Background & Summary

In a publicly funded, universal health care system, there is a fundamental expectation that all Canadians have equal and equitable access to primary healthcare services. Although there is general agreement about the importance and application of this principle, where primary healthcare is covered under a national single-payer system, the equity of access to healthcare services experienced across the country is in question1,2. In particular, the vast geography of Canada challenges the provision of timely and appropriate access to high quality healthcare for all, regardless of where they live3,4.

While accessibility to health and care services is multi-dimensional, access to 24-hr EDs can be conceptualized as a primarily geographic problem, where space and place must be explicitly included in analytic approaches. Therefore, it is simply not enough to illustrate the physical location of a set of healthcare resources, we must also measure geographic access, with travel time and distance important components5,6. This paper uses accessibility in reference to the multi-dimensional geographic concept, and access in reference to spatial proximity.

Accessibility often refers to the degree of interconnection from a starting point to all other points in a space and is a key factor in evaluating availability of healthcare facilities for resident populations5,7. A place-based measure of accessibility includes proximity and travel time to services as measures of access8,9,10. A distance decay association is documented in most Canadian studies of the effect of travel time to health and care resources, indicating reduced utilization of resources for those who reside further from healthcare facilities11 and meaning that the physical presence of enough capacity in a health care system does not on its own determine the accessibility of that system.

The conceptualization of accessibility can also include individual-based measures, such as occupancy, degree of service capability, or staffing capacity12,13, these measures were excluded given the objectives of this paper. For the purposes of developing a transparent and easily updatable data set, this analysis will exclude on these individual-based accessibility measures and focus on place-based measures of access.

A resource for modelling healthcare accessibility that is currently lacking is a national dataset of access to hospitals that have 24-hour emergency departments (24-hr EDs). Leading up to the COVID-19 pandemic, access to 24-hr EDs was of significant concern and became increasingly important as the COVID-19 pandemic progressed. The COVID-19 pandemic highlighted that some provinces, regions, and communities have better access to healthcare resources than others14.

There is evidence that decreased access to 24-hr EDs negatively impacts health outcomes, particularly in relation to trauma, myocardial infarction, and stroke15,16,17,18. However, despite the evidence for ensuring geographic access to primary healthcare, there are few datasets available that provide nationally-uniform, validated, and reliable estimates of travel time or distance.

The overall objective of this project was to adapt the methodology successfully developed by Barbieri & Jorn in Australia19 for the Canadian context and develop an open-access database of estimated travel times and associated travel distances between all communities and 24-hr EDs. Analysis of estimated travel times in Canada have been completed for specific diseases or with tighter geographical constraints20,21,22, but this has yet to be done across all provinces, has not used a population-weighted approach or been validated across multiple methods. A national level analysis allows us to better understand the geographical disparity between provinces due to their differing geographies and population distributions. An overview of the process used to create this dataset is presented in Fig. 1.

Fig. 1
figure 1

Schematic overview of the process used to compute travel times between dissemination areas and hospitals as well as between census subdivisions and hospitals.

Full size image

This dataset and associated code are released into the public domain with the goal of enabling future research on access to healthcare services and health outcomes in Canada. This is of particular use for the increasing number of researchers using health administrative data to address policy-relevant questions. We also hope that this work will promote the importance of geographical equity to healthcare in Canada by showing the significant gaps experienced between regions and degrees of remoteness. This will encourage consistency in measurement of the geographic influence on healthcare accessibility and outcomes in Canada. Furthermore, the clear methodology is applicable to generating similar datasets for other healthcare or social resources within Canada, and for application in other countries.

Methods

Data quality – hospital Identification

This dataset includes all hospitals with a 24-hr ED in Canada. Given that healthcare services are managed separately for each province, it is difficult to obtain a single, reliable dataset of ED facilities.

The hospital EDs used for this analysis were generated based on the Open Database of Healthcare Facilities, a Canada-wide healthcare facility database compiled by the Centre for Special Business Projects at Statistics Canada23. Most of the data is sourced from municipal, regional, and provincial/territorial governments, federal agencies, or independent not-for-profit organizations specializing in the health information field. This list, generated in August 2020, contained 7,033 facilities and includes ambulatory health care services, nursing and residential care facilities, and specialized care hospitals such as cancer treatment centres, children’s hospitals, and women’s health hospitals. Supplementary data fields included in this database included the facility name, facility type, street address, city, postal code, province, census subdivision identifier, latitude, and longitude.

