The effects of geographical accessibility to health facilities on antenatal care and delivery services utilization in Benin: a cross-sectional study

Benin is located in West Africa on the Gulf of Guinea, and covers an area of 114.763 square km. Benin consists of 12 departments: Alibori, Atacora, Atlantique, Borgou, Collines, Couffo, Donga, Littoral, Mono, Ouémé, Plateau, and Zou [17]. These departments are divided into 77 municipalities and then subdivided into 546 districts [17]. In terms of maternal and neonatal healthcare, Benin has two national hospitals, six Departmental Hospital Centers, 28 Zone Hospitals, 12 other hospitals, 76 municipality maternity units and more than 825 district maternity units [18].

Benin is a country where the southern and northern parts belong to different geographical zones. Benin’s geographic gradient is well-marked from south to north. The southern part of the country is in the coastal zone, whereas the northern part is mountainous. The Atakora mountain chain culminates at 658 m above sea level. Regarding geographical accessibility within Benin, the main mode of transport used are two-wheelers (personal motorbikes and motorbike taxis) and personal cars for road transport [19]. The ownership rate of cars per household was low, at 4.0% in 2011/2012 and 3.9% in 2017/2018 [3]. On the other hand, the ownership rate of motorbikes per household was comparatively higher, increasing from 58.4% in 2011/2012 to 66.3% in 2017/2018 [3, 20]. Road infrastructure consists of “classified roads” (6076 km) and “rural roads” (23,000 km) [19]. “Classified roads” include national roads (3898 km) and interstate national roads (2178 km), but only 30% of “classified roads” are paved [19]. Regarding “rural roads”, only 1/3 (a total of 7400 km) are properly constructed [19]. The low percentage of asphalted roads in the country leads to the high ownership of intermediate means of transport among households, such as motorcycles, tricycles, animal-drawn carts, rickshaws, bicycles etc., which are flexible and efficient on rough roads [19, 21].

We used two latest cross-sectional data from the Benin Demographic and Health Survey (BDHS), 2017/2018 [3] and 2011/2012 [20]. Table 1 shows the summary of the surveys. Regarding the sampling design, in the first stage of BDHS 2017/2018, 555 primary sampling units (clusters) were drawn from the list of 12,633 enumeration areas [3]. In BDHS 2011/2012, 750 clusters were selected from 7352 enumeration areas [20]. In the second stage, 26 households and 24 households per cluster were selected in BDHS 2017/2018 and BDHS 2011/2012, respectively. BDHS 2017/2018 surveyed 14,156 households and 15,928 women [3], while BDHS 2011/2012 surveyed 17,422 households and 16,599 women [20]. We used data of 18,105 women, 8994 women from BDHS 2017/2018 and 9111 women from BDHS 2011/2012, who had live births within five years preceding the surveys as a study sample.

Our main explanatory variable is geographical accessibility. Geographical accessibility is defined as “the physical distance or travel time from the service delivery point to the user” [22]. In addition, several studies have treated the altitude of women’s residence as a significant aspect of geographical accessibility [23, 24]. Accordingly, our study analyzed distance, travel time, and the altitude of women’s residence as important aspects of geographical accessibility. We used the GIS module of the BDHS and the ArcGIS software to calculate these variables.

First, we used the “Euclidean distance (km)”—a straight-line distance between two points—to measure the distance from one’s residence to the closest health facility [25]. The advantage of using this method is that it can be generalized for other similar topography and cultural contexts in Sub-Saharan Africa [25]. Several previous studies used the Euclidean distance to estimate geographical accessibility to health facilities in Sub-Saharan Africa [12, 26, 27]. However, using straight-line distance to assess geographical accessibility is sometimes regarded as less accurate than applying travel distance because it is the simplest measure [28‐30]. In addition, Euclidean distance may lead to an underestimation of travel distance [26]. Despites these limitations, Euclidean distance is considered a valid measure of accessibility in both rural and urban areas in Sub-Saharan Africa [25, 31].

