2.1. Accessibility as a measure to assess transport projects

Boisjoly and El-Geneidy (Reference Boisjoly and El-Geneidy2017b) consider that accessibility is defined by a combination of factors: exogenous variables such as the distribution of activities in the territory (land use component) and the transportation network that determine the costs, time, and convenience of travel between an origin and a destination; individual characteristics such as income, education level, gender and vehicle ownership that affect the abilities and needs to access destinations; and finally time restrictions, not only imposed by the schedules and programming of the activities and the available transport but also at an individual level by personal schedules.

Thus, accessibility is determined by the spatial distribution of potential destinations, the ease of reaching each destination, and the magnitude, quality, and character of the activities found there, along with the cost, the selection of the destination, and the chosen mode of transport that also have an impact on the level of accessibility (Handy and Niemeier, Reference Handy and Niemeier1997).

Poor transport quality can contribute to social exclusion as it restricts access to work, education, healthcare, leisure, and various activities that enhance people’s life chances. Besides, deprived communities suffer disproportionately the negative impacts of insecure roads, contamination, and physical barriers that congested roads generate (Social Exclusion Unit, 2003). Accessibility analysis that focuses on the influence of transport investments on the range and number of opportunities reachable provides a holistic view that can nurture social approaches to transport planning to address social inclusion challenges (Scholl et al., Reference Scholl, Fook, Barahona, Rivas, Montes Calero, Montoya-Robledo, Pedraza, Noboa, Sandoval, Lee, Rodríguez Porcel, Juan Pablo Bocarejo, Vergel Tovar, Urrego, Moreno, Bertucci, Oviedo, Sabogal-Cardona, Serebrisky and Mojica2022).

In Latin America, equity and accessibility issues have been addressed in several works that analyze accessibility to travel attracting poles (Molfino et al., Reference Molfino, Albrieu and Depiante2007), accessibility levels to the labor market considering time and income constraints (Bocarejo S. and Oviedo H., Reference Bocarejo S. and Oviedo H.2012), public transport affordability and its differences among different household compositions by income (Falavigna and Hernandez, Reference Falavigna and Hernandez2016), and accessibility for non-commuting trip purposes by income groups via private and public transport (Oviedo and Guzman, Reference Oviedo and Guzman2020), among others.

Geurs and van Wee (Reference Geurs and van Wee2004) define four classes of accessibility measures based on the perspectives from which accessibility is analyzed (infrastructure, location of activities, person-based or utility-based measures) and the accessibility components that are considered (transport, land use, temporal and individual restrains).

Infrastructure-based measures, which include travel times estimations, are direct and easy-to-communicate indicators but have the restriction that they do not consider the land use factor with its impacts on accessibility and the impact that transportation can have on the distribution of activities. Besides, some authors discussed that travel time, as an indicator of mobility, does not capture the potential of interaction for opportunities, so it does not fully reflect access to destinations (Boisjoly and El-Geneidy, Reference Boisjoly and El-Geneidy2017a).

Location-based measures are widely used in the bibliography, the simplest ones called distance or connectivity measures, both linearly and considering the road network, with indicators such as travel times or average speeds; and other contour or isochronous measures that are based on counting the number of opportunities reached in a particular time, distance or cost, named as cumulative opportunities indicator (Geurs and van Wee, Reference Geurs and van Wee2004). Other measures based on the locations are called potential accessibility, such as the work of Hansen (Reference Hansen1959) or Shen (Reference Shen1998), which improves the previous definition by including the possible competition for the available opportunities considering the demand.

On the contrary, people-based indicators measure accessibility at the individual level, tracking personal times and modes of travel to different activities, along with people’s temporal and spatial restrictions, as shown in the work of Kwan (Reference Kwan2010) carried out through travel diaries.

