The study has assessed current catchment area mapping practices at health facility level within Blantyre, Malawi. We generated catchments for operational health facilities within Blantyre, utilising open-source data and a robust method frequently used in spatial epidemiology. In addition, we crowdsourced road network data, established travel speeds, estimated travel time to health facilities and generated health facility catchment areas.
The use of hand-drawn paper maps by public health facilities in Blantyre has two main disadvantages: 1) such maps are subject to distorted geographical scale, resulting in exaggeration and/or underestimation of distance and 2) such maps are difficult to share and embed easily into evidence-based health system decision making. Digital health applications present many advantages and it is important to build health-focused GIS that can be used for planning. Some efforts to develop this capacity has already been done in a low-income setting, with examples including engaging health workers in mapping villages to highlighting variabilities [29]. Nonetheless, the approach of determining geographic accessibility to healthcare via cost-distance algorithms depend on the existence of road network data, barriers to movement and accurate information about physical location of health facilities. Nevertheless, open-source tools for generating catchment areas such as AccessMod [28] are available, and road network data can be accessed through OSM [17], and in case when road network data are of poor quality, updating road network datasets can be crowdsourced as demonstrated in this study. The availability of such tools and data resources suggest that this approach can be scaled up to national scale and be used to update the catchment areas that were developed in the early 2000s [25]. Between 1994 and 2018, the city registered nearly 14 percent transition of bare to built-up land [30]. Alongside this transition, the population of the city grew. Likely it is possible that the built-up communities might have been established in areas lacking health facilities, resulting in the strain and overextension of existing community health services. Uneven distribution of health facilities was previously acknowledged in the UN Habitat Blantyre urban profile [31], and this work has highlighted spatial variations in geographical access to health services in Blantyre city. Consequently, existing facilities would face increased expenses in terms of both travel distance and financial resources needed to organize community health programs. To mitigate these inequities, expansion of existing facilities where needed and all of this should go hand-in-hand with a review of resource allocation in proportion to the size of the catchment population. This approach would help alleviate disparities in access to healthcare services and ensure that healthcare provisions are more evenly distributed across the growing urban landscape. Given future changes in the availability of health services, human settlement distribution and density, land cover characteristics and road network, maps highlighting travel time and facility catchment areas should be updated on a regular basis. Compared to the current study where we visited individual health facility to determine operational status and accuracy of geospatial information, future work could include verifying operational status via remote contact with facilities by telephone or email. There is a need to ensure that these datasets are up to date and maintained. We advocate for a thorough review of health facilities across Malawi. However, we are aware that such an undertaking may be time and resource intensive. Our study is unique in the field-validation of its derived catchment areas.
Another aspect of this research is crowdsourcing of mapping effort to update OSM. As it is already known, OSM has a vast collection of user generated maps that are free to use and editable [17], however as this platform depends on user contributions, it is not 100% complete or accurate for some geographies [32]. In this project, we addressed these data gaps through two mapping events. These events were additionally organized to build a workforce and capacity to support mapping of roads in both Blantyre city, and other Malawian geographies beyond this study. Most of the mapping workshops (mapathons) attendees were beginner mappers. It has however been noted that most of the contributed mapping was performed by advanced mappers with familiarity of OSM mapping techniques and previous, personal experience. As there are few advanced mappers, upscaling the mapping to a national level can be a slow undertaking. Open mapping communities, for example OpenStreetMap Malawi can have a role in engaging and facilitate skill transfer to less advanced mappers. Additionally, university-based mapping groupings such as the YouthMappers (https://www.youthmappers.org/ ) can be utilized to crowdsource mapping efforts. Already, it is known that between 2015 and 2021, the YouthMappers made nearly 7.1 million edits on OSM across African countries [33]. Another issue that was observed with OSM data are the accuracy of road tags, such as highways. Highway tags provided an opportunity to model different travel scenarios, but mappers need experience to properly assign a highway tag to a road. This should be balanced with local knowledge as the current OSM tags are based on road hierarchy from high-income countries and some tags might not be appropriate for mapping in a different context e.g., in the Malawi setting. A good portion of the roads were not classified (12.3% of the total roads and 16.8% of the total distance covered by the roads). Standardizing quality control and ensuring data completeness can make a big difference.
