Geographical access to care at birth in Ghana: A barrier to safe motherhood

Background: Appropriate facility-based care at birth is a key determinant of safe motherhood but geographical access remains poor in many high burden regions. Despite its importance, geographical access is rarely audited systematically, preventing integration in national-level maternal health system assessment and planning. In this study, we develop a uniquely detailed set of spatially-linked data and a calibrated geospatial model to undertake a national-scale audit of geographical access to maternity care at birth in Ghana, a high-burden country typical of many in sub-Saharan Africa. Methods. We assembled detailed spatial data on the population, health facilities, and landscape features influencing journeys. These were used in a geospatial model to estimate journey-time for all women of childbearing age (WoCBA) to their nearest health facility offering differing levels of care at birth, taking into account different transport types and availability. We calibrated the model using data on actual journeys made by women seeking care. Results: We found that a third of women (34%) in Ghana live beyond the clinically significant two-hour threshold from facilities likely to offer emergency obstetric and neonatal care (EmONC) classed at the ‘partial’ standard or better. Nearly half (45%) live that distance or further from ‘comprehensive’ EmONC facilities, offering life-saving blood transfusion and surgery. In the most remote regions these figures rose to 63% and 81%, respectively. Poor levels of access were found in many regions that meet international targets based on facilities-per-capita ratios. Conclusions: Detailed data assembly combined with geospatial modelling can provide nation-wide audits of geographical access to care at birth to support systemic maternal health planning, human resource deployment, and strategic targeting. Current international benchmarks of maternal health care provision are inadequate for these purposes because they fail to take account of the location and accessibility of services relative to the women they serve. © 2012 Gething et al.; licensee BioMed Central Ltd.

The methodological objective of this study was to model geographical access of women to health facilities offering care at birth. Figure ​Figure11 shows schematically the various data and modelling components used to achieve this aim. In brief, we assembled a national geo-referenced database of health facilities providing care at birth, and extremely detailed digital topographic data on transport networks (roads, tracks, footpaths) and barriers to travel (e.g. rivers) which were used with a cost-surface algorithm within a geographical information system (GIS) to estimate journey-time from every 100 m × 100 m grid square to the nearest health facility offering a given category of care at birth. Separate models were developed for mechanised versus non-mechanised forms of transport, and the likely proportion of women using each transport type to access care in different locations was estimated using a small-area-estimation (SAE) approach that combined sparse sample survey data with spatially complete census-derived proxies. Schematic diagram showing main input data, analytical steps, and primary outputs in generation of a calibrated nationwide journey-time model. SAE = small area estimation; EA = enumeration area; WoCBA = women of childbearing age. To maximise the realism of the model for mechanised journey-times, we carried out an initial calibration stage in which survey data on actual journey-times made by women in labour were used to find optimum model parameters which were then applied nationally. By combining the resulting per-pixel map of journey-times with a high-resolution population map for WoCBA, we generated estimates of the proportion of WoCBA able to access successive levels of care within two hours, within four hours, or more than four hours journey-time. Each of these components is now described in more detailed. We compiled national lists of health facilities from four main sources. First, a list was obtained from the Ghana Ministry of Health containing records of 2,021 facilities of all types nationwide that included for each a description of services offered and the region (first administrative level), district (second administrative level), and town in which each facility was located. Second, a list of geo-referenced facilities was compiled by the Centre for Remote Sensing and Geographic Information Services (CERSGIS), University of Ghana that contained listings of 1,915 facilities nationwide. These two lists were combined, cross-checked and reconciled. The Ghana Ministry of Health also maintains listings of health facilities by district on a web resource. These were cross-checked with the formal listings and any additional facilities added to our database. Facilities without latitude and longitude data were geo-referenced by manual matching of listed town names to mapped locations on Google Earth and, in the remaining unresolved cases, by telephone contact with district health offices to confirm locations. Facilities that do not offer maternity services were excluded (e.g. psychiatric hospitals, supplementary feeding centres, nursing training colleges and administrative offices). Finally, a recent project by the Ghana Ministry of Health and Ghana Health Service carried out an audit of maternity facilities nationwide with accurate assessments of the level of emergency obstetric and neonatal care (EmONC) that each offers. Using an established set of nine ‘signal-functions’ [30] of potentially life-saving birth care services, all government facilities were classified as non-, partial-, basic- or comprehensive-EmONC depending on the number of signal functions available [29]. All hospitals offering