Background The child active transportation literature has focused on walking, with little attention to risk associated with increased traffic exposure. This paper reviews the literature related to built environment correlates of walking and pedestrian injury in children together, to broaden the current conceptualization of walkability to include injury prevention.
Methods Two independent searches were conducted focused on walking in children and child pedestrian injury within nine electronic databases until March, 2012. Studies were included which: 1) were quantitative 2) set in motorized countries 3) were either urban or suburban 4) investigated specific built environment risk factors 5) had outcomes of either walking in children and/or child pedestrian roadway collisions (ages 0-12). Built environment features were categorized according to those related to density, land use diversity or roadway design. Results were cross-tabulated to identify how built environment features associate with walking and injury.
Results Fifty walking and 35 child pedestrian injury studies were identified. Only traffic calming and presence of playgrounds/recreation areas were consistently associated with more walking and less pedestrian injury. Several built environment features were associated with more walking, but with increased injury. Many features had inconsistent results or had not been investigated for either outcome.
Conclusions The findings emphasise the importance of incorporating safety into the conversation about creating more walkable cities.
Statistics from Altmetric.com
Child pedestrian injuries are a leading cause of injury-related death for Canadian children younger than 14 years.1 In children of ages 5–9 years, pedestrian collisions are tied with motor vehicle collisions as the primary cause of unintentional injury death (18%).1 Every year, approximately 56 child pedestrians die and 780 are hospitalised with serious injuries in Canada.1 While the burden of child pedestrian injuries and fatalities is high, there has been a decline of over 50% in Canadian hospitalisation and deaths from 1994 to 2003.1 ,2 Declining trends are also evident in the USA, Europe and New Zealand.3–7 This reduction may not be due to safer traffic environments, but rather because children are walking less often, thus reducing the exposure to risk of injury from collision with a motor vehicle.4 ,6 ,8 While children's walking trips to all destinations have fallen, this decline is most apparent for school trips. From 1986 to 2006, active school transportation (ie, walking, biking) declined from 53% to 43% in Canadian children of ages 11–13 years.9 Declines have also been noted in the USA, Great Britain and Australia.8 ,10 ,11
Initiatives to increase walking in children have been developed to promote healthy active living and are focused primarily on school trips.12–14 The Safe Routes to School (SRTS) concept began in Denmark in the 1970s, with programmes developed in Europe, Australia, New Zealand, Canada and the USA.15 In the USA, a national SRTS program was passed in 2005 as part of the US federal surface transportation bill with 11 000 schools funded by 2011.13 ,15 ,16 In Canada, SRTS programs are conducted at a grassroots/activist level with some pilot funding from a provincial government agency.17
When planning interventions to increase walking to school, the potential effects of increased walking exposure on pedestrian injury rates should be considered. Gropp et al18 recently found a dose–response relationship between longer school travel distances and injury related to active school transportation in a Canadian national survey. In Toronto, Canada, almost 50% of child pedestrian collisions were found to occur during school transportation times and the highest density of collisions occurred within 150 m of a school.19
Since the 1970s, research from the transportation and urban planning fields has investigated the ‘walkability’ of the environment. More recently, public health researchers have become interested in the effects of the built environment and walkability on physical activity and obesity. The definition of walkability is problematic as it varies by discipline and there is no standard set of factors describing a walkable environment. Walkability has been defined as ‘the extent to which the built environment supports and encourages walking by providing for pedestrian comfort and safety…’.20 This conceptual recognition of safety in walking has not been well addressed in the built environment active travel literature. Focus has been on what can be done to increase walking with little attention paid to the risks associated with increased traffic exposure. Others have also acknowledged the importance of linking road safety indicators to active school commuting.21–23
Reviews related to walkability and children have investigated the correlates of walking, which encompass many characteristics of the household, behaviours and material and social environments.16 ,22–26 There are few systematic reviews, however, that link features of the built environment to child pedestrian injury. Wazana et al27 found that risk factors for child pedestrian injury were related to the physical environment. Their review was limited to Medline articles from 1985 to 1995. Built environment roadway characteristics have been statistically linked with child pedestrian injury risk (OR=2.5); however, the effects of specific built environment features have not been examined.28
The purpose of this review is to use the published literature to develop an understanding of how specific features of the built environment relate to both walking in elementary school children and child pedestrian injury to direct further research. As this review incorporated a variety of papers drawn from a wide array of disciplines that use different reporting standards, traditional systematic review was challenging. However, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were adhered to as closely as possible.29
The search strategy was developed in consultation with a research librarian at the Hospital for Sick Children, Toronto, Canada. As the research question crossed many disciplines, nine electronic databases were searched until 1 March 2012: Medline (1980–2012), Embase (1980–2012), Transport (1980–2012), Dissertations and Theses (1980–2012), Web of Science (1980–2012), Scopus (2004–2012), PyscInfo (1980–2012), CINAHL (1985–2012) and SafetyLit (1995–2012). Search strategies were broad, given the variety of discipline-specific terminologies (see online supplementary appendix A). Two sets of searches were conducted on each database, one for child pedestrian injury and another for walking in children. Search results are illustrated using a PRISMA flow diagram (figure 1).29
Hand searches were done for references from systematic reviews and select journals between the years 2006 and 2011: Accident Analysis & Prevention, Injury Prevention, Traffic Injury Prevention, Transportation Research Part D: Transport and the Environment, and Health and Place. Google searches were used to identify articles/reports/grey literature from websites using the search terms: pedestrian, child pedestrian, built environment and pedestrian injury, walkability, active school transportation and active transportation.
