Road traffic injury imposes a yearly global burden of morbidity and mortality of approximately 1.35 million fatalities and 50 million injuries.1 In Canada, road traffic injury caused over 1800 fatalities and 10,000 serious injuries annually over the last decade,2 each of which incurs health and economic costs.3 As with many health outcomes in Canada, there is evidence indicating that marginalized populations bear a disproportionate burden of these injuries and fatalities.4–8 Data limitations impede risk disparity measurement. These limitations include the systemic underreporting of injuries involving active transportation modes, the limited collection of sociodemographic variables, and an absence of travel data to account for exposure to risk (e.g., time or distance traveled by a specific transport mode).

In Canada, as in much of the industrialized world, the data that are typically used to estimate population-level traffic injury are derived from police-reported collision data.2,9 Police data underreport bicycling and pedestrian injury, disproportionately comprise incidents involving motor vehicles, report mostly on those collisions of greatest severity due to policies related to when they attend a collision, and tend to contain only a few basic sociodemographic variables that limit comparisons of injuries between population groups.10–12 Hospital records of treated injuries present an alternative, but these also have limited sociodemographic data and, historically, are difficult to access at a national level.13 Previous research into differences in transportation injury risk between sociodemographic groups in Canada has generally been characterized by use of area-level indicators,5 which has been shown to have low agreement with individual-level indicators.14 Moreover, these sources only count the number of traffic injuries and therefore do not measure risk of injury.

To compare risk of injury between different travel modes and population groups, counts of injury must be adjusted by measurements of exposure to risk within these populations (e.g., time, distance, or number of trips by mode). However, Canada lacks a national household travel survey, such as those conducted in the US, UK, and other developed countries.15 As a result, Canadian research is limited in comparing transportation injury risks across sociodemographic groups and travel modes and understanding risk inequities.15

The Canadian Census Health and Environment Cohorts (CanCHECs) are a series of population-based cohorts that probabilistically link detailed sociodemographic and commute mode data from the Canadian Census to health administrative databases, including hospitalization records and mortality data. Previously, these linked data have been used to examine the relationship between sociodemographic groups16,17 and environmental factors18,19 with health outcomes such as mortality and cancer diagnoses. The CanCHECs provide a unique opportunity to enable investigation into differences between sociodemographic groups using individual-level census data. Nationally, census data on the journey to work is one of few population-based data sources on use of active transportation among Canadians. The journey to work census module captures the main mode of commute for Canadians during the week of the census for the working adult population.20,21 The CanCHECs, by linking census data to hospitalization and mortality data, address traditional data limitations associated with the use of police reports, hospital, and mortality records.

Our goal is to quantify differences in risk of injury resulting in hospitalization or death for adult bicyclists, pedestrians, and motor vehicle occupants who commute to work in Canada, across sociodemographic groups and accounting for an indicator of use of different modes of transportation, represented by the participants’ primary mode of commute. We pool multiple waves of the CanCHECs linked to hospitalizations and deaths and construct cohorts to examine these separately. We use survival analysis to compare injury risks and identify sociodemographic groups at higher risk of hospitalization and death as a bicyclist, pedestrian, or motor vehicle occupant.

METHODS Data and Cohort Construction

The CanCHECs are probabilistically linked individual-level long-form census data (and the 2011 equivalent National Household Survey) to nationally compiled hospitalization and mortality databases. The CanCHECs are accessed in Canada via the Canadian Research Data Centre Network. The analyses presented here were conducted at the Vancouver Research Data Centre. All research results from Canadian Research Data Centre Network projects are subject to disclosure guidelines including random rounding of hospitalization and fatality counts. This project received approval by the Toronto Metropolitan University Research Ethics Board (TMU REB # 2020-136).

The Canadian census is conducted every 5 years in early May. The long-form component of the census corresponds to a 20% sample of the population in the 1996, 2001, and 2006 cycles with a nearly complete response rate (93.8% in 2006). In 2011, the long-form census was replaced with a voluntary National Household Survey, which sampled 30% of the population but had a response rate of 68.6%.22 Hospitalization data is from the Discharge Abstract Database (DAD, compiled by the Canadian Institute for Health Information). The DAD captures all hospitalizations from acute care facilities in all provinces and territories, with the exception of Quebec.23 Mortality data is from the Canadian Vital Statistics Death Database (CVSD). The CVSD collects mortality data from each province and territory’s vital statistics agencies and captures all deaths that occur in Canada.24 The CanCHECs are only linked to CVSD for the population aged 19+, while the linkage to DAD data spans all ages.

