years) incurring more accidents than adult populations (between 18 and 65 years) (Martensen, 2014), these age categories are underrepresented in the literature. Studies that did analyze bicycle safety of these age categories used national surveys, with data weighted for all ages. Other studies that included broader age categories selected a cohort where only a minority of the subjects were younger than 18, or older than 65. These studies could not analyze bicycle safety among these age categories due to high standard deviation. No studies were found specifically focusing on bicyclists younger than 18 years, or older than 65 years. The second factor investigated was the built environment. Infrastructure was only described in studies analyzing bicycle safety in terms of different built environments (Aultman-Hall and Hall, 1998; Aultman-Hall and Kaltenecker, 1999; Lusk et al., 2011; Teschke et al., 2012; Vandenbulcke et al., 2009; Yiannakoulias et al., 2012; de Geus et al., 2012; Sayed et al., 2013). However, logic suggests it is a confounding factor along with age, gender or even weather. Cyclists adapt their route choice for the weather conditions, or children might prefer different roads to adults. We believe that infrastructure characteristics should be described even when the study’s main objective focuses on another factor. This information helps policy makers or designers of infrastructure decide whether or not certain suggestions could be applied in different regions or countries. The third factor investigated included the weather (de Geus et al., 2012), the season (Hoffman et al., 2010) and lighting conditions (Rodgers, 1995; Twisk and Reurings, 2013). The final factor investigated was behavior. This included aspects such as helmet use, clothing, and use of lights when riding in the dark (Mindell et al., 2012; Bacchieri et al., 2010; Blaizot et al., 2013; Thornley et al., 2008; Tin Tin et al., 2010; Hoffman et al., 2010; Johnson et al., 2010; Poulos et al., 2012). These factors were assessed through questionnaires, often those that were used to assess exposure. 4.4. Bicycle safety results 4.4.1. Demographic parameters The studies including gender analysis were contradictory. Among these, there were more male bicycle commuters and they had a higher IR than women. One exception to this was in the study of Poulos et al. which was prospective. However, after controlling

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

for confounding factors, the authors found a higher IR for women than men. All of the relevant studies showed similar results regarding age. Studies looking at data from people older than 18 years, found that, for cyclists over 50 years old, IR increased with age, especially in single bicycle accidents (Rodgers, 1995; Tin Tin et al., 2010). However, these studies used hospital information for their accident data and therefore focused on major bicycle accidents. It is possible that the severity of accidents among older cyclists increases with age, but not the IR. A single bicycle accident of a 20-year-old cyclist may have no severe consequences, whereas the same type of fall is more likely to have more severe consequences in a 50-year-old with a higher risk of osteoporosis, for example. Mindell et al. (2012) looked at all ages, and they found higher fatality rates and hospital admissions per million hours of cycling for male cyclists younger than 17 and older than 50, with an incremental increase for those over 70 years of age. For female cyclists, the fatality rates and hospital admissions were lower compared to men except when aged between 17 and 20 years. Studies including a younger population (<18 years) (Mindell et al., 2012; Tin Tin et al., 2010, 2011; Twisk and Reurings, 2013) used national surveys, with age weighted according to the precise study area. For Mindell et al. (Davison et al., 2008) data were only available for a few young cyclists (under 20 years old), especially women, resulting in large uncertainties for this age category. 4.5. Built environment Three prospective studies indicated the impact of infrastructure on bicycle safety (de Geus et al., 2012; Hoffman et al., 2010; Johnson et al., 2010), especially for minor accidents. de Geus et al. (2012) reported that just having good bicycle infrastructure was not enough to ensure the safety of bicyclists, especially if it was not well maintained. Hoffman et al. (2010) concluded that there is an important role for dedicated infrastructure. They found more traumatic events on residential streets and on bike lanes but unfortunately, they did not assess the exposure for different types of infrastructure and therefore the IR for different types of infrastructure cannot be compared. Johnson et al. (2010) found that cyclists generally have high situational awareness on the roads, and car drivers were deemed at fault in the majority of collisions or nearcollisions analyzed. They also concluded that drivers are not aware of the cyclists traveling alongside or behind them. Together, these findings would support a combined approach of increasing driver awareness of cyclists on the road and building appropriate bicycle infrastructure to improve bicycle safety. Indeed, safety campaigns and training oriented at drivers may reduce cyclist trauma. Sayed et al. (2013) used a third approach. They identified dangerous locations at a major intersection. This was the only prospective study identified that offers insight that could be incorporated into guidelines for infrastructure engineers and designers to improve bicycle safety. Poulos et al. did take built environment into account in their study of infrastructure, but did not have enough data to compare IR across different infrastructure types. Seven retrospective studies also examined different aspects of infrastructure and bicycle safety (Lusk et al., 2011; Tin Tin et al., 2011; Teschke et al., 2012; Vandenbulcke et al., 2009; Yiannakoulias et al., 2012). Aultman-Hall and Hall (1998), AultmanHall and Kaltenecker (1999) focused broadly on infrastructure. Using an adult cohort (>18 years), they analyzed the IR for cycling on-road, off-road and on a cycle path. The highest IR was found on sidewalk cycling, and the lowest IR on-road. The IR per km was 26–68 times higher for bicycles compared to automobiles. It has been suggested that cyclists fare best when they behave as, and are treated as, operators of vehicles (Forester, 2008). These assumptions are based on absolute accident numbers, that gave a higher IR

