University of Southern California Dissertations and Theses

Reducing maternal mortality by improving medical facility accessibility: a methodology demonstrated for the Democratic Republic of the Congo

(USC Thesis Other)

Reducing maternal mortality by improving medical facility accessibility: a methodology demonstrated for the Democratic Republic of the Congo

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Reducing Maternal Mortality by Improving Medical Facility Accessibility:
A Methodology Demonstrated for the Democratic Republic of the Congo

by

Lindsay Lynette Aazami

A Thesis Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)

December 2019

Copyright © 2019 by Lindsay Aazami

For my daughters, Tzipora Saoirse and Ellaria Shireen, may they become strong,
independent, successful, and empowered women. My hard work and drive have always
been to give them a better life than my own. I hope they see more, do more, learn more,
and experience more than I could ever imagine for them. Thank you to my husband
Kiarash, who sacrificed his time and especially energy to enable me to accomplish
writing this while I worked more than full time all while keeping up with a three-year-
old, a newborn, and then also moving across the world to New Zealand during my final
semester. Without him taking care of the home, the cooking, play dates, bath times,
bedtimes, grocery shopping, and an endless number of other tasks, there is no way I
could have finished this. He always believed in me, and I want to thank him for being
the only person that ever has.
iv
Table of Contents
List of Figures ................................................................................................................................ vi
List of Tables ............................................................................................................................... viii
Acknowledgments.......................................................................................................................... ix
List of Abbreviations ...................................................................................................................... x
Abstract .......................................................................................................................................... xi
Chapter 1 Introduction .................................................................................................................... 1
1.1. Objective .............................................................................................................................. 3
1.2. Overview of Methodology ................................................................................................... 3
1.3. Structure of the Thesis ......................................................................................................... 4
Chapter 2 Related Work.................................................................................................................. 6
2.1. Causes of Maternal Mortality .............................................................................................. 6
2.2. Spatial Analysis of Maternal and Newborn Health ............................................................. 8
2.3. Spatial Distribution of Medical Facilities ............................................................................ 9
2.4. Service Areas, Catchment Areas & Cost-Distance ............................................................ 11
2.5. Summary ............................................................................................................................ 13
Chapter 3 Data and Methods......................................................................................................... 14
3.1. Data .................................................................................................................................... 14
3.1.1. Study Area .................................................................................................................. 15
3.1.2. Health Zone Boundaries ............................................................................................. 16
3.1.3. Population Table ......................................................................................................... 16
3.1.4. Medical Facilities ........................................................................................................ 19
3.1.5. Elevation ..................................................................................................................... 24
3.1.6. Land Cover.................................................................................................................. 25
3.2. Methods.............................................................................................................................. 27
3.2.1. Prepare the Data .......................................................................................................... 28
3.2.2. Determine the Unserved Areas of the DRC ................................................................ 30
3.2.3. Generate Tessellation and Calculate Population......................................................... 32
3.2.4. Determine Distance ..................................................................................................... 33
3.2.5. Create Need-Based Ranking ....................................................................................... 33
Chapter 4 Results .......................................................................................................................... 36
4.1. Intermediate Results........................................................................................................... 36
v
4.1.1. Determine the Unserved Areas of the DRC ................................................................ 36
4.1.2. Generate Tessellation and Calculate Population......................................................... 38
4.1.3. Create Need-Based Ranking ....................................................................................... 41
4.1.4. Determine Distance ..................................................................................................... 41
4.2. Final Needs-Based Ranking ............................................................................................... 42
Chapter 5 Discussion and Conclusion .......................................................................................... 48
5.1. Lessons Learned................................................................................................................. 49
5.2. Improvements for Future Applications of the Methodology ............................................. 50
5.3. Value of this Research ....................................................................................................... 52
References ..................................................................................................................................... 54
Appendix - Model Builder Diagram of the Complete Methodology ............................................ 57

vi
List of Figures
Figure 1 The Causes of Maternal Mortality in Sub-Saharan Africa ............................................... 1
Figure 2 The boundaries of the DRC and the 519 Health Zones .................................................. 16
Figure 3 Adult female population count by health zone ............................................................... 18
Figure 4 Adult female population density within the health zones .............................................. 19
Figure 5 Healthsite.io data lifecycle ............................................................................................. 21
Figure 6 Sample of the Medical Facility Data showing the accuracy .......................................... 22
Figure 7 The locations of the 1,086 existing medical facilities within the health zones. ............. 23
Figure 8 The 1,086 existing medical facilities displayed as a heat map ....................................... 24
Figure 9 Elevation data of the DRC .............................................................................................. 25
Figure 10 The GLCC raster trimmed to the boundary of the DRC .............................................. 26
Figure 11 The workflow of the methodology used. ...................................................................... 28
Figure 12 The land cover raster reclassified based on travel speed. ............................................. 37
Figure 13 The results of the path distance tool displaying Unserved Areas in the DRC ............. 38
Figure 14 The 9,911 generated hexagons covering the Unserved Area of the DRC. ................... 39
Figure 15 Distribution of the estimated adult female population by hexagon. ............................. 40
Figure 16 Visualization of the Adult Female Population Rank .................................................... 41
Figure 17 Visualization of the ranked distance of the hexagons from medical facilities ............. 42
Figure 18 Visualization of the hexagons based on need for a medical facility ............................ 43
Figure 19 Histogram displaying hexagon distance from existing medical facilities .................... 44
Figure 20 Histogram displaying the estimated adult female population per hexagon .................. 44
Figure 21 Histogram displaying normalized adult female population data .................................. 45
Figure 22 Scatter plot of the relationship between distance and the adult female population ...... 46
vii
Figure 23 Scatter plot with population normalized....................................................................... 46
Figure 24 Scatter plot with distance and population normalized .................................................. 47