Census long-form data linked to DAD and CVSD enable stratification of transportation injuries by work commute mode data and sociodemographic information captured in the census.25 We conducted analysis of hospitalizations and fatalities separately and create two analytic samples of commuting adults that we will refer to as the hospitalization cohort and the fatality cohort. The construction of each cohort was predicated on the criteria for being included in the linkage to either DAD or CVSD as well as the criteria for inclusion in the census data we examined including commute modes and basic sociodemographic indicators. Only respondents that had employment between 1 January and the week of the census were asked questions regarding commute modes and are included in our analysis. The fatality cohort consisted of persons aged 19 and over who completed the long-form census in 1996, 2001, 2006, or 2011, resided in a Canadian province, and had a regular place of employment at some point since 1 January of that year they completed the census (Figure 1). The hospitalization cohort consisted of persons aged 15 and over who completed the long-form census in either 2006 or 2011, resided in a Canadian province other than Quebec, and had a regular place of employment at some point since 1 January of that year they completed the census (Figure 2). Participants in both cohorts enter on the census day corresponding to their completed census cycle: 14 May (1996), 15 May (2001), 16 May (2006), and 10 May (2011).

Figure 1.:

Fatality cohort construction.

Figure 2.:

Hospitalization cohort construction.

Outcome Ascertainment

For both cohorts, we identified three types of injury outcomes within the DAD and CVSD respectively, including hospitalizations and/or deaths as a (1) bicyclist; (2) pedestrian; or (3) motor vehicle occupant. In the fatality cohort, bicyclist deaths were identified using International Classification of Diseases 10th revision external cause codes V100–V199, pedestrian deaths using V010–V099, and motor vehicle occupant deaths using V400–V699 (but only including codes that specified the victim as a driver, passenger, or unspecified). Deaths that occurred between 1996 and 2000 were identified using International Classification of Diseases 9th revision external cause codes (eTable 1, https://links.lww.com/EDE/C96).

We identified bicyclist, pedestrian, and motor vehicle occupant hospitalizations using the same International Classification of Diseases 10th revision external cause codes as in the fatality cohort. Deaths as a bicyclist, pedestrian, or motor vehicle occupant not associated with a hospitalization were also identified using these International Classification of Diseases 10th revision codes. We included repeat hospitalizations if the subsequent injury occurred after a period of 30 days.

Travel Mode

We measured a participant’s use of different modes of travel using a categorical variable that describes the participant’s main mode of transport they used for commuting to work. This variable is derived from the journey to work module in the long-form census.15 Since 1996, the long-form census queries respondents on the “main mode of commuting,” which refers to the main mode of transportation used to travel to and from their home and workplace.15 This question is restricted to the population in private households with employment and applies to the job held the week of the census (early May) or – if not working at the time of the census – the job held for the longest period since 1 January of that year. If a person used more than one mode, they were asked to choose the mode they used for the greatest distance. Respondents could select from several options, some of which have changed from 1996 to 2011 (Table 1). We simplified the census variable to categories of commuting modes as (1) car, truck, or van; (2) bicycle; (3) walking; or (4) other (Table 1). We use “motorized vehicle” to refer to car, truck, or van modes.