17

for off-road cycling than for on-road (Aultman-Hall and Hall, 1998; Aultman-Hall and Kaltenecker, 1999), suggesting that shared-use paths and two-way paths should be preferred. However, Lusk et al. (2011) analyzed the IR of two-way cycle tracks and found lower IR for cycling on a two-way cycle track compared to cycling on the street, concluding that the construction of specific infrastructure for bicycles should not be discouraged. These findings are in line with the results of Teschke et al. (2012), who found a lower IR on quiet streets and on dedicated infrastructure. The RR for two-way cycle tracks was lower or similar to cycling on the street (Lusk et al., 2011). Higher IRs are found on major streets with parked cars, rail tracks (including trams) and at construction sites. Given the study designs of these papers and their findings, we conclude that retrospective study designs are the most appropriate for assessing bicycle safety and infrastructure. A detailed analysis of a specific situation may require a prospective design, such as the study reported by Sayed et al. (2013). 4.5.1. Weather The study of de Geus et al. (2012) gave an IR of 0.099 (95% CI 0.053–0.145) for cycling during weeks when the roads were snowy or icy and an IR of 0.048 (0.036–0.060) for weeks without snow or icy roads. In the study of Hoffman et al. (2010), December and January were the months with the highest IR. 4.5.2. Behavior Wearing visible clothing, having more bicycling experience, or wearing a helmet were not shown to reduce the RR of being involved in an accident (Bacchieri et al., 2010; Hoffman et al., 2010). However, several studies showed that wearing protective clothing reduces accident severity, presumably in collision situations (Persaud et al., 2012; Wood et al., 2012; Amoros et al., 2012; Berg and Westerling, 2007). In order to reduce the RR of bicycle accidents, a greater range of guidelines and measures are needed. 5. Discussion Both prospective and retrospective study designs have their strengths and weaknesses. In prospective study designs, exposure can accurately be recorded as the subjects are closely followed. Additionally, bicycle accidents that are not officially reported (e.g. by police or hospitals) can be included. However, prospective studies are time consuming, expensive, collect a limited volume of data and are probably less generalizable than retrospective studies. Prospective studies are useful for analyzing bicycle safety parameters in detail and usually have fewer potential sources of bias than retrospective studies. One potential source of bias is the loss to follow-up but Tin Tin et al. (2014) did not find any substantial bias of seven association estimates from primary research. Unlike the prospective cohort design, retrospective studies can provide an immediate answer. Retrospective data collection is easier and less costly than prospective data collection. Retrospective study designs also allow accident data to be collected on a larger scale. Potential drawbacks of the retrospective study design are selection and recall bias, resulting in fewer details about the accident being remembered. Combining retrospective and prospective study designs would thus provide the most accurate safety data. 6. Conclusions To guarantee that findings are consistent, the investigated parameters need to be controlled for exposure and confounding factors. Despite the importance of accident exposure for bicycle safety research, 98% of the studies included from our search on bicycle safety or accidents did not collect exposure. Additionally,