viii
List of Tables
Table 1 Data used in this study and sources ................................................................................. 14
Table 2 Sample of the adult female population data aggregated by health zone .......................... 17
Table 3 The classification of land use designated by the GLCC .................................................. 27
Table 4 Land Use reclassification based on travel speed ............................................................. 29
Table 5 A Sample of the adult female population within each hexagon by GRID_ID ................ 33

ix
Acknowledgments
I would like to thank my advisor, Dr Karen Kemp, for her patience, guidance, and for not giving
up on me when she saw the first draft of this thesis (I’m sure she got a good laugh though).

x
List of Abbreviations
2SFCA Two-Step Floating Catchment Area
AVHRR Advanced Very High-Resolution Radiometer
CHD Centre for Humanitarian Data
DRC Democratic Republic of the Congo
GIS Geographic Information System/Science
GIST Geographic Information Science and Technology
GLCC Global Land Cover Classification
GMTED2010 Global Multi-resolution Terrain Elevation Data, 2010
HDX Humanitarian Data Exchange
MCSP Global Maternal and Child Survival Program
NDVI Normalized Difference Vegetation Index
NGA National Geospatial-Intelligence Agency
OCHA Office for the Coordination of Humanitarian Affairs
OSM Open Street Map
UNFPA The United Nations Population Fund
(formerly: The United Nations Fund for Population Activities)
UNICEF The United Nations Children's Fund
(formerly: United Nations International Children's Emergency Fund)
USAID United States Agency for International Development
USGS United States Geological Survey
USC University of Southern California
WHO World Health Organization

xi
Abstract
The Democratic Republic of the Congo (DRC) is the fourth most populated country in Africa,
with approximately 87 million people, of which 44 million are female. Unfortunately, it also has
the 10th highest maternal mortality rate of any country in the world at 693 deaths per 100,000
births in 2015. High maternal mortality in the DRC is due in large part to pregnant mothers being
remotely located from medical facilities and routinely dying from preventable complications.
Cars are not prevalent in the DRC, and the most common means of travel is by foot due to the
destruction of the infrastructure caused by the First and Second Congo Wars in the late 1990s.
Walking long distances during pregnancy or while in labor and especially at night is a significant
barrier for women seeking medical care. This study’s objective was to develop a simple
methodology that could be used to identify ideal locations for new birth facilities where large
populations are the furthest distance from existing facilities. The locations were identified
through the generation of a tessellation grid over the areas of the DRC with low walking
accessibility to medical facilities. For each tessellation grid cell, the distance to the nearest
medical facility and the population within the cell were calculated. Based on a combination of
population size and distance from a medical facility, a rank of locations for new facilities was
created. Facilities built in these highest ranked locations would have the maximum impact by
supporting the largest population that is the furthest distance from medical facilities. The
resulting increase in medical accessibility could greatly decrease birth complications and
preventable death.

1

Chapter 1 Introduction
Pregnancy and childbirth are a vulnerable time for women, and anyone could fall victim to minor
complications, excessive bleeding, preterm birth, or infections. Mothers who have quick access
to trained medical care have a better chance of surviving preventable complications. Maternal
mortality prominently plagues Sub-Saharan Africa, as indicated by the United Nations Children's
Fund (formerly known as the United Nations International Children's Emergency Fund,
UNICEF) who have determined that the Sub-Saharan Africa region has the highest maternal
mortality rate in the world (UNICEF 2019). According to UNICEF, there is evidence that almost
all maternal deaths are preventable given the huge disparities in the mortality rates between
wealthy and impoverished nations. Figure 1 shows the primary causes of maternal death in Sub-
Saharan Africa, all of which are preventable if trained health services are accessible (Say et al.
2014). Indirect causes include anaemia, malaria, and heart disease (Nour 2008).