Two reviewers from the research team independently reviewed titles and abstracts initially and then assessed full-text versions using standardised checklists. There were no language restrictions. Only literature from highly motorised countries (ie, Australia, Japan, New Zealand, North America and Western Europe) was considered. This classification of high motorised countries (HMCs) versus low motorised countries (LMCs) was developed by the Transport Research Laboratory and used by WHO to describe road fatality trends.30 ,31 Motorisation level is measured by the number of motor vehicles/1000 population and LMCs have typically much less motorisation levels (<400 Motor vehicles (MV)/1000 population) than HMCs.30 Data were not included from LMCs as the traffic environment is very different, with less developed traffic safety structures including traffic separation/calming, safety education and law enforcement.6 Inclusion and exclusion criteria were as follows.
Outcomes included measures of either one or both:
Child pedestrian roadway collisions (incidents or severity). Studies with samples composed of both child pedestrian and cyclist casualties where proportions of each were not indicated were included in the review if located in Australia/New Zealand/North America or Europe (excluding The Netherlands and Denmark). Nationally, the proportion of pedestrians is greater than cyclists in those locations except in the Netherlands and Denmark, where cycling rates are greater than walking rates32–34
Walking in children (or inversely, being driven)
Specifically identified built environment features (not summary measures, eg, ‘walkability’ or ‘perceived unsafe environment’ where traffic features were not specifically identified)
Full published papers, online reports or dissertations
Quantitative analysis with statistical significance testing (bivariate, multivariate) or where possible to calculate statistical significance
Highly motorised countries
≥50% sample composed of children ages 4–12 years or separate models for children
Specific disease or at-risk groups (eg, obesity)
Stratified outcomes (eg, comparing mid-block versus intersection collisions)
Unspecific outcomes (eg, physical activity not disaggregated by travel mode)
Discrepancies in eligibility were resolved through inter-reviewer discussion and in consultation with a third researcher as required.
Data extracted included; publication year, location, data source year(s), study design, authors’ disciplines, journal, population, ages, outcomes, exposures (irrespective of statistical significance), measurement techniques, statistical methods and covariate adjustment. Significant associations with the outcome were identified, where p≤0.05 for correlations/comparisons of means or where 95% CIs of ORs/relative risks excluded 1.0. Built environment variables were categorised into the 3 D groups, Density, Diversity and Design, using the framework described by Cervero and Kockelman.35 Density variables were related to population density, diversity variables reflected different land uses and design variables were related to roadway characteristics.35
Studies were identified which considered demographic and socioeconomic covariates by having restricted samples (eg, specific age group), stratified analysis, testing for inclusion in multivariate models or matching in case-control studies. Socioeconomic status (SES) covariates included: income, home ownership vehicle ownership, one parent families, employment, free school meals, crowding, public assistance, immigrant, education, dwelling size and overall deprivation indices. Other demographic covariates included: age, sex, race/ethnicity and household size.
