Approaches to defining health facility catchment areas in sub-Saharan Africa
1Population Health Unit, KEMRI-Wellcome Trust Research Programme, Nairobi, Kenya
2Centre for Health Informatics, Computing, and Statistics, Lancaster Medical School, Lancaster University, Lancaster, UK
3Ignite Global Health Research Lab, Global Research Institute, William and Mary, Williamsburg, Virginia, USA
4KEMRI-Wellcome Trust Research Programme, Kilifi, Kenya
5Centre for Tropical Medicine and Global Health, Nuffield Department of Medicine, University of Oxford, Oxford, UK
*Correspondence: pmacharia@kemri-wellcome.orgAbstract
The geographical area around a health facility characterizing the population that utilizes some or all of its services – a health facility catchment area (HFCA)-, forms the fundamental basis of estimating reliable population denominator for disease mapping and routine healthcare planning. Consequently, the approaches used to delineate the catchment area have a direct impact on the health of a population. To date, there is no systematic literature review documenting different approaches that have been used to define HFCAs while elucidating the implications on derived population denominators. To fill this gap, we systematically reviewed literature and documented approaches that have been used to define HFCA in sub–Saharan Africa (SSA). Simple to complex approaches have been used to define catchment areas in SSA with varying degrees of complexity and limitations in the last four decades. These approaches are mainly driven by lack of geocoded data on the residential address of care seekers and their care-seeking behaviour. To generate closer-to-reality HFCA, for robust disease mapping and healthcare planning, additional data and innovative approaches balancing between model complexity and routine programmatic use are required.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
PMM is supported by the Royal Society Newton International Fellowship (number NIF/R1/201418). JNO is supported by the Global Research Institute Fellowship at the College of William and Mary. EM is supported by funds from Professor Robert Snow Wellcome Trust Principal Fellowship (number 212176). EAO is supported as a Wellcome Trust Intermediate Fellow (number 201866). PMM, EM, AM and EAO are grateful for the support of the Wellcome Trust to the Kenya Major Overseas Programme (number 203077).
Introduction
A health facility catchment area (HFCA), also known as a sphere of influence, tributary area, service area or demand field, represents a geographical area around a health facility describing the majority population that uses its services [1,2]. A HFCA is needed to define the catchment population (denominator), which is essential for disease mapping, optimizing timely routine and emergency access, immunization campaigns and vaccination programmes, distribution and allocation of essential health commodities, and planning the location of a new health facility [3–9]. Therefore, knowledge of HFCAs is important for efficient healthcare planning, and resource allocation within a population [10].
Defining a representative HFCA is non-trivial [11]. Its definition is substantially dependent on the availability of geo-positioned residential addresses of patients linked to the sought facility and robust data on their health-seeking behaviour. The healthcare-seeking behaviour is influenced by social-economic, cultural, and religious factors, transport systems, weather patterns, and the characteristics of facilities such as size, services offered, stock-outs, and competition from other health facilities [1,2,6,12]. However, in the majority of sub-Saharan Africa (SSA) and other low-resource settings, such data are not readily available due to limited resources in the context of many competing needs [2]. The problem is more pronounced in rural poor settings where formal address systems are almost non-existent [13]. In addition, privacy concerns may limit the use of precise spatial locations for the residential address of the patients [13,14].
As a result, reliable HFCAs have not been adequately defined by the ministries of Health (MoH) [2,12,15] which hampers routine planning and surveillance [12]. Current attempts have involved the use of the most fundamental data (health facility location and a set of simple auxiliary factors) to define HFCA [6] using a range of simple to complex approaches. However, these approaches are conveniently implemented, disregarding the implications of the defined HFCA on the accuracy of the catchment population and consequences for public service planning. To date, there has been no review of approaches that have been used to define HFCA in SSA to harness the best practices and innovations in defining a closer-to-reality HFCA. Here, we review approaches that have been used to define HFCA in SSA while documenting their pros and cons. We conclude by proposing a pragmatic approach based on the best practices of published literature that can be applied in the SSA context.
Methods
Our review followed updated guidelines from the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) [16].
Search strategy
To identify eligible papers, a comprehensive search strategy was developed under the guidance of an information and library expert and a group of spatial epidemiologists. First, we developed a search strategy that leveraged the unique search optimization features and indexing of each of the three electronic databases: PubMed, Scopus and CINHAL. The final literature search was conducted in April 2022. The main search terms were HFCA and its synonyms such as hospital catchment, health service area, facility service area combined with defining/creation/modelling/estimation/delineation/planning and the list of SSA countries. Boolean operators and asterisks were used to optimize the search process. We also screened the bibliography of the selected papers for additional papers. We used Mendeley and Rayyan to serve as bibliographic software for managing references in the review.