Table 1. - Main Mode of Commute Variable and Its Corresponding Categories From the Census Main Mode of Commute Census Main Mode of Commute Years Applicable Car, truck, or van Car, truck, or van – as a driver 1996, 2001, 2006, 2011 Car, truck, or van – as a passenger 1996, 2001, 2006, 2011 Bicycle Bicycle 1996, 2001, 2006, 2011 Walking Walked to work 1996, 2001, 2006, 2011 Other Public transit 1996, 2001, 2006 Motorcycle 1996, 2001, 2006 Taxicab 1996, 2001, 2006 Subway or elevated rail 2011 Passenger ferry 2011 Light rail, streetcar, or commuter train 2011 Motorcycle, scooter or moped 2011 Other 1996, 2001, 2006, 2011
Covariates

To examine injury risk across basic sociodemographic characteristics, we included the following covariates: (1) age at baseline; (2) gender; (3) low-income cutoff (LICO) (after tax); (4) self-identified racialization; and (5) recent immigration status. LICOs are based on a threshold set by Statistics Canada where an individual spends 20% points more than average of their after-tax income on necessities of life (food, shelter clothing).26 This threshold varies for an individual based on their family size and the population characteristics of the area they live in.26 Since 1996, Statistics Canada has collected information on racialized population groups based on the framework of belonging to a visible minority, Indigenous, or White population group. Visible minorities are defined by Statistics Canada as a person identifying as non-Caucasian in race or nonwhite in color.27 Between 1996 and 2011, Statistics Canada used the term Aboriginal but has since been replaced and we will use Indigenous as the current terminology. Indigeneity is self-identified as belonging to at least one Indigenous group including First Nations (North American Indian), Métis, or Inuk (Inuit). The census also collects information on the year a person immigrated to Canada. Recent immigration was originally defined as settlement in Canada less than 5 years before census date, but due to small cell sizes, we expanded this definition to include settlement less than 10 years before census date.

Statistical Analysis

To examine associations among exposure, covariates, and hospitalizations or fatal injuries, we fit a series of Cox proportional hazards models that estimated risk for bicycling, pedestrian, and motor vehicle occupant injury separately. For each covariate we estimated hazard ratios (HR) and 95% confidence intervals (CIs) for time to: (1) bicycling hospitalization; (2) bicycling fatality; (3) pedestrian hospitalization; (4) pedestrian fatality; (5) motor vehicle occupant hospitalization; and (6) motor vehicle occupant fatality. For each outcome, participants were censored at the end of the follow-up period (Table 2) or upon the date of death from any cause. In the hospitalization analysis, we modeled repeated events using a counting process marginal rates approach.28 Analyses are unweighted. Given the complex interplay among sociodemographic indicators and transportation injury in the prior literature, we modeled injury against mode of travel mutually adjusted for all sociodemographic indicators. Because age and sex have particular bearing on mode of travel related injuries, we also included a minimally adjusted model with only age and sex as covariates. Finally, we include unadjusted results in eTables 2–4, https://links.lww.com/EDE/C96. Analyses were conducted in R version 3.5.3 https://links.lww.com/EDE/C97; https://links.lww.com/EDE/C98.

Table 2. - Follow-up Periods for Health Outcome Data by Canadian Census Health and Environment Cohort CanCHEC Ages Included CVSD DAD 1996 (version 3) 19+ 14 May 1996–31 December 2016 n/a 2001 (version 3) 19+ 15 May 2001–31 December 2016 n/a 2006 (version 1.1) 0+ 16 May 2006–31 December 2019 16 May 2006–31 March 2017 2011 (version 2) 0+ 10 May 2011–31 December 2018 10 May 2011–31 March 2017

CanCHEC indicates Canadian Census Health and Environment Cohort; CVSD, Canadian Vital Statistics Death Database; DAD, Discharge Abstract Database; n/a, not available.

RESULTS

The hospitalization cohort included 4,815,970 persons who contributed 39.3 million person years between 2006 and 2018 follow-up period (Table 3). The fatality cohort included 10,501,605 persons and 129.4 million person years between the 1996 and 2018 follow-up period (Table 4). Most members of both cohorts reported using a car, truck, or van (as driver or passenger) as their main mode of commute were White, were long-term residents or Canadian born, and were not low income (Tables 3 and 4). Most members of both cohorts were under the age of 65, reflecting our inclusion of only participants who were working outside the home at baseline (Tables 3 and 4).

