Did saving mothers, giving life expand timely access to lifesaving care in Uganda? A spatial district-level analysis of travel time to emergency obstetric and newborn care

Introduction: Interventions for the Saving Mothers, Giving Life (SMGL) initiative aimed to ensure all pregnant women in SMGL-supported districts have timely access to emergency obstetric and newborn care (EmONC). Spatial travel-time analyses provide a visualization of changes in timely access. Methods: We compared travel-time estimates to EmONC health facilities in SMGL-supported districts in western Uganda in 2012, 2013, and 2016. To examine EmONC access, we analyzed a categorical variable of travel-time duration in 30-minute increments. Data sources included health facility assessments, geographic coordinates of EmONC facilities, geolocated population estimates of women of reproductive age (WRA), and other road network and geographic sources. Results: The number of EmONC facilities almost tripled between 2012 and 2016, increasing geographic access to EmONC. Estimated travel time to EmONC facilities declined significantly during the 5-year period. The proportion of WRA able to access any EmONC and comprehensive EmONC (CEmONC) facility within 2 hours by motorcycle increased by 18% (from 61.3% to 72.1%, P < .01) and 37% (from 51.1% to 69.8%, P < .01), respectively from baseline to 2016. Similar increases occurred among WRA accessing EmONC and CEmONC respectively if 4-wheeled vehicles (14% and 31% increase, P < .01) could be used. Increases in timely access were also substantial for nonmotorized transportation such as walking and/or bicycling. Conclusions: Largely due to the SMGL-supported expansion of EmONC capability, timely access to EmONC significantly improved. Our analysis developed a geographic outline of facility accessibility using multiple types of transportation. Spatial travel-time analyses, along with other EmONC indicators, can be used by planners and policy makers to estimate need and target underserved populations to achieve further gains in EmONC accessibility. In addition to increasing the number and geographic distribution of EmONC facilities, complementary efforts to make motorized transportation available are necessary to achieve meaningful increases in EmONC access. The 4 SMGL-supported districts in Uganda—Kabarole, Kamwenge, Kibaale, and Kyenjojo—form a contiguous unit in the western region of the country. Among the combined 2017 population of just over 2 million were an estimated 538,706 women of reproductive age (WRA) aged between 15 and 49 years (Table 1).19 Population density is low, with over 78% of the 4-district area designated as rural and the largest urban population residing in Kabarole.34,35 Demographic Factors, SMGL-Supported Districts in Uganda, 2016 Abbreviations: km2, kilometers squared; SMGL, Saving Mothers, Giving Life; WRA, women of reproductive age. Transportation challenges are common in the SMGL-supported districts. The topography is mountainous, particularly in Kibaale district. Large national parks are mostly impassable forest and rugged terrain, and numerous rivers and lakes create geographic barriers (Figure 1). Only a small portion of the rural road network is passable by 4-wheeled vehicles, and only 2 paved roads connect Kamwenge and Kyenjojo districts with Fort Portal town, the district capital of Kabarole. Kibaale district did not have any paved roads during the SMGL implementation period. Visual Representation of Data Sources Used in Health Care Accessibility Modeling Analysis Abbreviation: WRA, women of reproductive age. The measurement of EmONC functionality used in our analysis was based on facility performance of a core set of lifesaving interventions, known as “signal functions,” in the 3 months prior to the health facility assessments (HFAs).36 EmONC facilities are defined as having the ability to, at a minimum, (1) administer parenteral antibiotics, (2) administer uterotonic drugs for active management of the third stage of labor and prevention of postpartum hemorrhage, (3) use parenteral anticonvulsants for the management of pre-eclampsia/eclampsia, (4) perform manual removal of placenta, (5) perform removal of retained products, (6) perform assisted vaginal delivery, and (7) perform neonatal resuscitation. CEmONC facilities have the additional capability to perform cesarean deliveries and blood transfusion.15 Although the Ugandan Ministry of Health has further mandates about the distribution of government facilities,37 our analysis applies WHO benchmarks for EmONC and CEmONC of at least 