Quality assessment was conducted using the Epidemiological Appraisal Instrument (EAI).36 The EAI builds on the Downs and Black checklist commonly used for epidemiological studies.37 ,38 The EAI revision improves the validity and reliability and adapts the checklist for different study designs. EAI has 43 items with item-specific instructions that specify applicability of each item to different study designs. Criterion and construct validity have been tested, with an inter-rater reliability of 90% and a weighted κ in the excellent range (0.80–1.00). No EAI summary scores were calculated; instead, the instrument was used as a descriptive assessment of study components according to guidelines in the PRISMA statement.29
Features with any statistically significant negative associations with incidence of child pedestrian injury or severity were categorised as ‘Less Injury (Safer)’ whereas those with statistically significant positive associations with injury were categorised ‘More Injury (Less Safe)’. Features with statistically significant positive associations with walking were categorised ‘More Walking’ and those with statistically significant negative associations were categorised ‘Less Walking’. Correlates were grouped together where appropriate (eg, playgrounds and parks) and were categorised into the 3 D categories. Those with inconsistent findings or had not been tested for either outcome were identified.
A total of 12 884 unique papers were originally screened, including 19 unpublished papers found through Google searches (figure 1). Of these, 12 256 papers were excluded, leaving 628 papers, 433 related to children walking and 195 related to child pedestrian injury. No studies were identified that addressed both correlates of children walking and pedestrian injury. Eligibility assessment of titles and abstracts produced 134 walking papers (plus 26 from reference lists) and 107 child pedestrian injury papers (plus six papers from reference lists). Full-text screening resulted in 50 walking and 35 pedestrian injury papers from which data were extracted. This represented 0.65% of the original articles identified. The average inter-rater agreement was κ=0.80 for pedestrian and κ=0.84 for walkability papers.
The walking papers were from a wide variety of disciplines, with 48 from scientific journals, one report and one meeting paper (see online supplementary appendix B). The papers were from: USA (22, 44%), Australia/New Zealand (12, 24%), Canada (6, 12%), UK (5, 10%), and one each from Germany, Holland, Norway, Sweden and Switzerland (1, 2%). No papers were published before 1996 that fulfilled inclusion criteria, and more than half had been published since 2008.
All studies were observational; 94% (47/50) were cross-sectional, one was longitudinal and two were time series studies (see online supplementary appendix B). The conceptualisation of the walking outcome varied (column 4, see online supplementary appendix B) with 47 (94%) of studies measuring walking as a prevalence proportion and the remaining three studies as a rate/week. In all, 42 studies (84%) measured walking to school, whereas the remaining eight measured walking in general or to leisure activities. Walking was measured in 29 (58%) papers via parent report (questionnaires, face-to-face interviews and telephone interviews), 9 (18%) via child report including questionnaires, face-to-face interviews, hand counts), 6 (12%) via both child/parent report, and 6 (12%) via travel diaries. Built environment correlates of walking were measured by routinely collected administrative data processed using Geographic Information System software (GIS, 23, 46%) and parent reports (22, 44%). Several studies used field surveys (4, 8%) or child/parent reports (4, 8%). Most studies (41/50=82%) accounted for both SES and other demographic covariates.
Child pedestrian injury
Child pedestrian injury papers were drawn primarily from health-related fields, and then transportation research/planning, civil engineering, urban planning and geography (see online supplementary appendix C). Most were published scientific papers (33, 94%) from: USA (12, 32%), Australia/New Zealand (7, 18.9%), Canada (9, 24%), the UK (7, 19%), Ireland (1, 3%) and Germany (1, 3%). Overall, 50% of the papers were published between 1990 and 1998. Only 7 (19%) were published since 2008.
All child pedestrian injury studies were observational, with the majority using retrospective data situated within discrete time intervals, referred to as a cross-sectional retrospective design (21/35, 60%, online supplementary appendix C). There were 10 case-control (29%), four time series (11%) and one case-crossover study (3%). Eighteen were ecological studies (51%). Injury was conceptualised either as an injury or fatality incident, or injury severity. Thirty-four of the cross-sectional studies measured injury rates per spatial unit or per population (92%), with the remaining two studies measuring incident cases. All case control, time series and case-crossover studies measured incident injury cases. Child pedestrian injury was measured using police-report databases (19, 54.1%), hospital surveillance (13, 35.1%), coroner surveillance (4, 10.8%), police surveillance, trauma databases and other databases (each 3, 8%). Built environment correlates of child pedestrian injury were measured using databases/GIS (26, 70.3%), field surveys (10, 27.0%), child/parent reports (3, 8.1%) and parent report (2, 5.4%). Overall, 13 of the 35 child pedestrian injury papers (37%) controlled for both SES and other demographic variables, whereas 14 controlled for one or the other (40%). Eight studies did not control for SES or demographic these variables (23%).