For the purposes of this study, only hospitals that had a 24-hr ED were included. However, the open database did not contain details on whether EDs operated on a 24-hr basis. To account for this, 24-hr EDs were identified through provincial or regional health authority websites, depending on the province’s structure of health administration. Inappropriate facilities for this analysis were removed by identifying facilities with appropriate North American Industry Classification System (NAICS) (Code# 62211) and Standard Industrial Classification (SIC) (Code# 8611) codes and keyword search in the name of the facility. This identification was then cross-verified by accessing each hospital’s open data resources to ensure they advertise themselves as offering 24-hour emergency services. Through this cross-verification process, the addresses and geolocation of the hospitals were also confirmed to ensure accurate geolocating of each facility.

Prior to this database development, no cleaned and succinct data source identifying 24-hr EDs has been developed for Canada. Given the limited availability of data and different community structures, the northern territories have been excluded from this analysis. Differing analysis, appropriate for the systematic and structural differences experienced within the territories will be conducted.

As well, some facilities identified as having a normally functioning 24-hr ED were found to have modified capabilities due to staff shortage. This was particularly prevalent in the Atlantic region, were only limited 24-hr ED services were offered with redirection to appropriate services. Since these circumstances were advertised as temporary these facilities were included in this analysis as under normal circumstances these facilities would be functioning in a way that is appropriate for this study. The location of 24-hr EDs have been presented in Fig. 2a.

Fig. 2
figure 2

(a) (top left) Map of all facilities with a 24-hr ED in Canada; (b) (top right) Map of the shortest time routes from DA centroid to closest 24-hr ED in Canada; (c) (bottom left) Gradient map of the shortest time routes from DA centroid to nearest 24-hr ED in Canada; (d) (bottom right) Population weighted average travel time by Census subdivision in Canada.

Full size image

Computation of dissemination area centroids

The 2021 DA24 and CSD25 shapefiles were obtained from Statistics Canada. A DA is a small, stable geographic unit containing 400 to 700 people based on the previous national Census26. A CSD is equivalent to a municipal boundary (or similar definition according to the jurisdiction) or an area that is deemed to be equivalent and is used for statistical reporting purposes. DAs were linked to their associated CSDs through the dissemination geographic relationships provided the Postal Code Conversion File (PCCF) from Statistics Canada27. This allowed the generation of average of values based on the larger spatial region using information generated at the DA level28. Also needed to generate a population weighted average is a rounded population count file for each DA based on the population identified through the 2021 Census cycle29.

However, given the known sparse population distribution and size of DAs throughout Canada, particularly through the rural spaces of the country, there was an identified need to implement populated weighted centroids at the DA level to ensure the populations observed in a DA are properly represented. To do this we required the centroids and population of the units that comprise DAs, called Dissemination Blocks. (DB). Traditionally used to identify the national road network, DBs are the smallest unit in which population is provided for, rounded to a base of 5 and are exclusively contained within a singular DA30. To identify centroids, population, and linkage between DB and DA, the Geographic Attribute File (GAF) from Statistics Canada was used31. The GAF provides users the ability to aggregate DBs to all geographic hierarchal levels31. Through this, we generated average centroids for DAs based on the information provided.

Computation of travel times from dissemination area centroids to closest hospitals

To generate the shortest travel time between DA centroids and the closest 24-hr ED, we generated a road network within ArcGIS Pro based on the 2021 Census Road Network file32. The 2021 Census Road Network (RNF) file depicts the digit road line coverage for all of Canada, including road names, types, directions, intersection types, and speed limits32. It is based on the National Geographic Database (NGD), a joint endeavour by Statistics Canada and Elections Canada to develop and maintain a national road network database that serves the needs of both organizations32. The purpose of the 2021 RNF is to provide a framework for mapping and spatial analysis and to support Geographic Information System (GIS) applications, such as this project. Given the intended representativeness of the data source, based on the requirements of both the distribution of the Census of Population and Elections Canada, it can be assumed all residential roads are properly represented.

The RNF was converted to a network analysis file and used as the physical restriction for identifying the nearest distance between the DA centroids and 24-hr EDs. Due to the limitations of the RNF and temporality of road conditions, this modelling assumes ideal road conditions at any time of day. This means the network analysis excluded roads that were flagged as seasonal (i.e. Ice roads, dirt camping roads). It also excludes pathing that may commonly be utilized by individuals but are not formal recognized (i.e. personal boating routes). However, it does include public means of transportation that are not bound by cars such as public ferry systems. Since population and centroid datasets are developed based on the usual residents per spatial used and we are seeking to understand experiences by the resident population of these spaces, we sought to focus on a generalised approach, rather than a seasonal or temporal analysis.