Figure 1 The Causes of Maternal Mortality in Sub-Saharan Africa. *Nearly all (99 per cent) of
abortion deaths are due to unsafe abortions. **This category includes deaths due to obstructed
labor or anemia. Source: Say et al. 2014

2

In 2016, the United Nations Population Fund, formerly the United Nations Fund for
Population Activities (UNFPA), reported that women who do not receive medical care during
pregnancy or at the time of birth are three to four times more likely to die during childbirth.
Understanding how to optimally implement new medical facilities throughout Sub-Saharan
Africa could prevent or reduce maternal mortality. In 2015, a meeting was conducted by the
United States Agency for International Development’s (USAID) global Maternal and Child
Survival Program (MCSP) where 72 participants from over 25 global health organizations,
government agencies, donors, universities, and other groups suggested that GIS mapping and
data visualization could play a significant role in ending preventable maternal mortality.
Examples suggested were mapping of hospital service areas of coverage based on distance or
identifying locations more at risk for maternal mortality. Mapping populations in areas not
practically serviced by a medical facility could enable policymakers to establish birthing centers
in these locations for women in need.
The Democratic Republic of the Congo (DRC), the focus of this research, is the largest
country within Sub-Saharan Africa and the fourth most populated country in Africa with a
population of approximately 87 million people, of which 44 million are female (United Nations
2019). Unfortunately, in 2015, the DRC also had the 10th highest maternal mortality rate of any
country in the world at 693 deaths per 100,000 births (World Bank 2015). This rate had dropped
from 879 deaths per 100,000 births in 1990, but it is still unnecessarily high.
High maternal mortality in the DRC is due in large part to pregnant mothers being
remotely located from medical facilities and routinely dying from preventable complications
such as hemorrhage, unsafe abortion, hypertension, and sepsis. Medical facilities in the DRC are
sparse and only located in major cities, leaving much of the population throughout the country to

3

choose between traveling by foot over a long distance or to deliver at home and hope everything
goes perfectly. Cars are not prevalent in the DRC, and the most common means of travel is by
foot due to the destruction of the infrastructure caused by the First and Second Congo Wars in
the late 1990s. Walking long distances during pregnancy or while in labor and especially at night
is a significant barrier for women seeking medical care. The lack of access to trained medical
care facilities is the direct link to maternal mortality death from preventable causes.
1.1. Objective
This objective of this study was to develop a simple GIS-based methodology for Sub-
Saharan Africa using readily accessible data that can be used to identify ideal locations for new
medical centers in order to increase accessibility to medical care. An increase in accessibility
will reduce incidences of preventable maternal death. The methodology was developed with a
focus on the Democratic Republic of the Congo (DRC) given its high maternal mortality rate
related to a lack of access to medical facilities. This research identified and ranked by need, areas
with the highest populations that are also the furthest distance from existing medical facilities
within the DRC. The ranked list could be used to target and support the populations in the most
need of medical services within the DRC. Further, the project sought to create a reproducible
methodology that can be applied in other nations with limited medical care accessibility.
1.2. Overview of Methodology
This study identified the isolated areas of the DRC unserved by present medical
facilitates, then identified and ranked by need, those isolated locations with the highest
populations that are also the furthest distance from existing medical facilities. The isolated areas
were identified by first determining regions of the country that are located within a maximum
three-hour walking distance from present medical facilities. This three-hour walking distance

4

service area takes into account straight-line distance as well as how land use type and elevation
affect travel time. A traditional method utilizing the Generate Service Area tool could not be
used as the infrastructure network had been destroyed and currently lacks functionality. As a
result, the Path Distance tool, along with a Cost Surface, was used to determine the service area.
After the part of the DRC that fell within the three-hour medical facilities service area
was identified, the remaining part of the country became the study area. With the unserved areas
of the DRC identified, a tessellation grid was generated to break up the country’s unserved area
into 210.44 sq km hexagons. The generated grid was intersected with a polygon layer of health
zones that had been joined with a health zones population table, the lowest level of aggregation
of DRC population data publicly available. Finally, the grid cells’ distances from existing
medical facilities were calculated utilizing the Near function.
Ranks in ascending order were created for each hexagon for both the grid distance from
existing medical facilities and the population that was assigned to each the grid polygon. The
combination of both of these ranks in descending order creates the needs-based rank for the
entire unserved parts of the country.
1.3. Structure of the Thesis
The remainder of this thesis is composed of 4 additional chapters. Chapter 2 discusses
background research on related topics such as causes of maternal mortality, spatial analysis of
maternal and newborn health, spatial distribution of medical facilities, and research related to the
methodology used such as cost-distance. Chapter 3 details the data utilized and the source from
which they were obtained. Chapter 3 also explains in detail the methodology used. Chapter 4
contains the results of the applied methodology on the designated study area. Chapter 5 is a

5

discussion of the limitations, potential improvements, and opportunities for further expansion of
this research.