18

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19

most studies measuring exposure had a retrospective design with an inherent recall bias, resulting in an underreporting of minor bicycle accidents. These retrospective studies indicated a higher IR for men compared to women, and an increased RR of injury for cyclists aged 50 years or older. Well-maintained bicycle infrastructure, in terms of dedicated or shared paths and intersections, improves bicycle safety. For evaluating the safety of new infrastructure, large-scale traffic counts for bicyclist should be carried out, before and after the construction phase. When analyzing bicycle safety in a more general context, time spent traveling or distance traveled with a bicycle are commonly used parameters. However, a more detailed description of injury severity is required. A prospective design including a follow-up method, will collect the most accurate data in terms of exposure and the characteristics of accidents. The recall bias is minimal, and minor bicycle accidents are also included. However, the limited accident data available in a prospective follow-up design makes it impossible to analyze the effects of infrastructure in any detail. When combining exposure data from a follow-up study, extrapolating the exposure data for a larger population and matching it with hospital and/or police data, a more complete analysis can be achieved. We identified only four studies using a prospective design, none of which included cyclists under 18 years old. These studies concluded that infrastructure and built environment are important factors influencing the outcome of bicycle accidents. But they are not able to offer details about specific infrastructure due to the limited accident data collected, and its diversity. Although the bicycle safety of adolescents and the elderly is of major concern in many countries, few studies have provided any safety guidelines for these age categories. This may be due to the difficulty of administering questionnaires or conducting followup studies for these age groups. Nonetheless, far more research focusing on these populations is needed. Lastly, several studies illustrated the importance of road infrastructure for bicycle safety. However, there is very limited research on bicycle-specific infrastructure that relates quality and presence to accident exposure. Future research should try to combine retrospective and prospective designs and use matched cohorts. Acknowledgements Support for this study was provided by the grant of an FWOLevenslijn project (G0C7113N). The Authors want to thank Simon Batterbury from the University of Melbourne (Australia) and Juliette Gray for the profound editing of the paper. References Amoros, E., Chiron, M., Martin, J.L., Thelot, B., Laumon, B., 2012. Bicycle helmet wearing and the risk of head, face, and neck injury: a French case–control study based on a road trauma registry. Inj. Prev. 18 (1), 27–32. Andersen, L.B., Schnohr, P., Schroll, M., Hein, H.O., 2000. All-cause mortality associated with physical activity during leisure time, work, sports, and cycling to work. Arch. Intern. Med. 160 (11), 1621–1628. Aultman-Hall, L., Hall, F.L., 1998. Ottawa-Carleton commuter cyclist on- and off-road incident rates. Accid. Anal. Prev. 30 (1), 29–43. Aultman-Hall, L., Kaltenecker, M.G., 1999. Toronto bicycle commuter safety rates. Accid. Anal. Prev. 31 (6), 675–686. Bacchieri, G., Barros, A.J., Dos Santos, J.V., Gigante, D.P., 2010. Cycling to work in Brazil: users profile, risk behaviors, and traffic accident occurrence. Accid. Anal. Prev. 42 (4), 1025–1030. Beck, L.F., Dellinger, A.M., O’Neil, M.E., 2007. Motor vehicle crash injury rates by mode of travel, United States: using exposure-based methods to quantify differences. Am. J. Epidemiol. 166 (2), 212–218. Berg, P., Westerling, R., 2007. A decrease in both mild and severe bicycle-related head injuries in helmet wearing ages – trend analyses in Sweden. Health Promot. Int. 22 (3), 191–197. Blaizot, S., Papon, F., Haddak, M.M., Amoros, E., 2013. Injury incidence rates of cyclists compared to pedestrians, car occupants and powered two-wheeler