Table 3. - Sociodemographic and Injury Characteristics of the CanCHEC Hospitalization Cohort Hospitalization Incidence Rate Per Million Person Years Variable Level Participants (%) Million Person Years (%) Bicyclist; N (%) Pedestrian; N (%) Motor Vehicle; N (%) Bicyclist Pedestrian Motor Vehicle Total 4,815,970 (100.0) 39.3 (100.0) 4,560 (100) 2,270 (100) 10,325 (100) 116.0 57.7 262.6 Main commute mode Car, truck, van 3,852,730 (80.0) 31.5 (80.1) 3,085 (68) 1,430 (63) 8,815 (85) 97.9 45.4 279.8 Bicycle 62,990 (1.3) 0.5 (1.3) 535 (12) 35 (2) 110 (1) 1,028.8 67.3 211.5 Walking 288,055 (6.0) 2.4 (6.1) 345 (8) 245 (11) 555 (5) 143.8 102.1 231.2 Other mode 612,195 (12.7) 4.9 (12.5) 595 (13) 565 (25) 845 (8) 121.4 115.3 172.4 Age at baseline 15–24 813,505 (16.9) 6.8 (17.3) 850 (19) 495 (22) 2,605 (25) 124.6 72.6 382.0 25–34 937,810 (19.5) 7.7 (19.6) 810 (18) 280 (12) 1,710 (17) 105.1 36.3 221.8 35–44 1,066,740 (22.2) 8.9 (22.7) 1,000 (22) 360 (16) 1,860 (18) 112.1 40.4 208.5 45–54 1,150,535 (23.9) 9.3 (23.8) 1,170 (26) 540 (24) 2,075 (20) 125.1 57.8 221.9 55–64 690,635 (14.3) 5.4 (13.7) 595 (13) 435 (19) 1,530 (15) 110.4 80.7 283.9 65+ 156,745 (3.3) 1.1 (2.9) 130 (3) 155 (7) 545 (5) 114.0 136.0 478.1 Gendera Women 2,329,990 (48.4) 19.0 (48.4) 1,270 (28) 1,035 (46) 4,080 (40) 66.7 54.4 214.3 Men 2,485,980 (51.6) 20.3 (51.6) 3,290 (72) 1,235 (54) 6,245 (60) 162.2 60.9 307.9 LICO Non-low income 4,474,220 (92.9) 36.5 (92.9) 4,210 (92) 1,955 (86) 9,495 (92) 115.3 53.5 260.1 Low income 341,750 (7.1) 2.8 (7.1) 350 (8) 315 (14) 830 (8) 124.6 112.1 295.4 Racialization Not visible minority 3,755,900 (78.0) 30.8 (78.4) 4,035 (88) 1,665 (73) 8,340 (81) 130.8 54.0 270.4 Visible minority 909,820 (18.9) 7.3 (18.5) 370 (8) 465 (20) 1,365 (13) 51.0 64.0 188.0 Indigenousb 150,250 (3.1) 1.2 (3.1) 155 (3) 140 (6) 620 (6) 127.0 114.8 508.2 Recent immigrant at baseline No 4,472,620 (92.9) 36.5 (92.9) 4,380 (96) 2,090 (92) 9,850 (95) 119.9 57.2 269.6 Yes 343,350 (7.1) 2.8 (7.1) 180 (4) 180 (8) 475 (5) 64.5 64.5 170.3

aStatistics Canada collects “sex” without information on nonbinary gender in these census years.

bStatistics Canada uses the term “Aboriginal” in these census years.

LICO indicates low-income cut-off.

