Distance is “a big problem”: a geographic analysis of reported and modelled proximity to maternal health services in Ghana

Background: Geographic barriers to healthcare are associated with adverse maternal health outcomes. Modelling travel times using georeferenced data is becoming common in quantifying physical access. Multiple Demographic and Health Surveys ask women about distance-related problems accessing healthcare, but responses have not been evaluated against modelled travel times. This cross-sectional study aims to compare reported and modelled distance by socio-demographic characteristics and evaluate their relationship with skilled birth attendance. Also, we assess the socio-demographic factors associated with self-reported distance problems in accessing healthcare. Methods: Distance problems and socio-demographic characteristics reported by 2210 women via the 2017 Ghana Maternal Health Survey were included in analysis. Geospatial methods were used to model travel time to the nearest health facility using roads, rivers, land cover, travel speeds, cluster locations and health facility locations. Logistic regressions were used to predict skilled birth attendance and self-reported distance problems. Results: Women reporting distance challenges accessing healthcare had significantly longer travel times to the nearest health facility. Poverty significantly increased the odds of reporting challenges with distance. In contrast, living in urban areas and being registered with health insurance reduced the odds of reporting distance challenges. Women with a skilled attendant at birth, four or more skilled antenatal appointments and timely skilled postnatal care had shorter travel times to the nearest health facility. Generally, less educated, poor, rural women registered with health insurance had longer travel times to their nearest health facility. After adjusting for socio-demographic characteristics, the following factors increased the odds of skilled birth attendance: wealth, health insurance, higher education, living in urban areas, and completing four or more antenatal care appointments. Conclusion: Studies relying on modelled travel times to nearest facility should recognise the differential impact of geographic access to healthcare on poor rural women. Physical access to maternal health care should be scaled up in rural areas and utilisation increased by improving livelihoods. The 2017 Ghana Maternal Health Survey (GMHS), a special DHS, sampled 25,062 women aged 15 to 49 years from 900 enumeration areas representative at national and regional levels [17]. The GMHS used a two staged cluster sample design with rural/urban stratification to collect data about women’s experiences and use of maternal health services between June and October 2017, achieving a 99% response rate. The GMHS also recorded the geographic location of clusters. This study includes information on women aged 15–49 who were asked about their last birth in the 5 years before the 2017 GMHS survey. Ghana Health Service (GHS) data on the location of health facilities providing birthing services in 2017 and spatial topographic data were used to model travel times. Spatial data representing terrain, land cover, roads, rivers/water bodies and topography were included to model the travel time to health facilities [26–28]. Travel time to the nearest health facilities providing birthing services was modelled in Accessmod version 5 [29], a free tool for measuring access to health services. To measure the travel time to each health facility, we first created an impedance cost surface (a gridded layer representing the difficulty of travel) by combining landcover, terrain, roads, and water bodies (e.g. rivers and lakes). Where roads of different classes met, the road with the maximum speed was prioritised. We assumed that patients would walk on all land cover types, then use mechanised transport on roads. Therefore, walking speeds were 5 kmh− 1 or less in forests, woodlands and croplands and other landcover types depending on how easily they can be traversed [30]. Following traffic regulations in Ghana, primary, secondary and tertiary roads were assigned 90 kmh− 1, 50 kmh− 1 and 30 kmh-1, respectively. The model ensures that the terrain’s steepness affects travel speed towards a health facility for persons walking or bicycling. Via the impedance surface, we estimated the travel times from each GMHS cluster location to the nearest health facility where birthing services are provided. Birth counts from routine GHS health data indicate health facilities providing birthing services. The longitudes and latitudes for DHS clusters are intentionally randomly displaced within two kilometres in urban clusters and five or up to ten in rural areas for data protection reasons. Therefore, we used two and five-kilometre buffers to compute the median travel times around urban and rural clusters respectively, calculating medians because travel times had skewed distributions within these buffers. These recommended distance buffers mitigate the likely effects of cluster displacement [31]. The DHS asks respondents if distance is a big problem in accessing healthcare when sick. The binary response from this question was the main outcome studied. The DHS questionnaire asked women the following question: “Many different factors can prevent women from getting medical advice or treatment for themselves. When you are sick and want to get medical advice or treatment, is each of the following a big problem or not a big problem: The distance to the health facility?” Response options: 1. Big problem 2. Not a big problem [32]. Secondly, we assessed the effect of proximity and socio-demographic variables on skilled birth attendance (SBA). For SBA, a woman was assisted by a skilled attendant if the most qualified person during childbirth was a midwife, doctor or nurse. The predictor variables were the number of antenatal care (ANC) appointments and postnatal care (PNC) within 48 hours after birth. Other variables included were age, rural-urban, wealth, region, health insurance and education [33]. In this study, ANC and PNC were defined as skilled if the service provider was a midwife, doctor or nurse. When two or more providers are present, the highest qualified service provider is used to classify skilled ANC and PNC. Based on the older WHO recommendation, the number of skilled ANC appointments were recoded into three (no ANC appointments, one to three, four or more) [34]. Similarly, timely PNC was defined as any woman who received a health check or visits from a skilled provider within 48 hours after delivery, while in the health facility or at home following delivery. The women’s ages were grouped into three classes (15 to 20 years, 21 to 30 years, 31 to 49 years). Household wealth quintiles were collapsed into three classes (poor, middle, and rich). We combined the “poorest” and “second” into “poor” and “fourth” plus “wealthiest” as “rich”. The DHS used household assets, livestock, drinking water source, type of toilet, type of cooking fuel, and building structure to construct the wealth index via a principal component analysis [35]. The highest education attained were recoded as no formal education, primary, secondary, and higher education. Due to the high proportion of missing values for women covered by health insurance, registration with a health insurance scheme was used as a proxy for insurance cover. The estimated median travel time between groups was reported with inter-quartile range (IQR). For categories with two levels, the differences in travel time were tested with the Wilcoxon rank sum test, whereas the Kruskal Wallis test was applied to groups with three or more levels. Non-parametric tests were chosen because the travel time distribution was skewed. To test for association between the reported distance and the independent variables, chi square tests were used. The descriptive statistics were presented in tables and plots to visualise the difference between groups. Logistic regression models were used to estimate the relationship between reported distance problems and SBA, controlling for socio-demographic and maternal health characteristics. The skilled birth outcome was chosen because it is the key determinant of maternal health outcomes [7]. Crude odds ratios were estimated between skilled birth attendance and each independent variable. All independent variables in the crude model associated with SBA at 10 % significance level were added to the adjusted models to allow for associations that can be insignificant in the crude model but change in the presence of other variables [36]. We tested all other associations at 5 % significance. Furthermore, a logistic regression analysis was conducted to estimate the relationship between socio-demographic backgrounds and reported challenges with distance. The outcome variable in this model was the binary self-reported distance. The independent variables were the socio-demographic variables and modelled travel time. Multicollinearity for both models was checked with a variance inflation factor threshold set at ten. We used the likelihood ratio test to compare the models. Finally, we included survey weights to correct for sampling and non-response error.