Eligibility criteria
The review sought to identify studies that were closely related to the measurement and conceptualization of HFCA in any SSA country. We did not limit the search by year; therefore, all years were included in the review. We excluded reviews, editorials, and conference presentations but included any relevant studies from their bibliography. We screened the identified studies in three stages: (1) screening by title, (2) screening by abstract, and (3) screening by reading the full text. Two authors independently reviewed all abstracts and full-text formats of the studies while a third author was used to resolve discordances. After screening, data were extracted from the remaining studies.
Data extraction
An online data extraction form was developed to obtain information about HFCA models and other important study characteristics. These characteristics were 1) bibliographic information, 2) facility level 3) health or study outcome, 4) analytical method used to define HFCA, 5) data needed to define HFCA, 6) sensitivity analysis, 7) and modelling gaps and recommendations that were acknowledged by the authors. Extraction discrepancies were resolved by consensus and by an independent arbitrator.
Data synthesis
Given the large scope of the review and the heterogeneity of the studies reviewed, a meta-analysis was not appropriate. However, a qualitative synthesis was conducted to identify approaches and methodological commonalities across studies and contexts.
Result
Overall, we retrieved 808 articles which were exported to the Mendeley reference manager. Studies were screened and excluded by title, abstract, and full text. Studies excluded after full-text review did not explicitly define an approach to model HFCA. Ultimately, 83 peer-reviewed articles met the inclusion criteria (Figure 1). The earliest manuscript was published in 1977 while the majority of the studies (84%) were published after 2008 with 2020-21 contributing 30% of all the studies. The studies varied by geographical scale and scope. Four studies (5%) were conducted across multiple countries whereas 79 studies (95%) were conducted within 21 individual SSA countries. Kenya (16%), Uganda (13%) and Zimbabwe (10%) had the highest number of studies while nine countries, each had at most two studies.
The studies were evenly distributed across the health system hierarchical structure, focusing on either primary (28.9%), secondary (33.7%), or both tiers (37.3%). The main health outcomes across the studies were highly variable. We identified 22 different health outcomes, with malaria (26.5%), HIV (15.7%), healthcare utilization (15.7%), vaccination (6.0%) and maternal and newborn care (6.0%) featuring in 70% of the manuscripts. Table 1 summarises data, threshold used to delineate HFCA, limitations (where indicated) and approaches, that have been used to define HFCA across the last 4 decades in SSA.
The use of subnational administrative boundaries (e.g., wards) to define HFCA was the second most common approach (21 studies). The boundaries of a polygon in which a facility was located and sometimes the neighbouring polygons formed the HFCA. Boundaries were used either independently or in combination with urban areas, disease estimates, population count or allocated by the MoH. The location of the health facility and the administrative boundaries combined with auxiliary datasets were the minimum dataset required. However, this approach ignores cross-border movement, and migration in and out of the catchment over time, particularly in rural areas where alternatives are limited.
Buffers around a health facility defining HFCA require the location of the health facility as the only input. Due to this simplicity, it was the most common approach (24 studies) including those that refined the buffers using population, administrative boundaries, or road networks (Table 1). The buffer size was based on a pragmatic distance that captured most patients or thresholds derived from literature, household surveys, local or international practices, facility level/function, and locality (urban or rural). The use of buffers was criticized because it is based on unrealistic straight-line distances which do not account for topography, transport modes, seasonality, mobility of people, the attractiveness of facilities, the inability of sick people to walk, and documented healthcare-seeking behaviour (such as bypassing the nearest facility). The authors justified the approach given the lack of updated data, especially healthcare-seeking behaviour data.
Closely related to the buffers is the use of Thiessen polygons also known as Voronoi diagrams to define HFCA (4 studies). They define a region incorporating all points that are closer to a given facility than any other facility, have similar data requirements and limitations as the buffers and can be combined with other approaches (Table 1). The need to account for the variable per-capita utilization rate within the HFCA was an additional limitation that was highlighted.
To account for some of the limitations in the use of administrative areas, buffers and Thiessen polygons, in defining HFCA, 20 studies applied a threshold on modelled travel time/distance to define a slightly improved HFCA. Time or distance was modelled through the path of least resistance via network analysis or cost distance surface accounting for transport mode, speeds, travel barriers (game park, reserves, water bodies and forest), travel factors (road network, land cover, topography) and sometimes simplified healthcare seeking behaviour. In some instances, modelled time was adjusted for facility capacity, and population or used in combination with Thiessen polygon, boundary or residential addresses (Table 1). The choice of the threshold was based on previous publications, policy recommendations by MoH or the international community or the use of a utilization decay curve.