6
Chapter 2 Related Work
This chapter explores research on related subject matters as well the methodology that was used
in this study. Subject matter related to this study includes the causes of maternal mortality,
spatial analysis of maternal and newborn health, and the spatial distribution of medical facilities.
Methodologies discussed that are related to this study are service areas, catchment area, and cost
distance.
2.1. Causes of Maternal Mortality
Maternal mortality is defined by the World Health Organization (WHO) as the death of a
woman while pregnant or in the postnatal period within 42 days of the end of pregnancy from
any cause related to or exacerbated by the pregnancy or its management, but not from accidental
or incidental causes (WHO 2019). Maternal mortality is a global issue but primarily affects more
impoverished nations due to the lack of access to medical facilities. In most cases, the reasons
are entirely preventable or recoverable with proper medical care.
The leading causes of maternal mortality in the DRC are primarily poverty in the country
and a general lack of the ability to pay for prenatal or childbirth medical expenses (The Guardian
2012). According to The Guardian article, even in cases where a woman was able to arrive at a
medical facility in time to obtain care for the birth, if she lacked funds, she may be repeatedly
transferred long distances between facilities. This report also noted that even if women have the
funds and can be admitted for care, preventable incidences of maternal death still occur.
Women in the DRC have six births on average in their lifetime, and the chance of
complications increases with each delivery (World Bank 2015). The pregnancy rate is primarily
due to lack of contraception and the lack education surrounding family planning. Due to the

7

limited use of contraception, multiple pregnancies that are closely spaced contribute to the high
maternal mortality rates in the country.
Additionally, according to the authors of “Socio-Cultural Factors in Maternal Morbidity
and Mortality,” women are likely to forgo utilizing medical services during pregnancy and
childbirth because of the cultural preference to engage the services of local traditional healers
(Okolocha et al. 1998). Local traditional healers offer cheaper and more accessible care than
modern medical facilities but do not have the medical training or supplies needed to save lives
when complications arise. A general lack of education regarding health care within the DRC
leads the majority of the population to believe traditional healers are more knowledgeable and
capable than medical doctors. The Okolocha et al. study showed that lack of education is a
primary contributing factor in maternal mortality. The more educated the mother is, the more she
can distinguish medical fact from superstition. Mothers with more education also tend to be
located in areas of the country with higher accessibility to healthcare.
Graham and Ronsmans (2006) point out that impoverished countries have had the least
success in the last 25 years in the global attempt to reduce preventable maternal mortality. The
impoverished nations have been unable to reduce the rate due to weak health care systems, high
fertility rates, and poor availability of data. The authors explain that poor and rurally located
mothers are most susceptible to maternal mortality. They found that the highest cause of
maternal death in Sub-Saharan Africa is severe bleeding or hemorrhaging along with
hypertensive diseases and infections, which are all preventable with access to trained medical
care. Access to nearby medical care increases the mother’s likelihood of utilizing care and
relates to a faster response time when complications arise. It was observed by Graham and
Ronsmans (2006) that women who have higher access to health care due to proximity have a

8

lower maternal mortality rate. This suggests that implementing birth centers in remotely
populated locations will increase the utilization of medical services and decrease maternal
mortality rates.
Further research adds strength to the argument that maternal mortality amongst poor rural
women in the DRC is directly related to lack of access to medical facilities. An analysis of World
Health Organization data on global maternal mortality trends by Khan et al. (2006) reveals that
while hemorrhage is the major contributor to maternal mortality, there are variations in causes at
the regional level. In less developed African nations, hemorrhage, sepsis, and hypertension are
the leading factors.
Finally, as reported by Ellison (2017), a lack of access to health care even in developed
nations is a contributing factor of maternal mortality. In more developed nations, the lack of
access to care is derived primarily from lack of financial means or insurance rather than being a
long distance from medical facilities.
2.2. Spatial Analysis of Maternal and Newborn Health
Researchers have begun to use GIS to analyze and map out maternal and newborn health.
One article by Ebner (2015) goes in-depth on spatial techniques used for analyses and
visualization such as thematic mapping, spatial analysis to assess accessibility, spatial modeling,
and small area estimation. These are all forms of analysis that could be used to both analyze the
data and display the results. The author noted that the biggest issue in maternal health analysis is
finding data that is aggregated at the lowest level possible as most data is typically published at
the national level only.
Spatial analysis has been used previously to identify communities that are at a higher risk
of a particular health issue. Chong et al. (2013) used spatial analysis to identify communities at a

9

higher risk of both smoking while pregnant and delaying the initial antenatal visit. The authors
reported that mothers that delay their initial prenatal visit are likely to experience adverse
pregnancy-related outcomes. Primiparous women (first-time mothers), young mothers, and
mothers with low socioeconomic status are likely to have delayed their initial antenatal visit.
Chong at al. began by georeferencing reported home locations of at-risk mothers as recorded at
the time of birth. After that, the authors utilized SaTScan to identify any spatial clusters that
showed significant correlation between smoking mothers and mothers with late initial antenatal
visits. The spatial clustering analysis revealed four areas that show correlated clustering. From
here, the authors were able to come to the conclusion that a higher proportion of at-risk mothers
lived in the most disadvantaged areas and a lower proportion were first time mothers.
A study by Molla et al. (2017) mapped the distribution of live births, pregnancies, and
population data in order to find a cause behind maternal mortality. Additionally, the authors
explained how mapping non-mortality factors that could support at-risk mothers, such as the
distribution of midwifery services, and mapping indicators such as the proportion of women who
need emergency obstetric care versus those that actually receive it, could reveal areas that
underutilize medical services and those that lack access due to low proximity. Mapping factors
and indicators such as these could enable governments to redistribute medical services to better
support a population.
2.3. Spatial Distribution of Medical Facilities
Understanding the spatial distribution of medical facilities is just as important as
understanding the distribution of the population being studied. The specific locations, the
distributions, and the density of medical facilities as it is related to population density all affect
the quality of healthcare available. Nwakeze (2011) compared the distribution of medical