The study has reported for the first time, observed travel speeds on different roads in Malawi. Similar works within Malawi and elsewhere used maximum speed indicated in road safety guidelines [35] and OSM travel speeds [14], however, excluding trunk roads, the observed travel speeds for walking on OSM roads in this study were higher than what is commonly used [35]. In previous studies, travel speeds were corrected to reflect the target population. Regardless, these travel speeds reported in this can be used to update travel estimates elsewhere, as has previously been done in Uganda [36]. It is crucial, however, to acknowledge that the current study did not encompass all factors influencing travel speed disruptions. For example, the authors know that minibuses offer common public transport, but they sometimes stop to wait for passengers. Equally wet weather can make roads difficult to use by some means of transportation. It was observed in Mozambique that following a tropical cyclone, travel time to health facilities increased from several minutes to up to 78 hours [37]. Furthermore, all the observations were made during weekdays and from the travel surveys were made on individuals who were not necessarily travelling to health facilities – they were ordinary road users, and more accurate estimates of travel times may be obtained through GPS trackers as opposed to roadside-based observations. In general, observations of travel speed may overestimate travel time as they fail to consider delays and pauses during travel, such as traffic congestion. Accordingly, the travel times presented here must be taken with caution as they might not purely reflect travel scenarios when people are accessing health services. This may in turn affect our understanding of the influence of treatment seeking behaviours and referral completion rates for infectious diseases requiring frequent treatment or follow-up at a health facility.
Our validation data show that correspondence between the catchment areas we derived and the actual locations of health facility users exceeded 90% in all cases. However, an average of 5.8 percent of the patients come from outside the generated catchment areas. A potential explanation of this deviation is that patients might not always seek health services from facilities that are located closer to their homes. This is a limitation of the approach used in this study, as it is known that apart from physical distance, choice of a health facility is also influenced by factors including users’ perception of the quality of health services delivered [38], seeking health services while at work or school, and fear of being seen at the hospital [39]. A typical example of this can be when only a subset of facilities have access to a particular diagnostic tool or treatment or when patients are concerned about privacy of their health status. In a previous study it was noted that for some diseases such as HIV, patients might choose facilities that are far from where they reside for fear of disclosing their health status to their communities or preference for better services elsewhere [40]. This suggest that if different health outcomes were considered for validation, the results may have been different and validation results should be seen with caution when the catchments are to be utilized for a different disease outcome. However, moving forward, we can consider conducting patient surveys focusing on facilities they may visit under various medical conditions, we will also consider utilizing close-to reality travel time derived from big datasets as in the recently published work by ONTIME consortium [41]. It is also important to note that travel time estimates is dependent on the service which patients seek at the facility. For example, trauma patients are more likely to be taken to the facility by car. This implies that a need to generate a disease or service specific catchment areas. We also intend to extend this work to derive fuzzy catchment areas where patients are on allocated to facilities on probabilistic basis. Once developed, this can be integrated in the digital health architectures being developed by the Ministry of Health. Equally, this can have denominator data acquired from satellite-based estimates such as WorldPop.
The generated catchment areas have the potential to be used for quantifying disease burden and estimation of facility catchment population sizes which can be used as denominators for disease prevalence or incidence estimation. This can become more practical with the availability of health records through platforms such as the District Health Information System 2 [42]. However, we acknowledge additional limitations to the approach employed in the current study. First, the travel speed observations did not account for the effect of slope on travel speed, and OSM tags associated with the roads were assigned after field observations. Better results could have been obtained if locally relevant OSM categories were properly defined prior to commencement of the study. Regarding the influence of slope, it is important to note that the observer did not record the age category of the walkers. Recognition of the influence of age has been noted in previous studies through the use of lower travel speed for children [35], and pregnant women. Another limitation is that the study did not account for the influence of barriers such as river channels on travel to health facilities. Furthermore, the study exclusively focused on the walking scenario for creating travel time maps and related catchment areas. Future studies should consider computing scenarios where patients access health services using bicycles, motorcycles, cars, or various combinations thereof. Nonetheless, the approaches used within this study provide a robust basis for future works refining catchment areas.