Finally, the indicators based on the utility theory consider that people decide on one option over others based on its ability to satisfy their needs and the utility (or not) it generates. The Multinomial Logit model (logsum) or the doubly constrained entropy model (Geurs and van Wee, Reference Geurs and van Wee2004) are examples of utility-based measures.

2.2. The GTFS format to describe transit

A GTFS feed is a collection of comma-delimited text files collected into a single ZIP compressed file. Each text file models a specific aspect of public transport information, such as stops, routes, or trips (Google Developers, 2022). The schedules provided by GTFS data are significant for calculating travel times by public transport. The estimation of door-to-door travel time is a challenge because of the multiple steps that can integrate an entire trip by public transport, considering the access and egress from the system, the numerous itineraries that may exist to link an origin and a destination, the possible variations in departure times and the itinerary selection since the transport system availability is not uniform throughout the day.

Public transport accessibility studies were traditionally based on average values over peak hours, such as travel speed or programmed frequencies (Polzin et al., Reference Polzin, Pendyala and Navari2002; Wu and Hine, Reference Wu and Hine2003), or by the usage of average values from surveys or census data from the users (Kawabata, Reference Kawabata2003, Reference Kawabata2009; Kawabata and Shen, Reference Kawabata and Shen2006, Reference Kawabata and Shen2007). GTFS feeds are new information sources with details about public transport schedules that enhance travel time estimations. When used with computational routing algorithms, they allow the inclusion of variations in departure times, door-to-door travel times distribution, and the selection of alternative itineraries beyond the shortest route.

2.3. Accessibility to healthcare facilities

The World Health Organization (Evans et al., Reference Evans, Hsu and Boerma2013) sustains that universal access is needed to achieve the goal of universal health coverage, meaning that people obtain the health services they need without excessive financial risk. The availability of good health services within reasonable reach of those who need them, called physical accessibility, is one of the main elements to address this issue (Evans et al., Reference Evans, Hsu and Boerma2013).

Access to healthcare, as defined by Penchansky and Thomas (Reference Penchansky and Thomas1981), is a general concept to evaluate the degree of fit between the population needs and the services provided in five specific areas or dimensions of access, namely: (i) availability in volume and type of the required service; (ii) geographic or physical accessibility, regarding the location of the population and the location of the healthcare services; (iii) accommodation in terms of attention times and general quality of the service, and how the population can accommodate to these service characteristics; (iv) affordability, meaning the cost of the service related to the patient’s income; and (v) acceptability, including health service compliance and satisfaction. Analyzing these dimensions through surveys, the authors conclude that access variations may influence patient satisfaction and healthcare services utilization.

There is evidence in the literature that relates geographic accessibility to the quality of healthcare the population receives. Wang and Luo (Reference Wang and Luo2005) identified spatial (geographic barriers) and nonspatial factors (socioeconomic and demographic characteristics such as social class, income, ethnicity, age, sex, etc.) that influence the variation in access to healthcare across space. Considering nonspatial factors, socioeconomic disadvantages, including female-headed households, percentage of the population in poverty, nonwhite minorities, households without vehicles, home ownership, and housing units’ lack of basic amenities were the most influential factors that explained the variance.

In rural areas of Malawi, Varela et al. (Reference Varela, Young, Mkandawire, Groen, Banza and Viste2019) determined that the main barriers to timely access to healthcare were the lack of suitable transport, finances, and prolonged travel times and that inadequate road infrastructure, unavailability of public transport and expensive private transport play an important part in access to healthcare centers. Douthit et al. (Reference Douthit, Kiv, Dwolatzky and Biswas2015) conclude that there are significant differences in healthcare access between rural and urban areas and that some transportation difficulties, such as having (or not) a driver’s license or increased travel times, are prohibitive.