5 EmONC facilities, including at least 1 CEmONC facility, per 500,000 population. SMGL-supported facilities include those added to the study area during the SMGL initiative as well as existing facilities that were upgraded to provide EmONC. EmONC functionality was based on facility performance of a core set of life-saving interventions, known as “signal functions.” The initiative employed HFAs and other monitoring and evaluation methods to assess the progress and impact of interventions across the SMGL's implementation phases: Phase 0 (pre-implementation planning in 2011–2012), Phase 1 (June 2012 to December 2013), and Phase 2 (January 2014 to October 2017). To assess changes in facility infrastructure, functionality, and use, SMGL implementing partners in Uganda conducted HFAs in SMGL-supported districts at baseline, the end of Phase 1, and endline (November 2016).19 The 3 assessment periods were conducted in 111, 127, and 129 health facilities, respectively, which provided over 95% of all facility deliveries in the SMGL study area at each time point. HFAs documented performance of EmONC functions during the 3 months prior to the assessments as well as the geographic location of facilities (accuracy of ±10 meters).37 Our analyses used all facilities included in HFAs for any of the 3 assessment periods. SMGL-supported health facilities provided care for over 95% of all facility deliveries in the SMGL study area at each time point. We used land cover data obtained from the Regional Centre for Mapping of Resources for Development,33 initially collected with a 30-by-30 meter resolution and subsequently aggregated within AccessMod version 5, revision 5.1.18 (WHO, Geneva, Switzerland,) to a 92-by-92 meter resolution to match the resolution of other layers (Figure 1). This land cover raster used a 6-ecosystem scheme that accounted for forestland, grassland, cropland, settlement, wetlands, and other land cover. We created updated shapefiles for lakes and rivers using Uganda Bureau of Statistics (UBOS) data35 and OpenStreetMap (www.openstreetmap.org) shapefiles. When a data source had incomplete information about a river network, we manually digitized our master river shapefile with Digital Globe EnhancedView Web Hosting Service (https://evwhs.digitalglobe.com/myDigitalGlobe) satellite imagery obtained in June 2018. We considered water bodies as being completely impassable and rivers as being partially passable, if crossed by a primary or secondary road, by an assumed bridge. We merged and cleaned the shapefiles of the UBOS road network data from the 2014 Uganda Census and the OpenStreetMap road network digitized in mid-December 2017 via the Humanitarian OpenStreetMap Team.38,39 We created subsets of all primary (between district capitals) and secondary road shapefiles (between towns and major villages) for use in the AccessMod analysis (Figure 1). The resulting road shapefile was cross-checked against Digital Globe satellite imagery. We ascertained the proportion of roads that were paved and unpaved and changes in paving that occurred over time. Since the majority of the roads were unpaved and no substantive changes occurred within the project duration, we applied travel speeds for unpaved roads only to yield the most conservative travel time estimates. Elevation and slope data were obtained from the Shuttle Radar Topography Mission digital elevation model produced by the U.S. Geological Survey, both with a 92-by-92 meter pixel resolution.40 The model provided elevation information to the tool and was used to determine the relative slope of each raster pixel. National parks, from the World Database on Protected Areas, were derived by the United Nations Environment World Conservation Monitoring Centre and considered impassable unless a road passed through it. They were included in the final maps to provide context.41 Within AccessMod, the land cover, road network, river, and water body datasets were combined into a merged land cover raster file, with a 92-by-92 meter resolution, and used in the travel-time analyses. Household population data from all villages in the 4 Ugandan districts were collected in 2017 as a component of the SMGL Reproductive Age Mortality Study (RAMOS).19 While RAMOS's primary aim was to measure and identify main causes of maternal mortality, the study also enumerated households, household members, WRA, and all recent deaths.19 We cross-checked geographic coordinates collected in the 2017 RAMOS with UBOS geographic data and reconciled discordant coordinates.34,35,42 Overall, 538,706 WRA resided in 