riders, using a medical registry and mobility data, Rhone County, France. Accid. Anal. Prev. 58, 35–45. Boufous, S., De Rome, L., Senserrick, T., Ivers, R., 2011. Cycling crashes in children, adolescents, and adults – a comparative analysis. Traffic Inj. Prev. 12 (3), 244–250. Boufous, S., de Rome, L., Senserrick, T., Ivers, R.Q., 2013. Single-versus multi-vehicle bicycle road crashes in Victoria, Australia. Inj. Prev. 19 (5), 358–362. Briem, V., Radeborg, K., Salo, I., Bengtsson, H., 2004. Developmental aspects of children’s behavior and safety while cycling. J. Pediatr. Psychol. 29 (5), 369–377. Christie, N., Cairns, S., Towner, E., Ward, H., 2007. How exposure information can enhance our understanding of child traffic “death leagues”. Inj. Prev. 13 (2), 125–129. Daniels, S., Brijs, T., Nuyts, E., Wets, G., 2009. Injury crashes with bicyclists at roundabouts: influence of some location characteristics and the design of cycle facilities. J. Saf. Res. 40 (2), 141–148. Davis, R.M., Pless, B., 2001. BMJ bans “accidents” – accidents are not unpredictable. Br. Med. J. 322 (7298), 1320–1321. Davison, K.K., J.L., W., Lawson, C.T., 2008. Children’s active commuting to school: current knowledge and future directions. Prev. Chronic Dis. 5 (3), 1–11. de Geus, B., Van Hoof, E., Aerts, I., Meeusen, R., 2008. Cycling to work: influence on indexes of health in untrained men and women in Flanders. Coronary heart disease and quality of life. Scand. J. Med. Sci. Sports 18 (4), 498–510. de Geus, B., Joncheere, J., Meeusen, R., 2009. Commuter cycling: effect on physical performance in untrained men and women in Flanders: minimum dose to improve indexes of fitness. Scand. J. Med. Sci. Sports 19 (2), 179–187. de Geus, B., Vandenbulcke, G., Int Panis, L., Thomas, I., Degraeuwe, B., Cumps, E., et al., 2012. A prospective cohort study on minor accidents involving commuter cyclists in Belgium. Accid. Anal. Prev. 45, 683–693. de Geus, B., Degraeuwe, B., Vandenbulcke, G., Int Panis, L., Thomas, I., Aertsens, J., et al., 2014. Utilitarian cycling in Belgium: a cross-sectional study in a sample of regular cyclists. J. Phys. Act. Health 11 (5), 884–894. Dhillon, P.K., 2001. Assessment of hospital and police ascertainment of automobile versus childhood pedestrian and bicyclist collisions. Accid. Anal. Prev. 33, 9. Dumbaugh, E., Li, W., Joh, K. (Eds.), 2012. The Built Environment and the Incidence of Pedestrian and Cyclist Crashes. Ewing, R., Dumbaugh, E., 2009. The built environment and traffic safety. A review of empirical evidence. J. Plan. Lit. 23 (4), 347–367. Forester, J., 2008. Effective Cycling Instructors Manual. Ghekiere, A., Van Cauwenberg, J., de Geus, B., Clarys, P., Cardon, G., Salmon, J., et al., 2014. Critical environmental factors for transportation cycling in children: a qualitative study using bike-along interviews. PLOS ONE 9 (9), e106696. Harris, M.A., Reynolds, C.C., Winters, M., Cripton, P.A., Shen, H., Chipman, M.L., et al., 2013. Comparing the effects of infrastructure on bicycling injury at intersections and non-intersections using a case-crossover design. Inj. Prev. 19 (5), 303–310. Higgins, P.A.T., 2005. Exercise-based transportation reduces oil dependence, carbon emissions and obesity. Environ. Conserv. 32 (3), 197–202. Hoffman, M.R., Lambert, W.E., Peck, E.G., Mayberry, J.C., 2010. Bicycle commuter injury prevention: it is time to focus on the environment. J. Trauma 69 (5), 1112–1117, discussion 7-9. Hoffrage, U., Weber, A., Hertwig, R., Chase, V.M., 2003. How to keep children safe in traffic: find the daredevils early. J. Exp. Psychol.-Appl. 9 (4), 249–260. Howarth, C.I., 1982. The need for regular monitoring of the exposure of pedestrians and cyclists to traffic. Accid. Anal. Prev. 14 (5), 341–344. Johnson, M., Charlton, J., Oxley, J., Newstead, S., 2010. Naturalistic cycling study: identifying risk factors for on-road commuter cyclists. Ann. Adv. Automot. Med. 54, 275–283. Keller, J., Abdel-Aty, M., Brady, P.A., 2006. Type of collision and crash data evaluation at signalized intersections. ITE J. 76 (2), 30–39. Lusk, A.C., Furth, P.G., Morency, P., Miranda-Moreno, L.F., Willett, W.C., Dennerlein, J.T., 2011. Risk of injury for bicycling on cycle tracks versus in the street. Inj. Prev. 17 (2), 131–135. Maring, W., van Schagen, I., 1990. Age dependence of attitudes and knowledge in cyclists. Accid. Anal. Prev. 22 (2), 127–136. Martensen, H., 2014. @ RISK: Analyse van het risico op ernstige en dodelijke verwondingen in het verkeer in functie van leeftijd en verplaatsingswijze. Belgisch Instituut voor de Verkeersveiligheid, Brussel, België. Kenniscentrum Verkeersveiligheid. Martinez-Ruiz, V., Jimenez-Mejias, E., Luna-del-Castillo Jde, D., Garcia-Martin, M., Jimenez-Moleon, J.J., Lardelli-Claret, P., 2014. Association of cyclists’ age and sex with risk of involvement in a crash before and after adjustment for cycling exposure. Accid. Anal. Prev. 62, 259–267. Meuleners, L.B., Lee, A.H., Haworth, C., 2007. Road environment, crash type and hospitalisation of bicyclists and motorcyclists presented to emergency departments in Western Australia. Accid. Anal. Prev. 39 (6), 1222–1225. Mindell, J.S., Leslie, D., Wardlaw, M., 2012. Exposure-based ‘like-for-like’ assessment of road safety by travel mode using routine health data. PLOS ONE 7 (12). Mueller, N., Rojas-Rueda, D., Cole-Hunter, T., de Nazelle, A., Dons, E., Gerike, R., et al., 2015. Health impact assessment of active transportation: a systematic review. Prev. Med. 76, 103–114. Oja, P., Titze, S., Bauman, A., de Geus, B., Krenn, P., Reger-Nash, B., et al., 2011. Health benefits of cycling: a systematic review. Scand. J. Med. Sci. Sports 21 (4), 496–509.