Other studies conducted in urban areas showed that accessibility decreases significantly from the city center outwards, related to the concentrated distribution of hospitals and the better road and public transport conditions in the central area (Tao and Cheng, Reference Tao and Cheng2019; Chen et al., Reference Chen, Cheng, Chen, Chen and Cao2021). Previous works also showed significant disparities in access to healthcare facilities by private and public transport, in which trips by individual vehicles have lower travel times and consequently higher levels of accessibility (Peipins et al., Reference Peipins, Graham, Young, Lewis and Flanagan2013; Langford et al., Reference Langford, Higgs and Fry2016; Tao et al., Reference Tao, Yao, Kong, Duan and Li2018; Ermagun and Tilahun, Reference Ermagun and Tilahun2020). Additionally, some authors discussed that the differentiation of healthcare facilities according to levels of attention and medical and surgical capabilities is substantial to appropriately address the gap between the population’s primary needs and the quality of healthcare received (Zheng et al., Reference Zheng, Xia, Ambinakudige, Qin, Li, Xie, Zhang and Gu2019; Zhao et al., Reference Zhao, Li and Liu2020).

This unequal geographical distribution in the intersection with nonspatial factors shown in several works that it has more impact on vulnerable population groups such as the elderly (Tao and Cheng, Reference Tao and Cheng2019; Ermagun and Tilahun, Reference Ermagun and Tilahun2020; Chen et al., Reference Chen, Cheng, Chen, Chen and Cao2021), people coming from low-income neighborhoods (Ermagun and Tilahun, Reference Ermagun and Tilahun2020; Zhao et al., Reference Zhao, Li and Liu2020), families with children (Ermagun and Tilahun, Reference Ermagun and Tilahun2020) and racialized groups (Peipins et al., Reference Peipins, Graham, Young, Lewis and Flanagan2013; Kim et al., Reference Kim, Ghorbanzadeh, Horner and Ozguven2021). For instance, Peipins et al. (Reference Peipins, Graham, Young, Lewis and Flanagan2013) found that majority-black census tracts had longer public transportation travel times than white tracts, while Ermagun and Tilahun (Reference Ermagun and Tilahun2020) concluded that areas of low accessibility have a higher percentage of African-Americans, Hispanics, Asians, low-income workers, low-educated citizens, and elderly people.

Travel time or distance as a primary constraint in urban areas was reflected in several previous works, and travel time values differ among them. Peipins et al. (Reference Peipins, Graham, Young, Lewis and Flanagan2013) calculated a median travel time to the nearest radiation facility of 56 min by public transport and 8 min by private. Tao et al. (Reference Tao, Yao, Kong, Duan and Li2018) set a maximum travel time threshold of 124 min to ensure that the elderly could access at least one hospital within the catchment area using a previous survey as a reference, while Chen et al. (Reference Chen, Cheng, Chen, Chen and Cao2021) threshold travel time used was 30 min, as the most of the minimum travel times modeled were under that value. However, Cao et al. (Reference Cao, Huang, Zhang, Liang, Xian, Gan, Xu and Lai2022) discussed that the gravity-based models or floating catchment (FCA) methods that researchers usually adopt use the same inaccessible time of 60 or 120 min across different areas, while a more accurate approach would be to use an acceptable travel time. Cao et al. (Reference Cao, Huang, Zhang, Liang, Xian, Gan, Xu and Lai2022) also obtained that more than 90% of the residents in Inner Mongolia accepted travel times lower than 30 min to a primary healthcare institution and that women tend to accept higher travel times while the elderly accepted shorter travel times than the youngest due to physical limitations such as deteriorating health conditions and reduced travel ability.

2.4. Accessibility to public hospitals in Córdoba

Mobility patterns to public hospitals in Córdoba have been characterized in Albrieu et al. (Reference Albrieu, Pastor and Galarraga2011) work, where through surveys, censuses, and vehicle counts, the authors performed a comparative analysis of trips to 7 locations of public and private hospitals. The main results showed that 95% of the patients in private hospitals came from the areas with the best socioeconomic background. On the contrary, the most vulnerable sectors do not use private hospitals. Also, 59% of the people who went to the public hospital did so by public transport, and only 24% used a private car. In opposition, 61% of people attended a private hospital using a private car, compared to only 29% of public transport usage.