Second, we used the “road network distance” as a more realistic measure. It is the distance along road infrastructure from a residence to the closest health facility. It can be defined as the Euclidean distance from the residence to the road network, plus a distance from the road network to a health facility [25]. To estimate the road network distance, we used the Benin Road Infrastructure dataset from the World Bank website that contains shapefiles of all roads as of 2017 [32]. We also used the geolocation points of the clusters provided by the BDHS and those of health facilities from the World Health Organization (WHO) website [33].

Third, we calculated travel time (in minutes: by walking and via car) from a residence to the closest health facility. The walking time was calculated from the women’s residence to the closest health center, following paths and roads for pedestrians. Walking speed was set at 5 km/h. Considering the types of roads, the driving time was calculated from the women’s residence to the closest health center. Driving speed was set at 80 km/h for “classified roads” and 60 km/h for “rural roads.” We used Dijkstra’s algorithm of ArcGIS, which can estimate the driving time from the residence to the closest health center by using the fastest route.

Fourth, we examined the effects of the geographic altitude of a residence (in meters) on women’s utilization of maternal healthcare resources. It was assumed that it would be more difficult for women living in highland areas to reach health facilities as confirmed in Ethiopia [23]. We used ArcGIS to conduct the abovementioned analysis.

Finally, we analyzed whether the availability of means of transport at the community level was associated with maternal healthcare utilization. The BDHS contains questions about whether the household owned a bicycle or motorbike. Because bicycles and motorbikes are popular means of transport in Benin, they can be used to travel when seeking maternal healthcare at health facilities [34]. Since the ownership of motorbikes and bicycles at the household level seems less important than its availability, we used the ownership rates of them per cluster (community), which would reflect more realistic situation of Benin. Thus, we calculated the ownership rates of bicycles and motorbikes per cluster and used them as proxy variables for community-level availability of means of transport.

We applied multivariate logistic regressions to analyze the impact of distance to the closest health center on maternal healthcare utilization. Data analysis was performed using Stata version 14. Because BDHS applied a two-stage cluster sampling design, we used the svy (survey) commands of Stata to correct for unequal sampling probability, clustering, and stratification to calculate descriptive statistics and perform regression analysis. Additionally, we conducted propensity score matching (PSM) analysis to check the robustness of the logistic regressions. The PSM attempts to estimate the effects of a specific policy or treatment in observational studies by reducing the bias arising from confounding factors that might predict outcome variables. It matches treated and untreated units based on a set of basic characteristics and attempts to balance both groups. According to the National Health Development Plan (NHDP) 2009–2018 of Benin, geographical accessibility to healthcare services in Benin was defined as “the percentage of the population living within 5 km of the closest health center” [35]. Therefore, we created two groups, the treatment group (women living more than 5 km from the closest health center) and a comparison group (women living within 5 km from the closest health facility) to assess whether the threshold (5 km from the closest health center) had adverse effects on women’s use of appropriate maternal healthcare.

We used the following outcome variables: (1) whether the woman made at least one ANC visit during her latest pregnancy (“any ANC”); (2) whether the woman made four or more ANC visits during her latest pregnancy (“ ≥ 4 ANC”); (3) whether the woman used a health facility at the birth-delivery (“Facility delivery”); and (4) whether the woman was attended by a professional health worker (i.e., doctor, nurse, auxiliary nurse, or midwife at birth (“delivery by SBA”). Because all the outcome variables were binary, they were coded 1 if the mother had received appropriate healthcare (ANC, facility delivery, or SBA) during pregnancy or childbirth, and 0 otherwise.

We used mother-, household-, and community-level characteristics as control variables. Mother-level variables comprised age and educational achievement (no education, primary, secondary/higher). Household-level variables included religion of the household head (Muslim, Protestant, Catholic, Vodoum/other traditional, and No religion/others) and asset quintiles. Community-level variables included geographical zones of a residence (south or north of the country), as well as the ownership rates of bicycles and motorbikes.