10

facilities as well as the ratio of available physicians in each region to the population distribution
in Niger. The article argues that the sheer number of medical facilities is not what matters most.
More important is the condition and size of the medical facility as well as the quality and
quantity of health care professionals within it. Nwakeze blames a lack of organized leadership
and inefficiencies in implementing the developed medical policies for the lack of health care
availability. This is a similar issue in the DRC where three decades of civil war have left the
country with no public funding for medical care, weak national leadership, and zero regulations
enforced on the health sector leaving the population dependent on external aid (Ntembwa and
Lerberghe 2015).
Previous research has been conducted on the distance a mother is living from existing
medical facilities. Gabrysch et al. (2011) analyzed this relationship to find what influence the
travel distance had on the quality of health care these mothers received. The article reported that
the distance a mother is from medical facilities influences whether or not she is more likely to
choose to give birth at home unassisted. This is an issue that is at the core of the research in this
paper. Decreasing the travel distance to medical facilities and thus increasing the accessibility
will help increase a mother’s likelihood that she will seek out services when in need. Gabrysch et
al. also analyzed the household’s ability to pay for medical services based on assets owned. They
concluded that the inability to pay may influence expectant mothers’ willingness to travel for
medical assistance during birth.
It is optimal to have new locations for services be ideally located so that the most
remotely located childbearing aged female populations will have access. Ruiz (2010) discussed
the implementation of maternity waiting homes and their effectiveness in Mozambique, Africa.
Maternity waiting homes were created to supplement areas that were beyond a medical facility’s

11

service area. The maternity waiting homes were specifically created in order to increase the
number of births in medical facilities and decrease maternal mortality. Ruiz noted that while
these maternal waiting homes helped the populations that previously lacked access to medical
services, there are still women located in remote regions that are unserved and at risk for
complications.
GIS had been used previously to identify optimal locations for additional medical
facilities. Massey (2011) wrote that by just looking at the numbers, there may appear to be
enough health workers for a population, but this relationship is likely skewed due to distribution.
The author notes that the health workers and medical facilities are likely clustered in the most
urban areas of the country, leading to a disparity in coverage. Massey used spatial analysis with a
priority index to identify areas in need of trained midwives. His methodology calculated the
current number of midwives by region in Senegal and divided it by the WHO-recommended
number of midwives per population which is 1 midwife per every 300 women. Then that result
was subtracted from 1 and multiplied by 100 to result in the percent of shortage of midwives.
Another methodology Massey implemented in his article identified the priority area for
midwife service expansion. Massey did a hot spot analysis based on percent shortage of
midwives (at regional level), percent of childbirths occurring unassisted (at regional level),
percent of women that received no antenatal care (at regional level), and percent of childbirths
that take place at a health care facility. In his result, he was able to identify the eastern region of
Senegal as the area most in need of midwifery services.
2.4. Service Areas, Catchment Areas & Cost-Distance
A network service area is a region that encompasses all streets accessible from a point in
a designated amount of time. This analysis can be completed using a tool within ArcPro called

12

Generate Service Areas. Given a layer of points to start from, it can generate a map of the service
areas of these points on the network based on travel time selected. Catchment areas, on the other
hand, identify the locations from which individuals are attracted to access service or institutions.
Catchments are defined by multiple factors like distance or population size, or they may simply
partition the entire space into nearest neighbors. Becker (2016) utilized catchment areas to
understand the accessibility and availability of veteran healthcare services utilizing a two-step
floating catchment area (2SFCA) model. He elaborates on previous research done in this area by
incorporating variables such as wait times, patient satisfaction, and acceptability of care. In his
research, he was able to identify locations that do not meet the primary care service standards
designated by the federal government.
Accessibility surface models using cost-distance algorithms have also been used to create
catchment areas. Blanford et al. (2012) specifically addressed the ease of physical access to
healthcare in Niger in both the wet season and the dry season. It was stated that due to walking
being the primary form of transportation, access to medical facilities can be grossly
overestimated based upon the season. Blanford et al. discovered that in Niger, women in rural
areas walk on average 26 miles (about 42 km) to seek medical assistance. Walking 26 miles
would take a woman about 8 hours. They noted that the utilization of medical facilities
diminishes with both the travel distance, the quality of transportation, as well as road conditions.
This issue is similar to the situation within the DRC due to a lack of infrastructure from damage
caused by civil wars. Blanford et al. created an accessibility surface model based on distance
using a cost distance algorithm. They also created a friction surface for this calculation to
account for travel times of various travel methods across various surfaces. After that, using a
subnational subdivision of population, they determined the distribution of the population that fell

13

outside a four-hour travel distance. Blanford et al.’s study provided the framework for the
development of the methodology used in this study.
2.5. Summary
The related research in this chapter provided a brief overview on maternal mortality and
its causes and the uses and benefits of mapping maternal and newborn data. The complexity of
mapping the distribution of facilities increases when taking into account whether the amount of
trained medical personnel can support the population they are obligated to serve. There was also
a brief overview of service areas, catchment areas, and cost-distance to find service areas. This
study was unable to use network service areas as there is no traditional transportation
infrastructure network to generate it. It was also unable to utilize catchment areas, which are
designed for a population that is fully covered by services such that the catchment areas indicate
which service facility each individual belongs to (in the way that school zones are assigned to
house addresses). Blanford et al.’s (2012) study did provide some valuable guidance that is
discussed further below.