Over 85% of walking studies had clearly stated hypotheses/objectives, described outcomes and statistical methods, reported main findings, provided estimates of statistical parameters, and adequately adjusted for covariates/confounders (table 1). Less than 50% explicitly stated the study design; however, the design of 52% could easily be inferred (‘Partial’). Most studies lacked clear identification of exposure (84%) as typically these papers investigated a group of covariates.
Assessing external validity was problematic. In all, 68% of studies reported participation rates; participant characteristics were described/partially described in 70% and only 8% accounted for subject loss/unavailable records (table 1). No sample size calculations were evident in any of the studies. Only 12 (24%) provided exposure measurement reliability and 9 (18%) provided validity information. Only 8 (16%) provided information regarding reliability of outcome measures and 5 (10%) reported validity information.
Child pedestrian injury
Over 90% of cross-sectional/longitudinal/time series injury studies had clearly reported hypothesis/objectives, outcome population source/sampling frame, eligibility criteria, statistical methods used and main findings (table 1). Only 20% of studies had explicitly stated the study design; the design of 76% was implicit or stated in the abstract (‘Partial’). Most of these studies did not identify an exposure (68%). There were no sample size calculations or reported exposure or outcome reliability and validity. Only 8 (32%) studies performed adequate individual covariate adjustment, but 17 (68%) performed adequate environmental adjustment. Estimates of random variability were missing in over 60% of papers.
All case control/crossover studies had clearly described outcomes, population sources, eligibility criteria, participant characteristics, main findings, estimates of statistical parameters, and measured exposure and outcome similarly for cases and/or controls. In approximately 25% of papers, the hypothesis and objectives were unclear and lacked a clear statement of study design. Over 80% of papers provided estimates of random variability and statistical methods were described clearly with confounders accounted for. Details regarding participation rates were frequently lacking. Only 2 (18%) of studies described observer blinding, and only 55% described the timing of case/control recruitment. There were no reports of covariate/outcome reliability or validity.
Safety and walking
Less injury (safer) and walking correlates
Correlates with significant associations with both increased walking and decreased injury were design and diversity features, traffic calming (eg, roundabouts, speed humps) and proximity to/presence of playgrounds/recreation areas/parks/open space (figure 2). Indicators of overall traffic calming measures were generally examined with only two papers focused on specific features, roundabouts and speed humps.39–43 The positive association between recreation areas/playgrounds and safe walking stresses the importance of incorporating these land use features and facilities into neighbourhoods.
More injury (less safe) and walking correlates
Correlates associated with a less safe traffic environment and with increased walking were from each of the 3 Ds. Design features were: higher road density/length and numbers of crosswalks. Density features were: higher pedestrian volume and higher population density. Diversity features included: number of schools, land use mix and proximity of services (figure 2). These environmental features may be indicative of locations of greater exposure to traffic, where there are more child pedestrians/vehicles and higher traffic speeds. Features such as crosswalks may act as confounders related to increased exposure (as their presence may indicate more children walking) and/or increased child pedestrian injury outcomes (due to inadequate design/use). Marked crosswalks have been associated with an unadjusted twofold elevation of risk of child pedestrian injury.44
Inconsistent/untested correlates of injury and walking
Many correlates had either positive or negative associations with injury but had inconsistent results or were not tested for walking (results not shown). Traffic control mechanisms such as lights were protective against injury but the relationship with walking was inconsistent.12 ,43 ,45–47 Sidewalks were associated with increased child pedestrian injury but the relationship with walking was inconsistent. Sidewalks, similar to crosswalks, might be a confounder related to both increased exposure and to increased child pedestrian injury outcomes. Sidewalks were, however, associated with injury after controlling for vehicle speed and volume.44 ,48–50 Children may treat sidewalks as extended play areas, or they may be more cautious when walking along roads where sidewalks are absent.49 One-way streets and school crossing guards were associated with increased child pedestrian injury, but had not been tested for walking. Wazana et al found that child pedestrian injuries occurred 2.5 times more frequently per kilometre of one-way street than on two-way streets. More children walking might explain this effect, along with less driver attention or children not looking for traffic in the appropriate direction.51 Cloutier et al found the numbers of crossing guards were positively related to pedestrian risk around schools. This may be a result of crossing guards being placed in locations that are particularly dangerous for child pedestrians, with their safety effect not being enough to overcome the excess danger.52
Other factors such as street parking were related to more walking, but had an inconsistent relationship with injury. Street parking slows traffic down and provides a barrier between vehicles and pedestrians.53 ,54 However, street parking can also be a visual obstruction for pedestrian crossings and contributes to traffic congestion, which may increase potential for collisions.55 ,56 Other correlates of walking related to the design of the built environment which warrant further investigation into their association with injury are the presence of trails, perceptions of safety, cul-de-sacs and dead ends, public transit, and walking network connectivity.