partial-EmONC or higher were extracted from this report, cross-referenced with our existing database, and geo-referenced using the process described above. Where more than one facility was listed at a single site (either because they shared the same building, or because they were geo-referenced using a village location), we retained only the highest order facility for subsequent analysis, thus avoiding any potential duplicated facility listings. We focused on those health facilities providing care at childbirth, rather than antenatal or postnatal/postpartum care, as this is the crucial period within the continuum of care chain when most mortality occurs, both for women and their newborn babies [5]. We stratified our analysis of geographical access to facilities providing three tiers of care. First, we considered a broad categorisation of 1,864 facilities of all types listed as offering any standard of care at birth (hereafter denoted as any-birth-care facilities, ABC). These spanned the complete spectrum of care from large tertiary hospitals to the most basic peripheral facilities including maternity homes as well as Community-based Health Planning and Services Initiative (CHPS) facilities. Measurement of access to this mixed level of care will overestimate true service availability because many people will be forced to bypass the simpler facilities, many of which offer only rudimentary care and no 24-h staffing. Indeed, even relatively well equipped hospitals in Ghana can offer less than 24-h cover [29]. We included this broad category of health facilities as the theoretical point-of-entry to the health system that also represents a hypothetical best-case scenario were all such facilities fully functional and offering robust referral services. Second, and representing the other extreme, we considered those 76 hospitals nationwide assessed as offering comprehensive-EmONC services (hereafter C-EmONC). This is a much more stringent designation, denoting hospitals providing all nine signal functions, including the availability of blood transfusion and surgical/caesarean section capability that are typically absent from other facilities. The reality for labouring women in Ghana often lies between these two extremes: a wider set of hospitals offer partial- or basic-EmONC services (where six/seven or eight signal functions are provided, respectively) that nevertheless represent a much higher degree of service than non-EmONC facilities and are able to respond appropriately to a range of birth complications. We therefore assessed access to a third intermediate category that included these partial- and basic- as well as comprehensive-EmONC hospitals (hereafter PBC-EmONC), representing 157 facilities nationwide with heterogeneous levels of care. The final geo-referenced facility database was imported into a GIS (ArcGIS 10.0) as a point shapefile for analysis. To support geospatial modelling of realistic journey-times to health facilities, detailed topographic datasets were obtained directly from CERSGIS. These included digitised topographic survey data on the national road network and additional tracks and trails, as well as other features such as rivers, lakes, and marshland that may act as a natural barrier to determine the route taken during journeys. These data stem from an unusually detailed national programme of land surveillance carried out by the Water Research Institute, CERSGIS, Department of Feeder Roads, Ghana Survey Department and the Forestry Commission of Ghana between 1995 and 2005. All layers were available as point, polygon, or line feature shapefiles and were imported into ArcGIS for subsequent analysis. These are described in more detailed in Additional file 1. Cost-surface algorithms are increasingly used within GIS software to estimate journey-times across modelled landscapes [41-47]. Users first define a gridded impedance surface in which the value of each grid cell represents the estimated time required to traverse it, taking into account the size of each cell and the type of landscape feature it represents. Low impedance values are assigned to high-speed features such as roads, with much larger values for off-road or rough terrain. Barrier features can be designated as impassable or assigned very large impedance values. Destination features (e.g. health facilities) are located on the modelled landscape, and the cost-surface algorithm computes the shortest cell-by-cell route from each origin cell to its nearest destination feature. The cumulative sum of impedance values along the route provides the estimated journey-time. The accuracy of these journey-time estimates is dependent on detailed landscape data and on appropriate choice of impedance parameter values. Cost-surface models have been used to estimate journey-time to health facilities in resource-poor settings, but have often focused only on journeys made on foot [32] or else have dealt simplistically with varying modes of transportation [48]. Pedestrian journeys made without mechanised transport are potentially more straightforward to model because average speeds will tend to be relatively similar across different settings and, because different categories of paths, tracks, and roads offer broadly similar walking speeds, journeys rarely deviate from the most direct route available. We used an established parameter set to represent an average non-mechanised travel speed of 5 kmh-1 on established roads or tracks, and 2.5 kmh-1 elsewhere [32,49]. Barriers such as rivers and lakes were given higher impedance, meaning journeys were likely to utilise established bridges and crossings where available, but in their absence could be traversed (e.g. by boat), with an appropriate time delay. Where journeys are made by car, bus, or motorbike, speeds can vary widely according to the type of road or track, and