14

Chapter 3 Data and Methods
This study sought to identify optimal locations to implement additional medical facilities for
pregnancy and childbirth to support unserved areas of the DRC. This chapter provides details on
the data and methods used to determine these optimal locations.
3.1. Data
The data needed to conduct the analysis in the study area of the DRC was a point file of
the medical facilities, digital elevation data, land cover data, a table containing population data
by health zone, and health zone boundaries of the DRC. Table 1 summarizes the content of each
dataset used and lists its source.
Table 1 Data used in this study and sources
Data Format Content Source URL
Study Area Polygon Boundary of the
DRC
OCHA CHD https://data.humdata.org/data
set/drc-administrative-
boundaries-levels-0-2
Health
Zones
Polygon Boundaries of the
519 Health Zones
OCHA CHA https://data.humdata.org/data
set/dr-congo-health-0
Population
Data
Table Adult Female
population data
by health zone
OCHA CHD https://data.humdata.org/data
set/rdc-statistiques-des-
populations
Medical
Facilities
Point XY locations of
all 1,086 medical
facilities in the
DRC
Healthsites https://www.healthsites.io/#c
ountry-data
Elevation Raster 30-arc second
(1km) elevation
data
USGS https://www.usgs.gov/land-
resources/eros/coastal-
changes-and-
impacts/gmted2010?qt-
science_support_page_relate
d_con=0#qt-
science_support_page_relate
d_con

15

Data Format Content Source URL
Land Cover Raster Classified land
use raster
GLCC USGS https://www.usgs.gov/centers
/eros/science/usgs-eros-
archive-land-cover-products-
global-land-cover-
characterization-glcc?qt-
science_center_objects=0#qt
-science_center_objects.

All of the data obtained from United Nations Office for the Coordination of
Humanitarian Affairs (OCHA) Centre for Humanitarian Data (CHD) (centre.humdata.org) was
downloaded from their Humanitarian Data Exchange (HDX) website (data.humdata.org). While
the HDX is an open data sharing site and data quality varies, the CHD strives for data of
sufficient quality that is useful for humanitarian purposes. Importantly, OCHA intends that this
site will increasingly become a source for the often difficult to find socio-demographic data for
developing countries, such as sub-national level population data. However, for the purposes of
this project intended to demonstrate a methodology, data quality is not critical, so it is assumed
that the data retrieved from this site are of sufficient quality to develop the methodology.
3.1.1. Study Area
The study area was the country of the Democratic Republic of the Congo. The DRC is
located in central Africa and by area is the largest country in Sub-Saharan Africa. The country
boundary polygon for the DRC was obtained from the United Nations OCHA in West & Central
Africa. The country boundary was used primarily to narrow the scope of data obtained to the
DRC area. This boundary is shown in several of the maps below.

16

3.1.2. Health Zone Boundaries
The health system in the DRC is organized into 519 health zones which typically cover a
population of 100,000 to 200,000 people. Medical facilities are not equally distributed between
the zones or the population of the country as a whole. Like the study area boundary, the health
zones polygon data were obtained from OCHA in West & Central Africa. These data are updated
annually and were last updated on June 13, 2019. Figure 2 shows the 519 health zones within the
DRC.

Figure 2 The boundaries of the DRC and the 519 Health Zones. Source: OCHA
3.1.3. Population Table
The population data for the Democratic Republic of the Congo was also obtained from
OCHA. The table is updated annually and was last updated 17 September 2018. The table has

17

population data for the 519 health zones. The attributes of the table are health zone codes, health
zone names, and total population per zone as well as population totals for male, female, adult
male, adult female, boys, girls, and elderly by zone. Since this study is about maternal mortality,
only the adult female population attribute was utilized in the analysis. The table was joined to the
health zone polygons using the health zone code. Table 1 displays a sample of the adult female
population data obtained for the 519 health zones and Figures 3 and 4 show maps of the raw
adult female counts and the adult female density in the health zones, respectively.
Table 2 Sample of the adult female population data aggregated by health zone. Source: OCHA
Health Zone Health Zone Code Adult Female Pop
Aba CD5307ZS01 24184
Abuzi CD4304ZS01 14548
Adi CD5409ZS01 30686
Adja CD5409ZS02 23964
Aketi CD5204ZS01 25811
Alimbongo CD6105ZS01 41835
Alunguli CD6301ZS01 15179
Ango CD5207ZS01 19757
Angumu CD5407ZS01 31135
Ankoro CD7406ZS01 48172

