Geographical Access to Health Sites in North and South Kivu,
Democratic Republic of Congo
Final paper and project
Michael Graham, GIS for International Applications
Introduction
With this project I sought to better understand the possibilities and limitations in using GIS to
analyze how difficult it was for Congolese to get to health sites.
Background
The Democratic Republic of the Congo has experienced a decade of conflict that has decimated
health infrastructure. In much of the country, access to health care is a 1-2 day walk away. In
the Kivus, at the epicenter of humanitarian funding and with the largest population outside of
Kinshasa, access is better relative to other regions, yet still severely lacking. While lack of data
prevents us from knowing the type or quality of care provide, we can still judge the physical
accessibility of villages to health structure in North and South Kivu.
In preparing for the analytical process, I encountered several projects that had analyzed
geographical accessibility for health centers in Latin America:
1.
Ebener et al. ‘Using GIS to measure Physical Accessibility to Health Care’, World Health
Organization, 2004. http://www.who.int/kms/initiatives/Ebener_et_al_2004a.pdf
Presents two models for analyzing accessibility, using Honduras as a case study. Uses road
networks, facility locations etc, as well as quality and slope of roads to come up with average
speed and travel times.
2.
Medicins Sans Frontieres. Access to healthcare, mortality and violence in the Democratic
Republic of the Congo. October, 2005.
https://www.doctorswithoutborders.org/publications/reports/2005/drc_healthcare_112005.pdf
This report surveyed multiple regions in DRC and asked what prevented use of health services.
In a couple settings, distance of facilities was a major factor (30%), where in others it was minor
(2-3%). It helps provide context- physical distance did not seem to prevent consultations in
most areas, but in some it was much more of a concern.
3.
Modelling and understanding primary health care accessibility and utilization in rural South
Africa: an exploration using a geographical information system. Tanzer et al. Social Science and
Medicine. Volume 63, Issue 3, August 2006, Pages 691–705
www.sciencedirect.com/science/article/pii/S0277953606000372
The most interesting conclusion of this article was the following, which contrasts to the MSF
study:
There was a significant logistic decline in usage with increasing travel time (p < 0.0001).
The adjusted odds of a homestead within 30 min of a clinic making use of the clinics
were 10 times (adjusted OR = 10; 95 CI 6.9-14.4) those of a homestead in the 90-120
min zone.
This demonstrates that distance from a clinic, at least in South Africa, plays a significant role in
usage of that facility.
In terms of analytical tools, the authors used far more robust and nuanced processes and
surveys than was possible with this initial project.
4.
Okwaraji YB , Cousens S , Berhane Y , Mulholland K , Edmond K (2012) Effect of Geographical
Access to Health Facilities on Child Mortality in Rural Ethiopia: A Community Based Cross
Sectional Study. PLoS ONE 7(3): e33564. doi:10.1371/journal.pone.0033564
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0033564
Like the Honduras study, this research looked at travel times to clinics, and found a drastic
impact on child mortality based on distances over 1 hour. It used similar slope analysis of roads:
In this study, a speed of 5 km/hr was assigned for all walking routes, slopes greater than
30 degrees were assigned a speed of 0.1 km/hr and traversing through water bodies
was also assigned a speed of 0.1 km/hr.ata
If I had better data and more time, I would seek to replicate these processes and do additional
field research involving interviews to estimate travel time and ground-truthing estimates.
Data
The crux of the project was acquiring health center data. I was able to get a hold of a partial
dataset from a contact working in DRC, which proved a mixed bag. While it was the most
complete seen yet, there was no metadata to explain its provenance, and no attribute data
about what type of facility it was- primary, tertiary, pharmacy etc. It was also not a national
dataset- it included seemingly random coverage of areas. Luckily the one region that seemed
to offer complete coverage was North and South Kivu. This is an interesting area to look at- it
was the epicenter of post-genocidal conflict in Congo, has other complete datasets such as
roads and localities, and has varied terrain that seemed as if they might play a critical role in
health access. I found datasets on roads, administrative boundaries, and other secondary
features.
I added SRDM terrain data to compute slope, and landscan population data.
Methods and Results
I gathered point data from the Congolese ministry of health and online sources, and used them
to create distance rasters for the locations of health structures and towns. A vector roads layer
was found, and two raster- SRTM-derived slope and population density – were also used.
These rasters were reclassed with rankings from 1-5 to build a single composite raster that
gives pixels a value based on combined rank of accessibility indicators. By transferring these
rankings to the town layer, the viewer is able to sort towns visually by level of access. Finally, by
adding raster population data to town points, I identified towns with both highest population
and least access to care, showing the first 20% of highest population as suggestive of a place to
prioritize with further intervention.
The rasters proved difficult to get right, with a great deal of time spent dealing with technical
challenges of cell spacing and raster calculations.
The analysis of the raster and town rankings found two things: first, it identified specific
physical areas (shown in shades of red) that are largely inaccessible to health facilities, given
lack of roads, high slope of terrain and physical distance from health structures. The same is
visualized in more detail for specific towns. The two together give us a good idea which villages
and regions are least accessible. Without knowing how many people live in these least
accessible villages or regions, this does little to identify gaps in access for Congolese. By
selecting from the vulnerable towns only those with populations greater than 100 (20% of total)
we were able to identify 60 towns (2% of total) that should be prioritized for health structure
access intervention.
While quality of input data makes confidence in this analysis low, with better data on health
facilities we could analyze. More complete health infrastructure data for the DRC on a national
level would allow for far more useful understanding of gaps in accessibility.
Conclusion and Lessons Learned
Data means everything, and without adequate health center data, the results of this analysis
have a low confidence level. It is impossible to know whether ‘health structures’ are in fact
clinics, or one-man pharmacies with no actual medicines. A detailed dataset of this would allow
for some very interesting analysis.
I felt that reclassifying could too easily change the fundamental analysis and message of the
project. Assumptions were everything- the assumption that roads were important, and slope
mattered. I don’t know that they do in eastern DRC to a large extent. Or maybe they do in
different ways in different villages. Some of the prior studies spent many months talking with
people on the ground about what mattered most. A real GIS analysis needs that sort of groundtruthing to have any chance at realistically assessing such challenges.
But ultimately I did think it was a useful exercise in the steps required, and possibly held some
degree of truth. It helped show that villages and clinics were located on top of steep terrain,
and not low in the fields. Roads followed these same paths. This would be important
information to know in further GIS analysis of accessibility.
In reading prior write-ups of analysis done by others, it was evident that travel times to clinics
plays an enormous role in many people’s health decisions. The further a clinic, the less likely
someone will be visiting regularly for preventive care, and the more likely death or serious
injury will be a result.
Below are the two results of the analysis. The first is a combined raster with weighted,
reclassed rasters that provides an estimate of difficulty to access any given health facility. The
second translates this into village terms, and see which villages find themselves worst situated
to access health care.