Table 4. - Sociodemographic and Injury Characteristics of the CanCHEC Fatality Cohort Deaths Incidence Rate Per Million Person Years Variable Level Participants (%) Million Person Years (%) Bicyclist; N (%) Pedestrian; N (%) Motor Vehicle; N (%) Bicyclist Pedestrian Motor Vehicle Total 10,501,605 (100) 129.4 (100) 240 (100) 960 (100) 3,170 (100) 1.9 7.4 24.5 Main commute mode Car, truck, van 8,500,560 (80.9) 105.2 (81.3) 155 (65) 700 (73) 2,770 (87) 1.5 6.7 26.3 Bicycle 119,135 (1.1) 1.4 (1.1) 25 (10) 15 (2) 35 (1) 17.6 10.6 24.6 Walking 617,255 (5.9) 7.8 (6.0) 20 (8) 80 (8) 155 (5) 2.6 10.3 19.9 Other mode 1,264,655 (12.0) 15.0 (11.6) 35 (15) 165 (17) 210 (7) 2.3 11.0 14.0 Age at baseline 15–24 1,256,860 (12.0) 15.6 (12.0) 25 (10) 100 (10) 550 (17) 1.6 6.4 35.3 25–34 2,291,655 (21.8) 29.5 (22.8) 35 (15) 135 (14) 610 (19) 1.2 4.6 20.7 35–44 2,719,145 (25.9) 35.6 (27.5) 60 (25) 230 (24) 790 (25) 1.7 6.5 22.2 45–54 2,590,830 (24.7) 31.2 (24.1) 75 (31) 225 (23) 730 (23) 2.4 7.2 23.4 55–64 1,362,570 (13.0) 14.8 (11.5) 35 (15) 185 (19) 380 (12) 2.4 12.5 25.6 65+ 280,545 (2.7) 2.7 (2.1) 10 (4) 80 (8) 115 (4) 3.7 29.7 42.8 Gendera Women 5,008,325 (47.7) 61.7 (47.7) 45 (19) 340 (35) 950 (30) 0.7 5.5 15.4 Men 5,493,285 (52.3) 67.7 (52.3) 195 (81) 620 (65) 2,220 (70) 2.9 9.2 32.8 LICO Non-low income 9,641,015 (91.8) 118.1 (91.3) 210 (88) 820 (85) 2,825 (89) 1.8 6.9 23.9 Low income 860,590 (8.2) 11.3 (8.7) 30 (12) 140 (15) 345 (11) 2.7 12.4 30.6 Racialization Not visible minority 8,769,890 (83.5) 109.7 (84.8) 225 (94) 770 (80) 2,840 (90) 2.1 7.0 25.9 Visible minority 1,493,360 (14.2) 17.0 (13.1) 5 (2) 120 (12) 220 (7) 0.3 7.1 13.0 Indigenousb 238,350 (2.3) 2.7 (2.1) 10 (4) 70 (7) 110 (3) 3.7 25.7 40.4 Recent immigrant at baseline No 9,845,210 (93.7) 121.4 (93.8) 235 (98) 895 (93) 3,075 (97) 1.9 7.4 25.3 Yes 656,395 (6.3) 8.0 (6.2) 5 (2) 65 (7) 100 (3) 0.6 8.2 12.5

aStatistics Canada collects “sex” without information on nonbinary gender in these census years.

bStatistics Canada uses the term “Aboriginal” in these census years.

LICO indicates low-income cut-off.

During the follow-up period, there were 4560 hospitalizations (Table 3) and 239 deaths due to bicyclist-related injuries (Table 4) and 2270 hospitalizations (Table 3) and 960 deaths for pedestrian-related injuries (Table 4). For motor vehicle occupant injuries, there were 10,325 hospitalizations (Table 3) and 3170 deaths (Table 4).

In our mutually adjusted models, primary mode of commute was strongly related to injury risk occurring in corresponding modes (Tables 5–7). This was most strongly observed for people whose primary mode of transportation to work was bicycle and who had increased risk for bicycle hospitalization (HR = 9.1; 95% CI = 8.3, 10) and fatality (HR = 11; 95% CI = 7.5, 17), compared with people whose primary mode of transportation to work was motor vehicle (Table 5). Those who walked to work experienced higher risk of pedestrian injury hospitalization (HR = 2.1; 95% CI = 1.8, 2.4) and fatality (HR = 1.5; 95% CI = 1.2, 1.9) (Table 6). By contrast, people who bicycled or walked, along with people who used “other” transportation modes (which includes public transport) had lower risk of motor vehicle-based hospitalization and fatality, as compared with people who mainly commuted by motor vehicle (Table 7). Collectively, these patterns suggest that mode of commute is a relevant indicator of exposure to risk. This finding is nuanced because bicyclists and users of “other” modes, including public transport, had elevated risks for both bicycling and pedestrian injury, suggesting that these commute modes may correlate to overall nonmotor vehicle transportation exposure. Unadjusted results (eTables 2–4, https://links.lww.com/EDE/C96) showed broad similarity in these patterns to minimally adjusted results.