18

Figure 3 Adult female population count by health zone. Source: OCHA

19

Figure 4 Adult female population density within the health zones. Source: OCHA
3.1.4. Medical Facilities
The existing medical facility location data was available for the Democratic Republic of
the Congo from a website called the HealthSites.io that specializes in mapping medical facilities
globally. According to their website, HealthSites’ mission is to establish accurate health care
location data that can be used to support first responders in the event of a natural disaster or
disease outbreak. HealthSites produces and manages the Global HealthSites Mapping Project,
which is an initiative to create an online map of every health facility in the world and make the
details of each location easily accessible. HealthSites noted on their website that their data has
proven extremely useful in the past, particularly during the Haiti Earthquake in 2010 and the
Ebola epidemic in West Africa from 2013 to 2016.

20

HealthSites imports health care location information from OpenStreetMap (OSM) and
other trusted partners such as Geomatica, Missing Maps, the International Committee of the Red
Cross, the International Hospital Foundation, Radiant Earth Foundation, cartONG, and Medecins
Sans Frontières. As Figure 5 shows, like other OSM crowd-sourced data validation workflows,
community users of HealthSites update and validate health care center data and then specialist
users validate the community user updates. The validated location data is then submitted back to
OpenStreetMap where HealthSites access the data. The data for medical facilities is consistently
being updated through crowd sourced users, trusted partners, and OSM field workers. Figure 5
shows HealthSites’ data lifecycle process, illustrating the care that is being taken to ensure that
data in the database is valid and accurate through multiple validations of the crowd sourced data.

21

Figure 5 Healthsite.io data lifecycle. Source: healthsites.io

22

The Global HealthSites Mapping Project provided a point shapefile of 1,086 sites, made
specifically for analysis within ArcGIS. The point shapefile attributes include the latitude and
longitude in decimal degrees with six decimal places, the type of facility (i.e. hospital or clinic),
and the name of the facility. A latitude and longitude in decimal degrees with six decimal places
offers a precision of 0.1 meters which is certainly precise enough to tag a building. Figure 6
displays the typical accuracy of a single point location of a medical facility.

Figure 6 Sample of the Medical Facility Data showing the accuracy of the reported point location
of the General Hospital of Banila
Figure 7 displays the location data for the current medical facilities. From first glance, it
is apparent that the 1,086 medical facilities are tightly clustered together and not evenly

23

distributed among the health zones or the DRC as a whole. Figure 8 shows a heat map of these
facilities that illustrates this clustering.

Figure 7 The locations of the 1,086 existing medical facilities within the health zones.

24

Figure 8 The 1,086 existing medical facilities displayed as a heat map to show the density of the
facilities. Source: OCHA, HealthSites
3.1.5. Elevation
The digital elevation data called the Global Multi-resolution Terrain Elevation Data
(GMTED2010) was obtained from the United States Geological Survey (USGS). This data set
was created in collaboration with the National Geospatial-Intelligence Agency (NGA) and was
obtained at a 30-arcsecond spatial resolution. The cell size is 0.0083 degrees (or approximately
928 meters at this latitude) and the datum is WGS 1984. The USGS and NGA’s sources for the
elevation data were the Digital Terrain Elevation Data (DTED) from the Shuttle Radar
Topography Mission (SRTM), Canadian elevation data, Spot 5 Reference3D data, and data from
the Ice, Cloud, and land Elevation Satellite (ICESat). Figure 9 shows the elevation distribution of

25

the DRC. The DRC is part of the Congo River Basin and a majority of the central portion of the
country is the low lying basin. This basin area increases in elevation as land moves east to the
inland part of the country. The basin area primarily consists of rainforests, rivers, terraces,
savannahs, and graslands.

Figure 9 Elevation data of the DRC. Source: USGS
3.1.6. Land Cover
Land cover data were obtained from the Global Land Cover Classification (GLCC)
created by the USGS in 1993. The land cover from the GLCC was used because the major land
types were already classified and just needed to be reclassified to suit this study. While this data
set is relatively old, it was easily acquired and deemed sufficient for this project given that the

26

focus of the study was on the methodology used to find optimal locations to implement medical
facilities.
The land cover raster is a representation of land cover based on the unsupervised
classification of 1-km AVHRR (Advanced Very High-Resolution Radiometer) 10-day NDVI
(Normalized Difference Vegetation Index) composites. These data were available both globally
and by continent, both with a spatial resolution of 1 kilometer. Figure 10 shows the land cover
trimmed to the polygon boundary of the DRC. To simplify the display, the values associated
with the 23 categories of land cover within the clipped area are displayed in Table 3. The original
dataset in its full fidelity can be found online at the source indicated in Table 1.