3,749 villages across the 4 districts in 2016 (Figure 1). To assess whether districts met the WHO benchmark of EmONC availability, we followed the WHO guidelines, which recommend a minimum of 5 EmONC facilities per 500,000 population, including at least 1 CEmONC facility in each district.37 For each district, we calculated the recommended number of EmONC facilities by dividing the estimated district population by 100,000. For each time period, we then computed the observed number of EmONC facilities and compared them to the recommended number of facilities. We estimated the minimum travel time to the nearest EmONC and CEmONC facilities using the AccessMod Accessibility module. AccessMod uses the least-cost path algorithm to calculate the quickest way of traveling between 2 points, using roads or off-road travel, as appropriate.43 Travel time is also dependent on travel speeds for each transportation mode—walking, bicycles, motorcycles, and 4-wheeled vehicles—with land cover influencing the speed of walking. Bicycles and motorcycles can be outfitted with sidecars as makeshift ambulances.43,44 We determined these speeds using direct observation combined with other published sources.34,43,46–51 Walking was the only mode of travel used for areas without primary or secondary roads. Speeds were reduced by two-thirds to account for slower transportation speeds of pregnant women and to further account for travel on unpaved roads. Tobler's function, which corrects walking speed based on the direction of slopes on the terrain derived from the digital elevation model, was used to adjust both walking and bicycling speeds.52 AccessMod uses the least-cost path algorithm to calculate the quickest way of traveling between 2 points, using roads or off-road travel, as appropriate. We performed AccessMod travel-time simulations for the 4 transportation modes to EmONC and CEmONC facilities, focusing on a 2-hour upper limit of the estimated travel time, consistent with WHO recommendations for EmONC access.15 All estimated transportation modes, except walking, assumed access to the nearest road by foot and travel by an immediately available vehicle to the closest facility providing EmONC care. We did not consider district boundaries as barriers to movement; however, we only estimated access to EmONC facilities within the SMGL-supported districts, allowing for movement between districts but not to facilities outside these districts. With ArcGIS Desktop version 10.3.1 (Environmental Systems Research Institute, Redlands, CA), we created a continuous distribution of estimated travel time needed to reach an EmONC facility for each transportation mode and categorized the continuous travel-time raster into 4 incremental 30-minute travel-time zones (0 to 30 minutes, 31 to 60 minutes, 61 to 90 minutes, and 91 to 120 minutes), plus a fifth category for more than 2 hours of travel time (>120 minutes). Instead of using AccessMod’s native Zonal Statistics module, we converted the raster into a shapefile of different travel-time zones in ArcGIS version 10.5. We mapped all travel-time zones to reach any EmONC and CEmONC services for each transportation mode. Combining the travel-time zones with georeferenced village population data, we estimated the number and proportion of WRA with access to EmONC and CEmONC services within each travel-time zone. We obtained the proportion of WRA within a travel-time zone by summing all WRA residing in villages located within each travel-time zone then dividing by the complete enumerated WRA population. We defined “adequate EmONC access” as the ability to reach an EmONC facility within 2 hours of travel time, and “poor EmONC access” as the inability to reach an EmONC facility within 2 hours. We assumed all travel to be from a woman’s home to a facility. We calculated the relative percentage change in the proportions of WRA residing within each travel-time zone and across each transportation mode, by subtracting the baseline percentage from the endline percentage and dividing by the baseline percentage. For the population percentages, z scores, based on the normal approximation to the binomial distribution, were used to calculate P values. The study protocol was reviewed and approved by recognized ethics committees in Uganda and complied with Ugandan Ministry of Health procedures for protecting human subjects. This study was reviewed and approved by the U.S. Centers for Disease Control and Prevention’s Center for Global Health Human Subject Review Board, which determined that it did not constitute human subjects research.