the most direct route is often not the fastest. One approach is to use statutory speed limits as a means of parameterising impedance values for different road categories, but this makes numerous assumptions about road, vehicle, and driver characteristics that may not be valid in many settings. To maximise the realism of our mechanised journey-time model a calibration exercise was undertaken using data on real journeys made by women in labour seeking care in Ghana. Such data have previously been obtained by the IMMPACT [50] project, originally designed to estimate out of pocket costs for birth care [51,52] in which a sample of women giving birth at health facilities in two regions in Ghana (Volta and Central) reported their origin (home) and destination (facility) locations, mode of transport, and time taken to make the journey. From these data, we extracted a total of 138 unique origin–destination pairs and, for each pair, an impedance grid was established from the national topographic data to model the surrounding landscape. These grids differentiated five categories of road or track, from the fastest national highways through to minor paved roads or unmade tracks (see Additional file 1: Figure A1.1). Recognising that the smallest tracks connecting households with the road network may not be captured in the database, a category for ‘background’ grid cells was defined, within which rivers and lakes were also defined as potential barriers (see Additional file 1: Figure A1.2). An automated algorithm was developed using Python 2.6 which allowed a large number of cost-surface models to be run within ArcGIS 10.0 for each origin–destination journey. In each run, a different set of candidate impedance parameters was assigned to the various landscape features and the resulting estimated journey-times were compared to the values reported in the IMMPACT survey. The overall performance of each parameter set was assessed by the median magnitude of errors between predicted and observed journey-times. A total of 1000 parameter sets were assessed in this way using a hierarchical grid-search that spanned the range of plausible values and the set returning the smallest median absolute error was identified. Optimum impedance parameters were identified as an average travel speed of 60 kmh-1 for national roads, 45 kmh-1 for inter-regional and regional roads, 5 kmh-1 for unmade tracks and trails and 1.75 kmh-1 for background pixels. The impedance parameters described above were used to define a mechanised and a non-mechanised impedance grid covering all of Ghana at 100 m × 100 m spatial resolution. Cost-surface algorithms were then implemented for both transport modes to calculate journey-times to nearest ABC, PBC-EmONC and C-EmONC facilities, resulting in a total of six journey-time surfaces. Clearly, journeys made by mechanised transport will almost always be substantially faster than those on foot, and so the availability of transport to women in labour is a critically important determinant of their geographical access to care. This availability will itself be influenced by complex socioeconomic factors that will vary from place to place. To estimate the fraction of women likely to be able to use mechanised transport to seek birth care, we first obtained data collected during the Core Welfare Indicator Questionnaire (CWIQ) survey carried out by the Ghana Statistical Service in 2003 [53]. This is a nationally representative survey which sampled 210,170 individuals from 49,003 households from all the 110 districtsa of Ghana and included a question on the mode of transport used in accessing health facilities. This survey was preferred to any direct data on, for example, car ownership, since the journey to seek care is out-of-the-ordinary and may represent a rare occasion when a taxi or bus ride is purchased, or when health facilities themselves may organise transport [6,29]. We used an SAE approach for a unit-level model [54] to relate the CWIQ data to a suite of potential correlates (literacy rate; dwelling ownership; marital status; urban population; material of roof, wall and floor; main source of drinking water; type of toilet facility and main fuel used for cooking) that were also available for all enumeration areas (EAs) from the 2000 Ghana Population and Housing Census [55] and, thereby, impute the fraction of women in each EA likely to make mechanised versus non-mechanised journeys to seek care. As a final step, the six national journey-time maps were combined with a population grid detailing the number of WoCBA (defined as 15–49 years) residing in each 100 m × 100 m grid cell (see Additional file 1: Figure A1.3). This surface is a new product produced by the AfriPop project (http://www.afripop.org) that combines high-resolution census data with satellite sensor imagery of settlements to create the most detailed population surfaces available for Africa [56]. We created policy-relevant summary statistics by summing the number of WoCBA within each district and region that fell within three levels of geographical access: less than two hours from a given facility type, greater than two hours, and greater than four hours. These thresholds follow earlier studies and are based on both clinical factors (two hours being the estimated modal time to death for postpartum haemorrhage [57,58] and empirical analyses showing significant successive increases in maternal case-fatality rates associated with journey-times of greater than two and four hours [59]. All data used in this study were either available on an unrestricted basis in the public domain, or provided with permission from the agencies described above.

• gething, peter w.
• johnson, fiifi amoako
• frempong-ainguah, faustina
• nyarko, philomena
• baschieri, angela
• aboagye, patrick
• falkingham, jane
• matthews, zoe
• atkinson, peter m.