Martinazzo and Falavigna (Reference Martinazzo and Falavigna2022) evaluated the potential accessibility to public hospitals between 2019 and 2021 by public transport in Córdoba to analyze and compare pre-COVID and during-COVID geographical accessibility to healthcare centers. The comparative analysis was made using the r5r package in R with the GTFS provided by the municipality, and the travel times were calculated with programmed schedules and frequencies and fixed values of maximum travel time and maximum walking distance.

The principal findings showed that for the three hospital categories evaluated, travel times increased around 20% in the study period, and the most vulnerable quintiles of the population increased their travel times 8 points above the less vulnerable quintiles. Overall, measures of central tendency and dispersion indicated significant differences between quintiles per percentage of households with unsatisfied needs (%NBI), where in general, zones with the highest proportions of households with unsatisfied needs suffered more the impact of the cuts in services during the pandemic, as shown in Figure 1.

Figure 1. Average travel times box plots per %NBI quintiles (Martinazzo and Falavigna, Reference Martinazzo and Falavigna2022).

The authors conclude that the differences in travel times for 2019 and 2021 among the least and most vulnerable groups show a higher detriment in potential accessibility for more deprived households, which illustrates part of the inequity they are subjected to.

3.1. General characteristics of the study area

Córdoba is the capital city and head of the Capital Department of the Province of Córdoba, in the central region of Argentina, with a territory of 573.2 km². According to the 2010 census, the Capital Department had 1,329,604 inhabitants and 414,237 households (INDEC, 2015). In contrast, preliminary results from the 2022 census indicate that the city population is 1,565,112 (INDEC, 2023), resulting in the second-largest city in Argentina. The city is also the reference for the province and the central region, as the consolidated area of Córdoba city concentrates the employment and service provider centers as well as the most significant number of educational establishments, almost all of the state hospital establishments, and essential and dynamic commercial, cultural, recreational and tourist attractions (Molfino et al., Reference Molfino, Albrieu and Depiante2007).

The percentage of households with unsatisfied basic needs (%NBI) measures socioeconomic status and structural poverty considering several elements apart from monetary income. In Córdoba, by 2010, 5.82% of the households showed at least one of the indicators included in the %NBI measure. To be a household with unsatisfied basic needs, at least one of the following deprivation conditions must be present: households that live in houses with more than three people per room (critical overcrowding), households that live in houses of an inconvenient type such as tenancy rooms or precarious residences, households that live in houses without flushing toilets, households with school-age children not attending school, a four (or more) member family group where only one of them is working and where the head of household attended two or less than two years to primary school (INDEC, 2000). The main deprivation conditions in Córdoba are overcrowding (3%), sanitary conditions (1.2%), and type of dwelling (1.1%) (Dirección General de Estadística y Censos de la Provincia de Córdoba, 2021).

As part of the 2010 population census, a broader survey was also conducted on a sample of households to address dwelling conditions, migration, healthcare services, social security, and employment. The results showed that, at the departmental level, 32.4% of Córdoba’s population depends exclusively on the public healthcare system, which includes state plans and people without any other type of assistance. Nearly 70% of people in this situation are between 15 and 64 years old.

The public healthcare system in Córdoba has municipal, provincial, and national institutions. In Figure 2, the territorial distribution of 14 of these institutions is shown with the 2010 census zones classified in quintiles per the percentage of households with unsatisfied basic needs (%NBI). The first quintile (Q1) groups the most vulnerable households, with the highest percentage of households with NBI, and the last quintile (Q5) corresponds to the less vulnerable households, with the lowest percentage of households with NBI.

Figure 2. Population census zoning by %NBI quintiles and public hospitals in Córdoba.