Traffic calming devices and the presence of playgrounds and recreation areas were the only factors consistently associated with more walking and less injury. Higher pedestrian volume, population and road density, schools, urban location, land use mix, proximity to services/facilities and crosswalks were associated with more walking, but with less safety. The majority of built environment factors had inconsistent associations with either walking or injury, or had not been tested for either one of the outcomes. Quality assessment, using a valid and reliable tool, revealed noteworthy inconsistencies of method, analysis and reporting in both the walking and the child pedestrian injury literature.
Many of the built environment correlates associated with pedestrian injury, such as higher pedestrian volume, urban environment and schools, may not be inherently dangerous, but rather may be markers for increased exposure to traffic in general and higher speed traffic in certain environments. WHO has identified speed as the principal risk factor for pedestrian–motor vehicle collisions and fatality.31 Well-designed studies are needed to study the effects of interventions directed towards separating child pedestrians from high speed traffic by space (eg, playgrounds), time (traffic lights) and the optimal use of traffic calming measures to slow vehicles down in areas where there are many child pedestrians.
Several potential explanations exist for inconsistent findings. There was age heterogeneity across studies, outcome variability by destination and differences in conceptualisation of outcomes (see online supplementary appendices B and C). Study locations were varied, and it has been noted that associations between built environment features and school travel-mode choice can vary across studies set in different locations.57 A variety of methods of both outcome and exposure measurement were used, each with different biases, and few studies reported measurement reliability and validity. More objective measures of walking should be explored such as using observational counts. Although databases may be a more objective source of built environment data, there are limitations depending on how and when the data were collected. When trying to explain walking behaviour in children, it may be most appropriate to model parent or child perceptions of the built environment as, ultimately, parents and children make the decision of whether or not the child walks to school.46 ,58 ,59 However, child and parent perceptions of the traffic environment differ and these likely differ from perceptions of a trained individual conducting a field survey.46 ,60 ,61 Several studies recommended that objective measurements be combined with perceptions of the environment.23 ,62 Correlates of active school transportation have been examined using both objective GIS measurements and parental perceptions of the built environment; however, only one study directly compared them using a GIS based overall walkability index.16 ,46 ,58 Another study found significant correlations between parent perceptions of overall walkability with walkability assessed via field audit.63
The limitations of this review include a focus on quantitative studies, which may have resulted in omission of environmental features. Qualitative work in this exists; however, the majority of the published work has used quantitative methods. Also, all walking outcomes were considered, but it is possible that built environment correlates could vary by destination. However, the majority of papers studied walking to school (84%). All significant correlates were included irrespective of whether there was control for confounding. The majority of the papers reviewed did control for SES and demographic factors (75/85=88%), but other factors such as vehicle ownership, weather and crime were not consistently controlled for in many of the studies. Meta-analysis was not conducted due to the heterogeneity of analytic techniques. This is problematic with meta-analysis of observational studies in general, as the diversity of study designs and populations results in summary statistics that are difficult to formulate and interpret.64 The purpose of this review was not, however, to assess the magnitude of effects but rather develop a list of correlates identified in different bodies of literature associated with more walking and less child pedestrian injuries.
Publication bias was difficult to assess as studies were observational and no registry exists. However, no language restrictions were imposed and internet searches were conducted to include unpublished reports and articles. Although difficult to test empirically, studies missing due to non-significant findings were not likely, as studies of built environment features incorporated many different exposures of which there was generally always at least one significant association. Studies reported both non-significant and significant associations for all exposures of interest. The analysis also indicated ‘any’ association with the outcome, and did not quantify the association or the number of studies where significant associations were found. Therefore, any omission of studies with completely null results would not have affected the findings. The PRISMA guidelines state that if publication bias exists, smaller studies would show larger estimates of the effects of the intervention.29 Comparisons of effect size by study size was not possible using, for example, funnel plots as consistent measurements of exposures and outcomes between studies were lacking (see online supplementary appendices B and C).
