Geographic accessibility to public health facilities providing tuberculosis testing services at point-of-care in the upper east region, Ghana

Figure 1 shows schematically the various data and modelling components used to achieve the objective of this study. The methodological aim of this study was to model geographical access of patients to health facilities providing TB testing services in the UER, Ghana. The region with 13 administrative districts is in the north-eastern corner of Ghana, bordered by Burkina Faso to the north, Togo and Upper West Region to the east, and Northern Region to the south. The region had 1,244,983 people in 2018 and is considered largely (79%) rural and scattered in dispersed settlements [10]. The main source of income for most of the population is farming.

The study spatially referenced or obtained locational data of all health facilities providing TB testing services via a prior survey aimed to assess the accessibility of POC diagnostic tests in the region [8], and the use of global positioning system (GPS). Thereafter, we applied the world geodetic system (WGS) Zone 30 North coordinate system to all spatial data in order to allow for results of spatial processes in a preferred unit of ‘meters’. Topographic data included roads, rivers and the digital elevation model (DEM). A previous survey in the region preceding this present study showed that motorized tricycle popularly known as “Moto king” was the most utilized public transport option [8]. We estimated travel time via road and paths/tracks based on the “Moto King” transport system. This involved recalibrating travel time per pixel (10 m by 10 m grid) for both roads and paths. This ultimately helped to estimate travel time from settlement areas of the population to health facilities offering TB testing for all the districts in the region.

Spatial and attribute (A-spatial) data on TB testing health facilities was collected and prepared by assigning one coordinate system to all the data to allow for synchronization. We then loaded the data in google earth to determine and confirm their location via landmarks and other reconciliatory features. The data was then imported into ArcGIS in the point shapefile format. The spatial data obtained primarily covered the ‘what’ and the ‘where’ that is, the clinic or hospital name and location (coordinates). Besides the name, other attribute data included the district where the facility is located (district name), the type of health clinic (health center, hospital, and others), whether they provided TB testing services and the commonly used public transport option. A cross-section survey prior to this study helped eliminate public health facilities that do not provide TB testing services from the analysis [8]. These were filtered out using the select by attribute function and eventually clipped.

This data covered digitised and georeferenced road networks and paths well as features like rivers and lakes. The DEM of the region was also obtained to help identify natural barriers such hills and valleys as well as undulating land that would inform the decision on estimating travel time. All data was obtained from Adu Manu Kwame (AMK) Consult and juxtaposed with data obtained from the University of Ghana Remote Sensing and Geographic Information Systems laboratory to validate accuracy. The DEM was then reclassified into highlands (more than 200 m high) and flatlands (between 119 and 200 m high). This was determined by the DEM data which showed that the highest point in the region was about 470 m whiles the lowest point identified as 119 m.

This was accomplished using a sequence of algorithms with the final output incorporating cost distance tool in ArcGIS 10.4. Cost distance calculates the shortest time to a source based on a cost dataset. To realize this, a cost surface algorithm was designed with these parameters: a grid cell of size 10 m was assigned to the spatial features and values were then assigned to the predetermined grids. Roads were assigned low values because traveling on roads is faster than travel via paths and travel over impediments such as water bodies and rocky areas takes a much longer time and hence assigned higher values. Since cost distance requires the cost surface dataset and the source, the raster dataset served as the cost surface dataset and health facilities providing TB testing served as the source for calculating the cost distance. The output is a map showing shortest travel time (cell by cell) from any point in the map to a health facility providing TB testing services in the region. Algorithms which allowed for carrying out conversion of data from vector to raster, map algebra (cost surface models) and cost distance was developed using Python 2.7. Designing the algorithm helped to avoid repeating the various processes for the varying districts via the ArcMap toolbox. Based on prior information on the most commonly used public transportation option in the UER, the travel speed of a motorised tricycle was pecked at 20 km/h. This served as a guide for determining travel time in the region. A detailed description of the model is presented in Additional file 1.

The Oakwood Ridge National Laboratory Landscan data set serves as a standardized and globally accepted population distribution data [11]. This data was mapped in MapInfo and an SQL query was employed in clipping out the population distribution within all the available travelling distances from health facilities offering TB testing.

The primary outcome of this study was geographical accessibility to health facilities offering TB testing in UER, Ghana. Geographical accessibility was measured as: 0-10 km = good geographical access; and > 10 km = poor geographical access. Considering ≤10 km as good geographical accessibility be arbitrary. However, evidence shows that access to healthcare beyond 10 km is associated with higher risks of adverse outcomes [12]. Also, categorising geographical accessibility using travel time would have been useful but, travel time depends on the mode of transportation option and the route hence our choice to base our categorisation of geographical accessibility on distance. The secondary outcome of this study was to identify locations for targeted improvement of TB testing services in the UER utilizing geographical models and the application of remote sensing through satellite imagery analysis. Remote sensing also known as earth observation, refers to acquiring information about objects located on the earth’s surface without being in direct contact with the said object. The application of remote sensing through satellite imagery afforded us the opportunity to observe locations that had settlements and help us properly cite proposed health clinics without necessarily having to visit the region.

Spatial autocorrelation tool or Global Moran’s I [13] was run in ArcMap 10.4.1 to determine the spatial distribution of the health facilities providing TB diagnostic services in the UER. The Global Moran’s I statistic was run with the region as the input feature class and the count of health facilities providing TB testing services in each district as the input field to determine the spatial distribution of the health facilities. Z-scores and p-values were generated and reported. Mean and standard deviation (SD) for distance and travel time were also reported.