Despite accounting for some limitations, drawbacks of modelled travel time/distance to define HFCA exist. Authors recognised the need to better account for care-seeking behaviour (bypassing the nearest facility, severity of illness, other dimensions of access, localised speeds, traffic, weather seasonality, urbanicity, realistic distribution and use of public transport, resources and infrastructure variation over time) supply-side factors (facility type and ownership, quality of services at facilities, referral patterns), and overlapping of two or more HFCA. Further, the authors argued that better data are needed to test model assumptions while balancing the trade-offs between model complexity, precision and routine application.
Albeit minimal, four studies used public participatory Geographical Information System (GIS) approaches. This involved community members and the patients defining HFCA, for example through data collection. Patients would identify their residential addresses from maps presented to them during health facility visits or interviews with long-term residents of an area or health staff to map HFCA. The main requirements were the participants, maps, imagery or a list of place names of the area. The approach was limited given the cost associated with acquiring high-resolution satellite imagery of the area and the incompleteness of existing spatial data.
On the other hand, the use of geocoded patient addresses (9 studies) linked with the health facility provided the most representative catchment area. The patient’s addresses were available at different spatial resolutions and were often refined or combined with boundaries, disease rates, and population. However, poor record keeping, bypassing of facilities, limiting the catchments within a region and the credibility of reported distances were reported as limitations.
Finally, to advance the approaches using modelled travel time, there were six standalone efforts to derive HFCA based on two-step floating catchment area [71,88], gravity models [89], spatial-statistical [35,90] and fuzzy choice models [6]. Mainly, these approaches had residential areas or enumeration areas contributing to HFCA with varying degrees of likelihood based on several factors (Table 1). Despite having some improvements, they did not satisfactorily account for variations in travel (mode of transport and speeds, road conditions, time of travel, traffic conditions, navigation errors), utilization rates and care-seeking behaviour, competition between facilities, population mobility, socio-demographic factors, geocoding inadequacies, cases not seen at a facility and non-governmental facilities.
Across the studies, a range of techniques were implemented as sensitivity analyses for the derived HFCA. These included deriving several HFCA for the same study area while using different; i) methods [23,34,59], ii) assumptions on healthcare-seeking behaviour [38], iii) population thresholds [80], iv) travel speed [71,88], v) radii for the buffer approach [54,73], vi) several teams validating the generated HFCA [81], and vii) using information criterion to select the best statistical model [35]. Finally, AccessMod and ArcMap were the most used software to derive HFCA. Other software included QGIS, Google Earth, GeoDa, Epi Info, R, STATA, FoxPro, and SAS
Discussion
The review has outlined approaches that have been used to define HFCA in SSA, a largely resource and data-constrained region. These approaches either rely on or are associated with techniques of defining geographical access as summarised in Ouma et al 2020 [91]. Overall, in SSA, there is a scarcity of geocoded data on patients’ residential addresses linked with the facility where care was sought which is the gold standard in defining a HFCA (Table 2). As a result, only six studies utilized such data [8–10,12,83–87], while six other studies either relied on MoH-derived HFCA [26,28,36] or used participatory GIS to collect data needed to delineate spatial extents of HFCAs [13,80–82]. The rest of the approaches used a variety of methods, with varying degrees of representativeness to delineate HFCA
Three commonly used approaches; administrative boundaries, buffers, and Thiessen polygons are limited because they oversimplify socio-demographic, epidemiological and health-seeking characteristics of communities when deriving HFCA (Table 1). These inadequate approaches will thus result in a non-representative catchment population and therefore, their use should be discouraged (Table 2). However, these approaches might be useful for applications that aggregate results to large subnational units. For example, a catchment derived using Thiessen polygons, but results presented at a district level. On the other hand, approaches based on travel time, gravity, and spatial statistical models while useful, also still require novel extensions to deal with their shortcomings (Table 1) to push the frontier to the next level. The advances should be made widely accessible at the programmatic level for routine use.
The key aspects that should be considered to open up a new avenue for HFCA definitions are cross-cross-border movement and overlapping catchments, mobility of patients, realistic travel times (that account for weather seasonality, transport modes within the public and private sector, localised speeds, road conditions, traffic, time of journey, navigation errors), competition between facilities, health-seeking behaviour (bypassing of the nearest facility, socio-demographic factors, severity of illness, cases not seen at a facility), facility characteristics (type and ownership, quality of service), residence (urban or rural) and referral patterns.