3.2. Transport and mobility system characteristics in the study area

The public transport system in Córdoba includes bus lines and trolley lines. The original network, licensed for 10 years in 2014, was structured with core and secondary lines. It also had six neighborhood lines, which connect areas of the city without entering the center; ring lines, which circulate clockwise and counterclockwise without going through the downtown area; and differential lines, which had air-conditioned buses. The system covers the extent of the city as it is an urban, not a metropolitan service, except for one line that goes to Bower on the southern periphery. Since 2021, an urban rail service is also provided but is not part of the urban transport system.

In 2019, three companies owned the concession of the 69 bus lines: Autobuses Córdoba (5/69), Coniferal (21/69), and ERSA (43/69). The municipal company Tamse provided the trolley service, consisting of three lines supplied by electric energy. During the 2021 first semester, the concession contract was still in force, but because of the decrease in demand due to the mobility restrictions for the COVID-19 pandemic, some bus lines stopped operating, and the distribution among the concession companies changed: Autobuses Córdoba (5/60), Coniferal (21/60), and ERSA (34/60). The trolley service provided by Tamse doubled the number of lines, but the power supply changed to a mixed scheme with part of the fleet using diesel engines, a system currently in force. By the end of 2021, Tamse took control of corridors 3, 8, and the neighborhood line B80, and by the end of 2022 also had lines 600, 601, 53, and 54, and the special Aerobus service that travels to and from the Airport. In the same period, Autobuses Córdoba stopped being a service provider. Tamse has 26/79 lines, Coniferal 22/79 lines, and Ersa 31/79 lines. These changes in the public transport system are described through its leading indicators in Table 1. In Figure 3, the public transport system routes were mapped using GTFS GO, a QGIS plugin to extract GTFS data to show routes and stops (MIERUNE Inc., 2023), where it can be seen how the lines cover more of the city center and less the periphery, and that almost all routes begin and end within the cities’ boundaries marked with the black line.

Table 1. 2019–2023 public transport system characteristics in Córdoba

* Data obtained through tidytransit on R using published GTFS, services from 6 a.m. to 8 p.m.

^** Information not published.

Figure 3. Routes of the public transport system in Córdoba 2019–2023.

3.3. Materials and methods

As proposed by Martinazzo and Falavigna (Reference Martinazzo and Falavigna2022), the r5r package in R (Pereira et al., Reference Pereira, Saraiva, Herszenhut, Braga and Conway2021) was used to calculate travel times between the centroids of the 2010 census zones (origins) and public hospitals from the IDERA-SISA database (destinations). Thereby, the travel time matrix obtained had 1,503 origins and 14 destinations. It was excluded from the analysis the mental health hospital of the city Hospital Neuropsiquiátrico, considering it is a facility that has a particular type of attention dedicated to hospitalizations and outpatient follow-up and in several of the hospitals analyzed there is a mental health service.

Besides the origin and destination information, the data requirements of r5r are an OpenStreetMap road network in .pbf format (obtained from Protomaps for this work) and a public transport feed in GTFS.zip format, using for this work the 2022 and 2023 GTFS published by the Municipality of Córdoba. The databases were initially obtained from the published 2010 INDEC population census, the IDERA database, where the hospitals’ information was obtained in a .kml format, and the open data webpage Gobierno Abierto from the Municipality of Córdoba, where GTFS for 2022 and 2023 were published.

The centroids of the census zones used as origins were formerly obtained using QGIS, and the .csv database is published on an open repository on GitHub. The destinations selected were the 14 public hospitals from IDERA’s database, grouped into categories to analyze the results: five hospitals belonging to CAT I (high-risk healthcare centers with specialized intensive therapy), five hospitals to CAT II (high-risk healthcare centers with intensive therapy) and four hospitals to CAT III (low-risk healthcare centers with simple hospitalization). This categorization was available on the original database according to different levels of attention and services provided. The grouping of results was made considering that patients can choose to travel to any of the hospitals in every category to receive the same healthcare attention, as hospitals in the same category have similar services and equipment.

