Menu
Menu
Menu
Menu
Background: Access to transportation is vital to reducing the travel time to emergency obstetric and neonatal care (EmONC) for managing complications and preventing adverse maternal and neonatal outcomes. This study examines the distribution of travel times to EmONC in Kigoma Region, Tanzania, using various transportation schemes, to estimate the proportion of live births (a proxy indicator of women needing delivery care) with poor geographic access to EmONC services. Methods: The 2014 Reproductive Health Survey of Kigoma Region identified 4 primary means of transportation used to travel to health facilities: walking, cycling, motorcycle, and 4-wheeled motor vehicle. A raster-based travel time model was used to map the 2-hour travel time catchment for each mode of transportation. Live birth density distributions were aggregated by travel time catchments, and by administrative council, to estimate the proportion of births with poor access. Results: Of all live births in Kigoma Region, 13% occurred in areas where women can reach EmONC facilities within 2 hours on foot, 33% in areas that can be reached within 2 hours only by motorized vehicles, and 32% where it is impossible to reach EmONC facilities within 2 hours. Over 50% of births in 3 of the 8 administrative councils had poor estimated access. In half the councils, births with poor access could be reduced to no higher than 12% if all female residents had access to motorized vehicles. Conclusion: Significant differences in geographic access to EmONC in Kigoma Region, Tanzania, were observed both by location and by primary transportation type. As most of the population may only have good EmONC access when using mechanized or motorized vehicles, bicycles and motorcycles should be incorporated into the health transportation strategy. Collaboration between private transportation sectors and obstetric service providers could improve access to EmONC services among most populations. In areas where residents may not access EmONC facilities within 2 hours regardless of the type of transportation used, upgrading EmONC capacity among nearby non-EmONC facilities may be required to improve accessibility.
Kigoma Region, which covers 45,066 square kilometers, is situated in the northwest corner of Tanzania and borders the Democratic Republic of the Congo, Burundi, and Lake Tanganyika (Figure 1). The region’s land cover consists of grassland (34%), cropland (8%), forest (34%), and water (14%), while the remaining area consists of human settlements and other terrains. Topographically, there is a wide range of altitudes in Kigoma Region; the lowest area is 800–1,000 meters above sea level, along Lake Tanganyika, while the highest areas are in the northern and southern highlands (1,500–2,400 meters above sea level).8 The Luiche, Malagarasi, and Ruchuigi rivers originate from the northern and northeastern highlands and move southward before draining west into Lake Tanganyika. Delivery Health Facilities, by Emergency Obstetric and Neonatal Care Status, Included in the Study, Kigoma Region, Tanzania, 2013 Abbreviations: BEmONC, basic emergency obstetric and neonatal care; BEmONC-1, partially functional BeMONC facility (i.e., BEmONC without assisted vaginal delivery); CEmONC, comprehensive emergency obstetric and neonatal care; CEmONC-1, partially functional CEmONC facility (i.e., CEmONC without assisted vaginal delivery); EmONC, emergency obstetric and neonatal care. In 2012, the region had a reported population of 2,127,930, of which an estimated 470,000 (22%) were women of reproductive age (i.e., 15–49 years).9 Kigoma Region consists of 8 administrative councils (Buhigwe, Kakonko, Kasulu, Kasulu Township Authority, Kibondo, Kigoma Rural, Kigoma Municipal-Ujiji, and Uvinza).10 It is characterized by its rurality (83% of surveyed households11), high birth rates (210 births per 1,000 women aged 15–44 years12), and relatively high maternal mortality (222 maternal deaths per 100,000 facility-based live births in 201313). In 2014, only 47% of the deliveries in Kigoma Region were attended by a skilled birth attendant.12 This analysis aims to estimate the minimum amount of time required for populations to travel to the nearest EmONC facility in Kigoma Region when seeking care, by using AccessMod version 4.0 (World Health Organization [WHO], Geneva, Switzerland), an add-on analytic extension to ArcGIS 9.3.1.14 This travel time modeling program uses a least-cost path (friction surface) approach to produce a raster layer across the target area, where each gridded cell represents the minimum travel time from the cell’s location to the target destination. Spatial raster data models are representations of continuously varying attributes in which the surface of the earth is divided into uniformly spaced pixels of cells, and each pixel carries one or more attributes of a value for that location. Raster data differ from vector data models (e.g., points, lines, and polygons) when representing spatially continuous processes and interactions such as distance, terrain, and travel time. This analysis aims to estimate the amount of time required for populations to travel to the nearest EmONC facility in Kigoma Region. The travel time model requires the following data inputs: (1) geographic coordinates of the health facilities providing EmONC services; (2) combined land cover raster dataset; (3) digital elevation model (DEM) raster dataset; and (4) travel speed specification, based on land cover classification and transportation