J. Vanparijs et al. / Accident Analysis and Prevention 84 (2015) 9–19 Panter, J.R., Jones, A.P., van Sluijs, E.M., Griffin, S.J., 2010. Attitudes, social support and environmental perceptions as predictors of active commuting behaviour in school children. J. Epidemiol. Community Health 64 (1), 41–48. Persaud, N., Coleman, E., Zwolakowski, D., Lauwers, B., Cass, D., 2012. Nonuse of bicycle helmets and risk of fatal head injury: a proportional mortality, case–control study. Can. Med. Assoc. J. 184 (17), E921-E3. Plumert, J.M., Kearney, J.K., Cremer, J.F., 2004. Children’s perception of gap affordances: bicycling across traffic-filled intersections in an immersive virtual environment. Child Dev. 75 (4), 1243–1253. Poulos, R.G., Hatfield, J., Rissel, C., Grzebieta, R., McIntosh, A.S., 2012. Exposure-based cycling crash, near miss and injury rates: the Safer Cycling Prospective Cohort Study protocol. Inj. Prev. 18 (1), e1. Poulos, R.G., Hatfield, J., Rissel, C., Flack, L.K., Murphy, S., Grzebieta, R., et al., 2015. An exposure based study of crash and injury rates in a cohort of transport and recreational cyclists in New South Wales, Australia. Accid. Anal. Prev. 78, 29–38. Roberts, I., Carlin, J., Bennett, C., Bergstrom, E., Guyer, B., Nolan, T., et al., 1997. An international study of the exposure of children to traffic. Inj. Prev. 3 (2), 89–93. Rodgers, G.B., 1995. Bicyclist deaths and fatality risk patterns. Accid. Anal. Prev. 27 (2), 215–223. Sayed, T., Zaki, M.H., Autey, J., 2013. Automated safety diagnosis of vehicle–bicycle interactions using computer vision analysis. Saf. Sci. 59, 163–172. Schepers, J.P., Kroeze, P.A., Sweers, W., Wust, J.C., 2011. Road factors and bicycle-motor vehicle crashes at unsignalized priority intersections. Accid. Anal. Prev. 43 (3), 853–861. Stevenson, M., 2014. Epidemiology and transport: good science is paramount. Inj. Prev. Teschke, K., Harris, M.A., Reynolds, C.C., Winters, M., Babul, S., Chipman, M., et al., 2012. Route infrastructure and the risk of injuries to bicyclists: a case-crossover study. Am. J. Public Health 102 (12), 2336–2343.