The function travel_time_matrix() is one of the six fundamental functions in the r5r package and was used to obtain a data table with the travel times estimates between census zones and hospitals. The package uses a specific extension of the routing algorithm RAPTOR that computes all Pareto-optimal journeys between each origin–destination pair, minimizing the arrival time and the number of transfers made between them (Delling et al., Reference Delling, Pajor and Werneck2015). The estimation is made by the schedules provided in the GTFS or by modeled schedules with a departure per minute during a time window if the GTFS is frequency-based. In both cases, a set of optimal paths is obtained in each round of the algorithm, with the difference that in the frequency-based GTFS, the schedules are randomly generated (Conway and Byrd, Reference Conway, Byrd and van der Linden2017).

The function also has arguments that can be set by the user, such as modes, departure date and time, time window, percentiles, maximum walking, biking, or by car time, maximum trip duration, fare structure, and max fare to consider monetary constraints, average walking or biking speed, among others. As selected in the previous Martinazzo and Falavigna’s (Reference Martinazzo and Falavigna2022) work, the analyzed time window was between 9 and 11 in the morning, and the departure date was set on an April Wednesday in 2022 and 2023. The maximum walking time was 10 min, equivalent to the 833 m walk used in the cited work, the maximum trip duration was 120 min, and the percentile selected was 85, meaning 85% of trips between every origin and destination pair are shorter than the obtained result.

The parameters were set to replicate an average day in an off-peak hour during the morning, as most of the facilities’ attention concentrates during the morning and weekdays. Both GTFS used had different services for weekdays, Saturdays and Sundays, so the date selected is an average weekday to represent more accurately an average trip. According to the last mobility survey, 54.6% of health-related trips had departure time before 11 a.m., more than 85% of health-related trips had a perceived travel time lower than an hour, and almost all of them were less than 90 min long (De Beláustegui, Reference De Beláustegui2011). A sensibility analysis of the departure time was performed to evaluate the selected time, and the results are shown in the next section. The maximum trip duration of 120 min was set to contain all trips beyond the average value of the mobility survey. The maximum walking time selected, which includes the time to access and egress public transport, is consistent with previous work in Córdoba about access to mobility services, which considers a 5-min walk appropriate to access a bus or tram stop (Martinazzo, Reference Martinazzo2022).

The resultant travel time matrix was processed to calculate an average per hospital category per zone. This way, in the grouped per category data frames obtained, each row corresponds to only one census zone with a single travel time value identified by its code, with additional information about the number of total households and the number of households with NBI. The percentage of households with NBI (%NBI) was calculated and used to classify the households into weighted quintiles per number of households so that the zones were classified accounting for 20% of households each.

The cumulative opportunities indicator was also calculated using r5r (Pereira et al., Reference Pereira, Saraiva, Herszenhut, Braga and Conway2021) for a maximum travel time of 60 min to complement the analysis of travel times from Martinazzo and Falavigna’s (Reference Martinazzo and Falavigna2022) work. The time frame was selected considering that, in Córdoba, over 95% of the trips last less than an hour (De Beláustegui, Reference De Beláustegui2011). The accessibility() function was used, setting the step decay function, a maximum walking time of 10 min, a maximum trip duration of 120 min, and the percentile selected was 85, as it was set for the travel time matrix.

To complete the analysis of travel times and evaluate how changes in the location of facilities during 2023 impacted the accessibility to healthcare, the minimum travel time per hospital category per zone was also obtained from the original 2023 travel time matrix. r5r (Pereira et al., Reference Pereira, Saraiva, Herszenhut, Braga and Conway2021) was also used to obtain a new travel time matrix considering the locations of three new facilities called “Hospitales de Pronta Atención” that were projected by the Municipality, which information was retrieved from the open data Municipality webpage Gobierno Abierto (Municipalidad de Córdoba, 2023a).