The nearest feature function of ArcGIS was utilised to identify the shortest distance between DA centroids to 24-hr EDs. This spatial join function joins the attributes of two layers based on the location of features between differing layers. For the purposes of this analysis, we chose to identify the geographically closest 24 hr ED to a DA centroid. Although this analysis is often conducted based on the shortest distance from a straight line, we desired to develop an understanding of closest feature, based on the available 2021 Census Road Network parameters. By following the road network, the nearest features analysis can identify travel time based on the road speed provided for each road segment and calculated distance based on the reported length of the road segments travelled based on the identified shortest time pathway. This analysis also incorporated standard road rules based on the 2021 RNF including one-way streets, vehicle type restrictions, or avoiding illegal turns, resulting in some areas having no route being found (e.g., no permanent roads available, seasonal travel, no public ferries). It also assumed all travel conducted are restricted to being in a personal automobile. Timing for traffic lights is not considered in this analysis. Additionally, driver is not allowed to exceed the posted speed limit for a road in an emergency, making the results of this output representative of a generalised experience. For both the Google Maps API and developed model, the time reported to use the public ferry ways were based on the reported sailing scheudling.

A map presenting the shortest identified route from DA centroid to 24-hr ED has been presented in Fig. 2b, and a gradient map with the estimated times have been presented in Fig. 2c.

Computation of population weighted travel times for census subdivisions

Once the shortest distance between a DA centroid and hospital was identified, these values were population-weighted based on CSD populations, through a custom script written in Python version 3.10. This calculation determined the average population weighted travel time and distance (travel pathing) between CSDs and hospitals (presented in Fig. 2d).

Figure 2a presented the 24-hr EDs identified in the provincial regions of Canada, Fig. 2b presents the shortest route from DA centroid to closest hospital based on travel time. Figure 2c presents a gradient value of the estimated travel times ranging from (yellow - <1 min) to purple (>59 minutes). Figure 2d presents the population weighted travel time experienced in a CSD to nearest identified hospitals ranging from light purple (<1 minute) to darker purple (>120 minutes).

Mirroring results have been presented in Fig. 3a–d for the Eastern Ontario area to provide insight to the small area assessment of travel times due to the wide nature of Canada’s landmass. This area was selected due to its mixture of formally recognized urban and rural spaces.

Fig. 3
figure 3

(a) (top left) Map of all facilities with a 24-hr ED in Eastern Ontario; (b) (top right) Map of the shortest time routes from DA centroid to closest 24-hr ED in Eastern Ontario; (c) (bottom left) Gradient map of the shortest time routes from DA centroid to nearest 24-hr ED in Eastern Ontario; (d) (bottom right) Population weighted average travel time by Census subdivision in Eastern Ontario.

Full size image

Data Records

The dataset is available at figshare (Estimated Travel Times to Canadian 24-hr Emergency Departments, and specific details on file size and format are reported in Table 1. Further detail of these datafiles have been provided through a codebook featured within the figshare project33.

Table 1 Format and size of data records.
Full size table

The “Popwgt_DA_Cent” table contains the 57,932 population-weighted DA Codes associated with the 2021 census of population, DAUID, the province Identification Code, PRUID, the reported latitude and longitude of the DA centroid, x and y respectively, and the estimated population of the DA based on the 2021 Census of Population, Population.

The “ED_Location_Geocoded” table contains the name, address information, and latitude and longitude of each identified 24-hr ED.

The “Can_Road_Network” shapefile (shp) contains the road network map that was used to create the network analysis file necessary to compute the travel times and distances. The”DA_to_24hrED_Merge” files contain the list of 57,607 DAs who were able to identify a closest 24hr ED through the provided road network, their reported travel times time in minutes, Total_Minu, and travel distance in kilometres, Total_Kilo, along with other identifying information provided through the previous files. In this file, the starting point variable labels begin with From_ and the end point variable names end with To_.

The “CSD_DA_Union” file contain the list of CSD unique identifiers CSDUID associated with each DA, DAUID.

Finally, the “Pop_weight avg_time” file contains the population weighted average travel time in seconds and minutes for each CSD and the “Pop_weight_avg_dis” file contains the population-weighted average travel distance by CSD.

Technical Validation

Since no previous estimates of corresponding results have been developed for 24-hr EDs in Canada, a technical validation was implemented to ensure the accuracy of developed results. For each Canadian province, 1,000 routes between the DA centroids and the closest hospital were sampled, without replacement. The Atlantic provinces (PEI, NL, NB, and NS) were combined into a single sampling frame as they did not have 1,000 centroid distances to measure individually. In total, 1,000 routes were sampled from the combined Atlantic provinces, Alberta, British Columbia, Manitoba, Ontario, Quebec, and Saskatchewan, resulting in a total of 7,000 routes being sampled.