19

Thornley, S.J., Woodward, A., Langley, J.D., Ameratunga, S.N., Rodgers, A., 2008. Conspicuity and bicycle crashes: preliminary findings of the Taupo Bicycle Study. Inj. Prev. 14 (1), 11–18. Timperio, A., Crawford, D., Telford, A., Salmon, J., 2004. Perceptions about the local neighborhood and walking and cycling among children. Prev. Med. 38 (1), 39–47. Tin Tin, S., Woodward, A., Ameratunga, S., 2010. Injuries to pedal cyclists on New Zealand roads, 1988–2007. BMC Public Health 10, 655. Tin Tin, S., Woodward, A., Thornley, S., Ameratunga, S., 2011. Regional variations in pedal cyclist injuries in New Zealand: safety in numbers or risk in scarcity? Aust. N. Z. J. Public Health 35 (4), 357–363. Tin Tin, S., Woodward, A., Ameratunga, S., 2014. Estimating bias from loss to follow-up in a prospective cohort study of bicycle crash injuries. Inj. Prev. 20 (5), 322–329. Twisk, D.A.M., Reurings, M., 2013. An epidemiological study of the risk of cycling in the dark: the role of visual perception, conspicuity and alcohol use. Accid. Anal. Prev. 60, 134–140. Vandenbulcke, G., Thomas, I., de Geus, B., Degraeuwe, B., Torfs, R., Meeusen, R., et al., 2009. Mapping bicycle use and the risk of accidents for commuters who cycle to work in Belgium. Transp. Policy 16 (2), 77–87. Vandenbulcke, G., Thomas, I., Int Panis, L., 2014. Predicting cycling accident risk in Brussels: a spatial case–control approach. Accid. Anal. Prev. 62, 341–357. Wood, J.M., Tyrrell, R.A., Marszalek, R., Lacherez, P., Carberry, T., Chu, B.S., 2012. Using reflective clothing to enhance the conspicuity of bicyclists at night. Accid. Anal. Prev. 45, 726–730. Yiannakoulias, N., Bennet, S.A., Scott, D.M., 2012. Mapping commuter cycling risk in urban areas. Accid. Anal. Prev. 45, 164–172.