modes. Tobler’s function, which corrects walking speed based on the direction of slopes on the terrain, was used to adjust the anisotropic (directional) walking speed.15 Motorized transportation did not require directional speed adjustments. Unfortunately, due to the limitations of our software installation, we were unable to adjust for anisotropic changes in bicycling speed due to a reported bug in AccessMod version 4.0. Table 1 summarizes the data source, the spatial resolution, and other specification details of the input geospatial data used for modeling the travel time and estimating accessibility to existing EmONC services in the region. All input geospatial datasets were cropped to the administrative boundaries of Kigoma Region, as derived from the ward boundaries of the 2012 Tanzania census.10 All rasters used in the travel time analysis were specifically at 30-meters resolution, while the live birth raster data used in the live births analysis were kept at 100-meters resolution, as it was recommended not to resample that dataset. All data layers were projected into the spatial reference frame, WGS84/UTM Zone 35S. Characteristics of the Input Geospatial Datasets, by Data Layer, for Modeling Travel Time and Estimating Accessibility Coverage Among Women Needing Delivery Care in Kigoma Region, Tanzania a The 2010 Tanzania land cover scheme I was used to specify non-road land classes. b Roads are reclassified to reflect the classification scheme used by John Snow, Inc. and Medical Supply Department in the cross-country medical supply route analysis in 2014.18 c The layer was created using the mapping methodology described by Tatem et al. (2014).46 The health facility dataset was derived from a 2013 region-wide health facility assessment led by the U.S. Centers for Disease Control and Prevention (CDC) staff in selected Kigoma Region health facilities to document the functionality of EmONC infrastructure and EmONC-related human resources.16 During the same time frame, the CDC and project partners used the Pregnancy Outcomes Mortality Surveillance (POMS) system to document changes in facility-based deliveries in the region and to assess corresponding pregnancy outcomes in the region.13 Together, these 2 surveys documented evidence for the observed practice of the 9 essential medical services necessary for treating and managing maternal and neonatal complications (referred to as signal functions17), both at the facility level (through the health facility assessment) and at the individual delivery level (through POMS). Included in this study were 127 health facilities in the region (Figure 1), composed of hospitals, health centers, and large dispensaries providing delivery services (i.e., dispensaries experiencing more than 90 births per year), accounting for 97% of all facility deliveries in 2012 in Kigoma Region. A total of 11 facilities were found to be providing EmONC levels of care. Eight facilities (4 hospitals and 4 health centers) were identified as fully functional CEmONC facilities, performing 9 signal functions for EmONC within the 3 months before the assessment; 1 facility was identified as a partially functional CEmONC facility (CEmONC-1), performing 8 signal functions excluding the provision of assisted vaginal delivery (AVD).16 Two additional health centers were found to be partially functional BEmONC facilities (BEmONC-1), defined as performing 6 BEmONC signal functions, excluding AVD. Despite being unable to provide AVD, the 3 partially functional BEmONC or CEmONC facilities were found to have strong enough transportation referral networks to ensure their inclusion in the roster of EmONC facilities. The geographic coordinates used to locate the facilities in this study were recorded in the health facility assessment, using Garmin eTrex 30 devices with an accuracy of 3–5 meters. A road network dataset, obtained from OpenStreetMap, was reclassified to reflect the classification scheme used by John Snow, Inc. (JSI) and the Medical Supply Department (MSD) of Tanzania for their 2014 cross-country medical supply route analysis.18 The road classes included: (1) major roads; (2) major roads crossing residential areas; (3) secondary roads; and (4) local roads18 (road network visualized in Figure 1). Local roads were further divided into categories, based on road width and OpenStreetMap classifications, including the following: (1) car-passable roads; (2) motorcycle- and bicycle-passable roads (i.e., tracks that are passable to motorcycle and bicycle, but not cars); and (3) walking-only roads.19,20 Boating routes were digitized to allow for travel approximately 60 meters away from the shore of Lake Tanganyika. Docks were identified using Bing satellite imagery and were digitized to connect to boat travel routes. Both river and road network vector datasets, obtained from OpenStreetMap, were transformed to raster datasets consisting of 30-meter gridded cells, and were then overlaid on the land cover raster dataset.21 This created a combined land cover raster dataset with 13 unique land feature classes. The 6 non-road land cover classes included forest land, grassland, cropland, settlements, wetlands, and other land. For the purposes of analysis, wetlands and rivers were considered to be impassable to any form of transportation. To provide a mapped model of land elevation, we obtained Shuttle Radar Topography Mission (SRTM) digital elevation model data, at a spatial resolution of 30 meters (1 arc-second), from the U.S. Geological Survey.22 To simulate the