Estimated travel times from samples were compared between ArcGIS Pro routing and Google Maps API Distance Matrix services. Parameters applied to the Google Maps API (version 3.56) Distance Matrix and ArcGIS Pro Model include limiting the travel mode to be driving, allow for tolls, highways, and ferries, and avoiding indoor pathing (walking paths, transit). Note that neither model considered the time that may be required to reach the road network, current traffic conditions, temporary road closures, or modes of transportation other than traveling by car34. Google Maps API does not account for traffic lights through their travel time estimates35. The same methodology was applied to travel distance.

Departure time for both the ArcGIS Pro and Google Maps API were not specified. Given this parameter the ArcGIS Pro model extrapolates the travel time/path based on the provided road network and known travel speeds of road types. The Google Maps API also develops a chosen route given the parameters based on the road network and the average travel time, independent of traffic. Mean differences in seconds (Google Maps API time – ArcGIS Pro time) and meters with corresponding standard deviation and 95% confidence intervals are reported in Table 2.

Table 2 Average differences in estimated travel times computed by Google Maps and ArcGIS Pro Nearest Route function across Canadian Provinces.
Full size table

The estimated travel times were comparable, with Google Maps being slightly more conservative, and 99% of the estimated differences being less than 5 minutes. Similar analysis was conducted for travel distance. For travel distance, the estimated travel distance was comparable, with the Google Maps API again being more conservative, giving longer distances that what was predicted through ArcGIS Pro. As well, 99% of the estimated differences reported were less than 1 kilometer.

For the 1% of routes with a time difference greater than 5 minutes (N = 389), time difference ranged from 6 minutes – 14 minutes, and for the 2% of routes with a distance difference greater than 1 km (N = 452), difference differences ranging from 1.01 km – 7.6 km. An assessment of these routes and reasons behind differences was conducted, and it was determined that often differences occurred because of the Google Maps API capturing roads not recognized by the 2021 RNF (i.e. Private roads, Dirt roads) or road closures that have occurred between May 10th 2021, the day the 2021 RNF was based on to June 15th 2024, the day the Google Maps API was implemented. Although notable difference, when these values were population weighted to the CSD level, differences in reported value were minimal.

In the final dataset, 325 DAs did not have a reported travel distance or time. Of these, 222 where is the territorial regions of Canada, and would have been excluded from the analysis given our parameters. The remaining 103 routes exist in the provincial regions of Canada but no route for publicly maintained roads, bridges or ferry ways leading to or near the DA centroid exists. Included in these 103 routes are three DAUIDs where a path was identified based on the population weighted centroid but upon further review would not be realistic as they required the individual to access a public ferry system within the ferry pathway. Although this is not problematic for the DAs that do not have a population reported (65/103), this could be problematic for the DAs that do contain a reported population (38/103). The total population of these DAs was 18,210, but when manually assessing why these spaces may not have a reported travel time through the Google Maps API output, the spaces were reported to be only accessible through alternative means of transportation (private boat, plane) not captured through the ArcGIS model or Google Maps API model as per the parameters provided. Based on this information, a secondary analysis assessing the healthcare accessibility of these areas, along with the territories, is required to understand how communities with limited infrastructure accessibility can access healthcare resources equitability. Most CSDs contained many DA centroids, while for some the CSDs only contained one centroid. If these single centroids did not have a population associated with it or did not have a route associated with it, then no average weighted travel information was calculated. However, if there was more than one DA in the CSD area, and the other DAs had a reported route, then the weight averages were calculated based on these valid values.

Usage Notes

To integrate this information, there are several ways this data can be used given data availability resources. If a researcher is using medical records, these resources often provide either the patients full address, Postal Code, or FSA. All these sources of information can be used to identify an estimated travel time from DA or CSD (as desired) through linkage based on the PCCF. If the data resource provided only provides the individuals DA or CSD, these values can be directly linked to the data resource provided.

This data resource can be used to provide a community or city level estimate of access which can be useful when building community profiles within regression models in health-based research. This type of linkage allows proper estimate of access as a factor to healthcare accessibility, utilization and outcomes. Future research optimizing on these measurements will allow for accurate representation and allow more detailed analysis of whether individuals are accessing the closest hospital, and if not, why are they deviating from the closest resources available to them.