To account for these aspects, better data will be needed. This will also aid in testing model assumptions, and deal with the perennial incompleteness of spatial data, poor record keeping and geocoding inadequacies [92]. With the advancement in data science (such as machine learning) and data collection techniques (such as remote sensing), a range of climatic and environmental data (e.g., land use, rainfall patterns), road networks and traffic patterns can now be easily collected [93–95]. Increasingly available household surveys and routine data will be valuable for tracking utilization rates in the population to derive better thresholds for different health outcomes and contexts. Further, the use of mobile phones has become ubiquitous across the globe and can be harnessed to record geographical location information, especially in SSA to improve HFCA definition [96]. However, privacy and data protection concerns will need to be addressed when utilizing data from mobile phones. This is also a challenge affecting sharing of patients’ addresses and locations of service providers in the routine health information systems in SSA.
The travel time or distance thresholds that patients can travel are critical in the delineation of HFCA irrespective of the complexity of the approach. The threshold varies depending on the local context, health condition, severity of illness, and services offered at a facility. The use of healthcare utilization data for a particular outcome to create a decay curve or medical relevant thresholds is more useful than random and generalized thresholds. It is probably the use of random thresholds, cross-border movements, and simplified approaches (administrative boundaries, Thiessen polygons or buffers) that may have led to health coverage exceeding 100% at the facility level in recent DHIS2 analyses [97].
We, therefore, propose three levels when choosing an approach to delineate HFCA guided by the data availability and study objectives (Table 13.2). Level 1 is the most appropriate approach where HFCA can be defined unambiguously. It will require patients’ addresses to be geocoded and linked with the service provider where care was sought and where possible to harness recent technologies to collect these data. The second level (Level 2) is based on travelled time or distance but requires innovative methods to deal with the outlined key shortcomings. Finally, level 3, is the least recommended and its use is discouraged due to unrealistic assumptions.
Further MoH-derived HFCAs should be available across countries in SSA as a fundamental baseline for healthcare planning. However, limited studies referenced the use of MoH-derived HFCA, which may imply the absence of guidelines within MoH on defining robust HFCA. This may be attributed to poor documentation, or that the role of HFCA is under-appreciated. In this line, though at nascent stages, is a promising initiative aiming to create a system that enables MoH and stakeholders to define, create and manage their HFCAs [98].
Much of SSA and other low-resource countries are currently striving to achieve the ambitious targets within the Sustainable Development Goals (SDGs) framework by 2030. The SDG mantra of leaving no one behind, and reaching those farthest, behind, first would require estimating the populations in need of essential health services and defining health care coverage gaps at the HFCA level for targeted resource location. This will be essential for universal health coverage (UHC) to ensure that all people have access to the health services they need, when and where they need them, without financial hardship. Therefore, the role of accurate HFCA is timely and cannot be ignored as a catalyst for health development in SSA. In addition, the concept of a catchment area extends beyond health facilities and similar cases (limitations and requirements) may be advanced for catchment areas related to schools [4], community health workers, and vaccination posts among other service delivery points [2].
The review should be interpreted while considering several limitations. The literature search was limited to studies published in English. Secondly, given the vast nature of grey literature, some insights on HFCA in SSA might have been missed and our findings can only be applied to SSA countries or similar contexts. Despite these limitations, the review shows that most of the studies derived HFCA using simplified approaches due to a lack of appropriate data. To move the frontier of HFCA to the next level, the majority of the limitations that were acknowledged should be accounted for to derive closer-to-reality HFCA for robust catchment populations (denominator) for healthcare planning.
Data Availability
All data produced in the present work are contained in the manuscript
Declarations
Funding
PMM is supported by the Royal Society Newton International Fellowship (number NIF/R1/201418). JNO is supported by the Global Research Institute Fellowship at the College of William and Mary. EM is supported by funds from Professor Robert Snow Wellcome Trust Principal Fellowship (number 212176). EAO is supported as a Wellcome Trust Intermediate Fellow (number 201866). PMM, EM, AM and EAO are grateful for the support of the Wellcome Trust to the Kenya Major Overseas Programme (number 203077).
Availability of data and materials
All data relevant to the study are included in the article. The data supporting conclusions made in this review are available in the detailed reference list.
Ethics approval
This study used secondary data only, all publicly available and can be accessed from reference list
Patient consent for publication
Not required
Competing interests
The authors declare that they have no competing interests