Hospital selection and definitions
All government national, tertiary, regional, and district hospitals in South Africa were identified from the South African National Department of Health (NDoH). Each health district and its corresponding DH had defined district boundaries. While patients could attend other facilities in acute emergencies, we assumed that district boundaries would likely be followed for the majority of surgical referrals given defined referral pathways from PHC and CHC to a DH.
As per the LCGS’ definition of 2HA, data for all three bellwether procedures performed at DH was not readily available in South Africa. Therefore, a surgical district hospital (S-DH) was defined by the presence of a functional operating theatre, a surgical provider, and provision of at least one caesarean section (CS) annually. CS data is routinely collected by NDoH for DH facilities and was obtained for 2015-16. Data for the presence of a functional operating theatre and surgical provider was done through telephonic surveys to facility managers by one of the authors (AD) between 2015-2016.(13, 14)
GPS locations
Geographical Positioning Satellite (GPS) coordinates for hospitals were obtained from the National Institute for Communicable Diseases. GPS coordinates were reviewed using logical checks and compared with NDoH datasets. Discrepancies were manually checked using a combination of Landsat images, Google Maps, Google Street View, telephone calls to facilities, and metadata from photographs.
Population data
The 2014 population estimates for 103,576 Enumeration Areas for South Africa were obtained from the Environmental Research System Institute (ESRI, Redlands, CA) IDEAL dataset.
Road network data
We obtained road map data for South Africa from the OpenStreetMap (OSM) project (http://www.openstreetmap.org/). Road speed limits from OSM were utilized, where available, to calculate travel time impedance. Where OSM road speed limits were not available, travel speeds of 110 kilometers per hour (km/h) were assigned to highways, 100km/h to regional roads, 80km/h to regional secondary roads, 60km/h to local roads, and 50km/h to unclassified roads and tracks in keeping with standard OSM algorithms. Standard OSM modifications for road surface (e.g. gravel/dirt=speed/2); and road smoothness (e.g. horrible=speed/2) were incorporated. The road network was compiled in ArcMap (version 10.3) and identified errors manually corrected.
Spatial analysis
2HA service areas for all hospitals were estimated using detailed non-overlapping polygons in the service area tool in ArcMap (version 10.3). Since the off-network travel time to the nearest road was not directly modelled, high and low 2HA estimates were generated for each analysis. Low 2HA estimates were generated by trimming the 2HA service area polygons to within one kilometer of the outer network edges, whereas high 2HA estimates were not trimmed resulting in larger service areas. Mask area weighting, incorporating mesozones with population counts of less than five people as mask areas, was used to estimate the proportion of the population residing within the 2HA service areas.