Resource-poor regions around the world have been suffering from inequitable health infrastructure for decades. As we look forward to the 2020 Virtual UN World Data Forum (October 19-21, 2020), hospital accessibility planning falls squarely within Thematic Area 2: “Leaving No One Behind.”
The World Bank’s Development Data Partnership brings international organizations together with private sector companies to facilitate the use of third-party data in research and international development. Using this platform and guidance of experts in the field, our team worked with several open and private data sources to address the challenge of measuring accessibility to health facilities in select regions. Here, we discuss our workflow and demonstrate findings in Indonesia and the Philippines as a proof of concept.
Indonesia and the Philippines are archipelagos, where geographic and administrative boundaries are similar. The collection of islands is home to different concentrations of ethnic, cultural, linguistic, and socioeconomic groups, with an imbalanced distribution of health facilities and physicians.
Through conversations with experts on the ground, we understood that health facility records are largely incomplete and on paper. By using a combination of public and private sector data, we aimed for a more comprehensive understanding of health facility accessibility.
To measure health accessibility, we focus on travel times to the nearest health facility. By using time as our comparison metric, we take into account key barriers like the quality of roads and the amount of traffic. In our analysis, we aim to understand what parts of a region/province are beyond the threshold for acceptable travel time.
To test our framework in various settings, we use national health and demographic survey data to focus on regions with high health indicators and a region with low health indicators in both countries. For Indonesia, we chose East Java (high health indicators) and Papua (low health indicators). For Philippines we focus on National Capital Region or NCR (high health indicators) and Bangsamoro Autonomous Muslim Mindanao or BARMM (low health indicators).Datasets
In order to conduct this analysis, we bring together three datasets:
- Origins: Geospatial representations of where people live and begin their travel to health facilities
- Destinations: Locations of hospitals and clinics
- Travel Networks: Road networks used to calculate travel times
Using Facebook Population Density Maps, we obtained a highly granular population dataset (up to 30m resolution). Facebook sources these data using a combination of Census data and computer vision algorithms on satellite imagery. They make maps available by country and different demographic groups (age, gender).
Using the Mapbox Matrix API, we calculated the travel time between a pair of origins and destinations. The API takes different transportation methods into consideration: driving, driving with traffic, walking, and cycling.Public Data
To address Indonesia and Philippines’ lack of cohesive and fully digitized information on locations of hospitals and clinics, we used OpenStreetMap POIs. This health facility location dataset is crowdsourced and edited by a wide community of mappers. Comprehensive mapping is a challenge in many countries, with multiple initiatives working to source this data, such as HealthSites.io Where it made sense, we combined OpenStreetMap’s data with Facebook’s Hospital Locations datasets derived from public imagery.
The World Bank’s Geospatial Operations Support Team initially developed GOSTNets to enable users to perform network analysis on OpenStreetMap data quickly and easily. We used the GOSTNets library to create a network graph for each region and to snap each population origin to a road for accurate travel time calculations.Methodology
For each region / population segment pair evaluated, we followed the workflow below:
- Used GOSTNets to create a network graph for the region.
- Used Facebook Population Density maps to break the region of into similar-sized windows, treating the center of each window as an origin.
- Used GOSTNets to snap origins to the closest road.
- Extracted hospital and clinical locations via OpenStreetMap POIs.
- Used Mapbox Matrix API to test hundreds of origin-destination pairs to find the closest hospital or clinic to each origin and calculate the associated travel time.
- Calculated “accessibility percentages” to quantify the portion of a region’s population that is beyond a set threshold (we use one hour) , based on different transportation methods.
- Created visualizations for what areas of the region are beyond the acceptable travel time threshold with regards to population density.
We have presented our findings in the form of interactive maps, which show the percentage of a population in a study region that cannot reach a health facility within 1 hour of travel time, by mode.Papua, Indonesia
East Java, Indonesia
This work is an initial analysis. However, we look forward to validating these results on the ground and working with local decision makers to put this framework into action. Based on where those conversations lead, we may:
- Perform this analysis by gender, hospital capacity and travel preferences to understand inequities in health accessibility at an ever narrower scope;
- Use this analysis to inform where new testing facilities or mobile clinics may be set up to expand the reach of health systems to vulnerable populations; and/or
- Create an interactive tool to perform this accessibility analysis.
You can find our code here and modify the analysis to your regions of interest.Acknowledgments
- Bruno Sanchez-Andrade Nuno’s previous workon health facility mapping
- World Bank Colleagues: Holly Krambeck, Mersedeh Tariverdi, Andres Chamorro, Data Development Partnership team, Geospatial Operations Support Team (GOST)
- Experts in the Field: Celina Agaton (MapPH), Laura Raulston (The World Bank), Allan Hsiao (MIT Economics)
- Facebook Data for Good Team, Mapbox Community Team