use of various primary transportation modes in real life, we specified 4 different travel scenarios according to the primary transportation modes (i.e., walking, bicycle, motorcycle, or car) described in the 2014 Kigoma Reproductive Survey (RHS) conducted by the CDC: We specified 4 different travel scenarios according to whether the primary transportation mode was by walking, bicycle, motorcycle, or car. Boats were included for every scenario, as there are many villages around the lake that may use boats for at least part of the trip. A boat route was included as part of a journey if its inclusion resulted in an overall shorter travel time. Walking was also included in every scenario, because all residents would need to walk at some point during their trip to an EmONC facility, as they travel through land areas with low road coverage. To simulate real-life travel experiences in which travel time may vary by terrains, road types, and transportation used, our travel time computation module specified a transportation mode with a corresponding travel speed for every combined land cover class, under each travel scenario. Various sources were used to ascertain the transportation-specific travel speed for each land cover type in the dry season, as summarized in Table 2; the sources consisted of the Global Accessibility Map,23 AccessMod version 3.0 publication literature,24 a cost-surface analysis conducted in Dar es Salaam,25 WHO’s Tanzania road safety brief,26 a motorcycle analysis conducted in Hanoi, Vietnam,27 and a cost-distance analysis conducted in the Biliran Island, Philippines.28 Table 3 describes the travel speed for each land cover class for all 4 travel scenarios employed in this analysis. Source and Rationale for Travel Speed Specification per Land Cover Type in Kigoma Region, Tanzania Summary of the original speed data sources and rationales from which the traveling speeds were derived per transportation and landcover type, for constructing a traveling scenario table required by AccessMod version 4.0 analysis modules to estimate the travel time and accessibility to existing emergency obstetric and neonatal care services in Kigoma Region. Travel Speeds, per Land Cover Type, to the Nearest Emergency Obstetric and Neonatal Care Facilities in Kigoma Region, Tanzania, by Travel Scenario a Tobler’s function was used for correcting anisotropic movement. Wetlands and rivers were considered to be impassable for the purposes of this analysis (i.e., walking speed of 0), and were not included in this table. Live birth count was used as a proxy measurement for women needing delivery care, which is the target population of EmONC services. A 2012 projected live birth raster dataset for Tanzania was obtained from the WorldPop Project.29 The value of each 100-meter gridded cell represented the estimated number of live births that would have occurred in an area of 100 square meters in 2012. A raster layer that describes the minimum travel time required to reach the nearest EmONC facility was created using the AccessMod version 4.0 extension’s modules for each of the 4 travel scenarios. For this accessibility analysis, the upper limit of the estimated travel time was set at 2 hours, a conservative time frame consistent with the WHO recommendations for access to EmONC facilities.30 Therefore, “good geographic access to EmONC care” was operationally defined as a woman’s travel time to EmONC care being at, or under, 2 hours, while “poor geographic access to EmONC care” was defined as a woman’s travel time to EmONC care exceeding 2 hours. Each scenario-specific travel time raster layer was reclassified and converted into 4 incremental 30-minute travel time zones (up to 2 hours) as polygon vectors in ArcGIS 10.3. All 2-hour service catchment areas for each corresponding travel scenario were merged to show the distribution of areas with good EmONC service access (i.e., areas within which one can reach EmONC services in less than 2 hours) based on each of the primary transportation modes. To compute the region-wide proportion of live births in a travel time zone or service catchment, we divided the total number of live births within a travel time zone or service catchment by the total number of live births in Kigoma Region. In addition, the proportion of live births with poor access to EmONC under each travel scenario was calculated per administrative council for each travel scenario (i.e., Scenarios 1–4 and the “all-modes” scenario, where women may use any of the 4 travel scenarios to reach the nearest EmONC facility as necessary), by dividing the total number of live births located outside the 2-hour service catchment in an administrative council by the total number of live births in the entire council. The live birth figures involved in these calculations were aggregated from the 2012 live birth raster dataset in ArcGIS 10.3, collected to a 100-meter resolution by various travel time or 2-hour catchment polygon vectors using zonal statistics. The estimate for the number of all births per catchment presented in this analysis was the product of the total population of women aged 15–49 years in Kigoma Region in 2012,9 the annual population growth coefficient (for projecting the population of 2013),9 the age-specific fertility rate in the 2014 Kigoma RHS,12 and the proportion of births occurring in that catchment. This study was reviewed and approved by the CDC’s Center for Global Health Human Subject Review Board and was determined not to comprise human subjects research.
N/A