The majority of studies reviewed were cross-sectional and well designed controlled studies that examine the built environment, walking and child pedestrian injury were lacking. Therefore, inferences could only be made regarding associations and not causality. Randomised trials are difficult to design and implement for traffic interventions due to issues including high costs and lack of denominator exposure data. Other more feasible study design options should continue to be explored. Several of the pedestrian injury studies used case-control and case-crossover methods, which with further refinement, would have better validity than cross-sectional studies.39 ,44 ,48 ,55 ,65 Quasi-experimental designs are also feasible and produce more valid results when studying traffic injury and could be extended to studying walking. These studies are needed to investigate the effectiveness of design features such as specific traffic calming devices, crosswalks and sidewalks for children. Further work is also required for features that either had inconsistent associations or had not been studied for the walking or pedestrian outcome.
The feasibility of modifying built environment features and the time-frame required are important to consider when designing traffic intervention studies.27 Diversity and density features may be less easily modified in existing neighbourhoods, but should be considered when planning new neighbourhoods. Targeted interventions addressed at road environment design features, such as traffic calming, may be more feasible in an established neighbourhood compared with dealing with the more general issues related to higher population densities and land use mix.
The results of the current review will be disseminated to the City of Toronto, Transportation Services Department, and to the Green Communities Canada, SRTS Program, with whom working relationships are already established. Transportation Services is responsible for designing traffic environments and also conduct retrospective evaluation of the effectiveness of these environments in terms of injury and pedestrian activity. There must be close coordination between scientific investigators and traffic planners to implement well designed prospective traffic intervention studies. Green Communities Canada currently conducts and evaluates programmes to increase walking to school in Canada.17 Researchers and staff can work together to inform future SRTS evaluations for incorporating injury prevention.
This review described the current knowledge regarding the relationship between specific built environment features and both walking and child pedestrian injury. Built environment features that either slow traffic down (traffic calming) or separate children in space from traffic (playgrounds) were associated with both increased walking and less pedestrian injury. Many built environment factors associated with more walking were also associated with a greater risk of injury. Walkability assessment and evaluations of walking promotion interventions should include a pedestrian injury component to ensure that increased walking does not have detrimental effects on child pedestrian safety. Likewise, evaluation of traffic safety interventions should also address the effectiveness of the intervention in promoting walking. An interdisciplinary approach including city planners, community organisations and health and planning scholars is essential to evaluate and design appropriate interventions to increase walking while ensuring safety.
What is already known on the subject?
Many features of the built environment are associated with children walking.
Built environment roadway characteristics are related to child pedestrian injury risk.
Walkability constructs are varied and do not include assessment of pedestrian safety indicators.
What this study adds?
Playgrounds/parks and traffic calming were consistently related to both increased walking and decreased injury.
Other built environment correlates were found to either increase injury as well as increase walking or decrease walking as well as decrease injury.
Walking rate and injury rate need to be considered together to optimise health outcomes of active transportation.
New UN report on global road safety
The UN Secretary-General has issued a report, ‘Improving global road safety’, that draws attention to key developments over the last 2 years. Although considerable progress has been made, to meet the goal of the Decade of Action for Road Safety 2011–2020—to save 5 million lives—much more is needed to protect vulnerable road users. Comprehensive safety laws need to be adopted and enforced. As well, UN calls for more financial support; implementation of better road safety management systems; new car assessment programmes; improved prehospital, trauma and rehabilitation care; and the development of a set of targets and indicators against which progress can be systematically measured.
We would like to acknowledge Elizabeth Uleryk for her assistance in developing the literature search strategy and Andi Camden for her assistance in reviewing the articles.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
Contributors LR was responsible for the conceptual framework and design, analysis and interpretation, and writing and editing of the manuscript. AH and RB contributed to the conceptual framework and design, analysis and interpretation, writing and critical editing of the manuscript. CM and TT contributed to the conceptual framework and design, and critical editing of the manuscript.
Funding This work was supported by a CIHR Doctoral Research Award. CIHR had no role in the study or in the decision to submit the article for publication. The views and conclusions expressed are the authors’ and may not reflect those of the funders.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.