Table 2 summarizes the characteristics of the study participants. After merging two rounds of BDHS, 18,105 women (49.9% from BDHS 2011/2012 and 50.1% from BDHS 2017/2018) and 26,996 births were included in this study. Regarding geographical zones, 55.7% of the sample (10,087 women) represented southern, while 44.3% (8018 women) represented the northern part of the country. Regarding outcome variables, 84.0% of the women received ANC at least once, and 54.8% received ANC four times or more. Of all the childbirth, 86.0% were delivered at health facilities, and 81.7% were assisted by SBA. These figures differed significantly between the geographical zones. Although 92.0% and 68.4% of the women received at least one ANC and four or more ANC in the South, only 73.3% and 36.8% of the women received the same care in the North. Regarding childbirth, 95.1% of the babies were born at health facilities and 92.0% were assisted by SBA in the South, but only 73.8% and 68.1% of the newborns received the same care in the North.

Regarding the main predictor, the mean straight-line distance from a residence to the closest health facility was 3.57 km in the total sample. The distance in the north (5.29 km) was twice as long as that in the South (2.26 km). Looking at as a distance dummy variable, 19.7% of the households lived more than 5 km away from the closest health center in the total sample. The percentage was almost five times higher in the North (36.7%) than in the South (6.8%).

Regarding mothers’ characteristics, their mean age was 29.3 years. Of all the mothers, 67.3% were not formally educated; only 15.1% had completed secondary education or higher. As for household characteristics, 28.6% of the household heads were Muslim, followed by Protestants (27.8%), Catholics (25.2%), Vodoum/other traditional religion (11.4%), and no religion/others (7.0%). In terms of community characteristics, the ownership rates of bicycles and motorbikes per cluster were 30.0% and 62.4%, respectively.

Table 3 present the results of the multivariate logistic regression analysis for antenatal care and delivery services utilization. The magnitude of the effects was assessed by odds ratio (OR), which can be interpreted as increasing (if OR > 1) or reducing (if OR < 1) the likelihood of women’s use of healthcare. Regarding the main predictor (straight-line distance to the closest health center in the Model 1), there was a statistically significant and negative effect of having any ANC, facility delivery, and SBA, indicating that the longer the distance to the closest health center, the less likely women are to receive the necessary healthcare during pregnancy and childbirth. The estimated ORs for any ANC (OR = 0.958, p < 0.001) indicated that if the straight-line distance to the closest health center increased by one km, the odds of a woman receiving ANC at least once was reduced by 0.042. Similarly, the estimated ORs for facility delivery (OR = 0.908, p < 0.001) and SBA (OR = 0.882, p < 0.001) suggest that one-kilometer increase in straight-line distance decreases the odds of delivering a baby at a health facility by 0.092 and reduces the odds of being assisted by SBA by 0.108. However, these results differed between geographical zones. In the South, a statistically significant and negative effect of distance on the use of maternal healthcare was not confirmed. On the other hand, distance had statistically negative effects on women’s uptake of ANC (OR = 0.955, p < 0.001), facility delivery (OR = 0.894, p < 0.05), and SBA (OR = 0.879, p < 0.001) in the North. Table 3 shows statistically significant and negative effects of travel distance (Model 2), travel time by walk (Model 3) and travel time by car (Model 4) on women’s uptake of any ANC, facility delivery, and SBA.

With respect to the other explanatory mother and household level variables, mothers’ educational attainment and household assets were consistently associated with a higher likelihood of utilizing ANC, facility-based delivery, and SBA. Regarding religion, women whose heads of household are Protestant and Catholic were more likely to receive appropriate MCH care services compared to Muslims. As for the community-level variables, the ownership rate of motorbikes was associated with a higher likelihood of having facility-based delivery and SBA in the North.

Table 4 presents the PSM results. In the total sample, the treatment dummy variable had a statistically significant and negative effect on receiving any ANC (OR = 0.952, p < 0.001), facility delivery (OR = 0.69, p < 0.001) and delivery by SBA (OR = 0.973, p < 0.001), indicating that women who lived 5 km or more from the closest health center were less likely to use these services compared to the comparison group. The negative effects of distance on receiving any ANC (OR = 0.943, p < 0.001), and delivery by SBA (OR = 0.973, p = 0.008) were confirmed in the southern part of the country. In north, the distance had statistically and negative effects on women’s uptake of ANC (OR = 0.965, p = 0.003), facility delivery (OR = 0.957, p < 0.001), and SBA (OR = 0.969, p = 0.003).