In all, 10 public health facilities providing TB testing services at POC existed in the region This represented an estimated population to health facility ratio of 125,000 people per facility in the UER. Of the 10 facilities, 9 (90%) were owned by the Ghana Health Services, and the remaining one (10%) owned by the Presbyterian church, a member of the Christian Health Association of Ghana. Majority (60%) of the health facilities providing TB testing services in the region were in districts with a total population greater than 100,000 people. Two (20%) of the 10 facilities were in the Bawku West district, and one (10%) each in the remaining 8 (69%) districts and municipalities in the region. Four districts (Builsa South, Nabdam, Pusiga, and Binduri) did not have any health facility providing TB testing services located in the districts. Table 1 shows the population distribution per public health facility providing TB testing services in the UER based on the sum of health facilities providing the service in each district.

The spatial autocorrelation analysis was carried out to determine how spatially distributed the health facilities providing TB testing services are, in the UER. The result shows that the health facilities were dispersed (z-score = − 2.3; p = 0.02), as illustrated in Fig. 2.

Figure 3 shows 13 modelled surfaces estimating distances values from the population to existing health facilities offering TB testing services in the UER. The maps provide visual prominent indications of the degree of variations in levels of geographical accessibility in relation to distance in the region. Of the 13 districts, 4 (31%) districts namely; Nabdam, Binduri, Builsa South, and Pusiga did not have any facility offering TB test. Suspected TB patients residing in Pusiga district would have to travel to the nearest health facility (Bawku Presbyterian hospital) which is situated in neighboring Bawku municipal for TB testing services. Suspected TB patients living in Binduri district however are bounded by four health facilities providing TB testing services: to the West by Zebilla Hospital (Bawku West), to the South-West by Binaba Health Centre (Bawku West), to the South-East by Garu Health Centre (Garu-Tempane), and to the North-East by Bawku Presbyterian Hospital (Bawku municipal). Nabdam district is also bordered by four districts (Bolgatanga Municipal, Bawku West, Talensi and Bongo) which have health facilities providing TB testing services and suspected TB patients referred from PHC clinics for testing would have to travel to the nearest TB facility. Patients resident in the Northern part of Talensi would be better served (by a short distance) if they opt to visit Zebilla hospital in the Bawku West district as the estimated distance according to the cost distance algorithm would be generally between 20 to 25 km. Patients residing in the central and southern part of Nabdam would have the most difficulty in reaching any health facility providing TB testing services partly due to lack of good roads to allow for faster travel and hence reduced travel distance. The spatial analysis reveals that all TB suspected patients’ resident in the central and southern part of Nabdam district are at least 30 km away from the nearest health facility providing TB testing. The closest referral health facility offering TB testing services to people living in the Builsa South is Sandema hospital which is averagely 30 km from all residential areas in the district.

Table 2 shows the distance and travel time distribution per population in the UER. Cumulatively, the spatial analysis shows that just about 38% of the population residents in the UER have good geographic access (less than 10 km) to a health facility providing TB testing services. Majority (62%) of suspect TB patients referred from PHC clinics for TB testing in the region however will have to travel more than 10 km to access a facility for the service. The mean distance ± SD to the nearest public health facility providing TB testing services in UER was 33.2 km ± 13.5. Whilst the mean travel time ± SD using a motorized tricycle speed of 20 km/h to the nearest facility providing TB testing services in the UER was 99.6 min ± 41.6. The longest mean distance and travel time was recorded in the Builsa South district that is, 53.8 km ± 13.1 and 161.4 min ± 40.2 respectively. Bawku Municipal recorded the shortest distance and travel time to the nearest facility providing TB testing services in the region that is, 21.1 km ± 8.3 and 63.3 min ± 25.7 respectively, as illustrated in Table 3.

To improve geographic accessibility to TB testing services in the region, this study proposed that there should be at least one health facility providing TB testing services located not more than 10 km from where people live. Base on this, the study employed geographical models to estimate 51 additional healthcare facilities for targeted improvement to meet the unmet TB diagnostic needs of the people resident in the UER. Figure 4 shows a map of the proposed locations of health facilities for targeted improvement to increase geographic access to TB diagnostic service in the UER.

To ensure that all the 51 proposed health facilities would not be in an uninhabited area (example, in the forest), remote sensing (satellite imagery) was first done to map out settlement zones and the areas within 10 km range of the already existing public health facilities providing TB testing services in the UER. An application of the satellite imagery analysis helped to reduce the initially identified 97 health facilities to 51 for targeted improvement to ensure geographical accessibility (less than 10 km) to TB testing services for all residents in the region. The remote sensing classification method of unsupervised classification was used to determine settlement and non-settlement areas (forests). In the supervised classification method, pixels are placed into clusters based on their digital numbers. For example, pixels that capture forest areas are similar and as such, the software classifies all such pixels into a cluster of forest area. The resulting classes were converted from raster to vector data. Using this, a clip function was performed to remove all proposed health facilities that fell in forest areas. Due reduction was achieved after deleting all proposed health facilities falling outside settlements leaving those critical to meeting the unmet diagnostic needs of populations reciting in poor geographical access areas. Figure 5 shows a map of residential zones and areas within 10 km coverage by the existing public health facilities providing TB testing services in the UER.