Figure 10 The GLCC raster trimmed to the boundary of the DRC. Source: USGS

27

Table 3 The classification of land use in the GLCC. Source: USGS
Value Description Value Description
1 Urban 47 Dry Woody Scrub
8 Bare Desert 51 Semi Desert Shrubs
14 Inland Water 56 Forest and Field
15 Sea Water 58 Fields and Woody Savanna
28 Montane Tropical Forests 59 Succulent and Thorn Scrub
29 Seasonal Tropical Forest 72 Mangrove
31 Crops and Town 78 Southern Hemisphere Mixed Forest
32 Dry Tropical Woods 91 Woody Savanna
33 Tropical Rainforest 93 Grass Crops
34 Tropical Degraded Forest 94 Crops, Grass, Shrubs
41 Hot and Mild Grasses, Shrubs 95 Evergreen Tree Crop
43 Savanna (Woods)

To use these data in the cost path analysis, it was necessary to reclassify the land cover
categories to reflect the difficulty of traversing across each type. As explained below in the
Methods section, the land cover data was manually reclassified based on Blanford et al.’s (2011)
work to a scale where the more difficult land cover to traverse was assigned a higher value. The
pixel size is 1000 meters and the datum is WGS 1984.
3.2. Methods
The following section describes in detail the methodology used to determine ideal
locations to implement medical facilities based on need. Figure 11 displays the generalized
workflow of the methodology used in this study. A complete model of the methodology can be
seen in Appendix A.

28

Figure 11 The workflow of the methodology used.
3.2.1. Prepare the Data
The population data table had eight columns of population data and only the adult female
population data was needed for this study. The health zone code, health zone name, and the adult
female population data columns were extracted. The columns were saved on to a new table and
imported into ArcPro. The population table was joined to the health zones polygon boundaries
by the health zone codes.
The layers were all obtained with a geographic coordinate system of WGS 84 and no
projected coordinate system. In order to perform the analysis, each layer was projected to WGS
1984 UTM Zone 33S which covers the DRC well.
The land cover data were reclassified using the Reclassify tool based on published travel
speeds through varying land types derived from Blanford et al. (2011). The travel speeds and
associated cell crossing times through various terrains used in this study can be seen in Table 4.
Since Blanford et al. did not include river travel in their model, an average speed on rivers was
assumed to be the typical speed a canoe could travel on a slow-moving river, 6 km/hr. The
average speed of 6km/hr was derived from an article by Munn (2006) in which he describes the
speed for an advanced paddler as 6km/hr, an intermediate paddler at 4-5km/hr, and a novice
paddler at 2-3km/hr (2006). The fastest speed was taken into account to determine the largest
three-hour service area a person would be capable of traveling regardless of ability. More
Prepare the Data
• Edit population table
• Join population table to
health zone polygons
• Reclassify land use raster
• Mosaic elevation rasters
Determine
Unserved Areas
• Path Distance tool used
to determine existing
medical facility service
areas
• Binary reclassification of
path distance result to
"served" and "unserved"
areas
Tesselate and
Determine
Population in
Unserved Area
• Generate tesselation
grid
• Tabulate intersection
against health zone adult
female population
Determine
Distance
• Near tool used to
calculate distance
Create Need-
Based Ranking
• Sort tool and field
calculator used to sort
and rank distances

29

difficult terrain to traverse has a higher cost while easy terrain has a lower cost value. Using the
travel times in Table 4, the value of the cell crossing time was determined by the cell size of
1000 meters and the speed to cross the land use type in the cell. Equation 1 shows the formula
used to identify the cell crossing time, t, and Equation 2 displays the equation if the 6 km/hr
terrain speed is applied for a single cell.
𝑡 = ( 1 ÷ 𝑆 𝑝 𝑒 𝑒 𝑑 ) × 𝑑 𝑖 𝑠 𝑡𝑎 𝑛 𝑐𝑒 (1)
𝑡 = ( 1 ÷ 6 ) × 1 (2)
Table 4 Land use reclassification based on travel speed. Source: Blanford et al. 2011
Land Cover Type
Average Speed
(km/hr)
Cell Crossing
Time (hr)
Rivers 6 km 0.16667
Open or sparse grasslands, croplands,
urban areas
3 km 0.33333
Shrubland, woodland, desert 1.5 km 0.66667
Lowland forest, swamp 1 km 1

The reclassified land use layer was then clipped to the extent of the DRC utilizing the
Clip Raster tool and the polygon layer of the DRC boundary.
The elevation data came as three separate rasters, so it was necessary to use the Mosaic to
New Raster function to merge the three rasters together to be used as one. The raster was then
clipped to the DRC boundary using the Clip Raster tool. The elevation raster then had to have the
cell sizes altered to match the land use raster. The Resample tool was used with the output cell
size set to the land use raster at 1000 meters. The bilinear resampling technique was used due to
the fact that bilinear interpolation calculates the value of each pixel by averaging (weighted for
distance) the values of the surrounding four pixels and it is suitable for continuous data.

30

3.2.2. Determine the Unserved Areas of the DRC
In order to determine the unserved areas, it was necessary first to identify the areas that
are served by medical facilities. The conventional approach, such as that used by Becker (2018),
for determining the service area of medical facilities using the ArcGIS Network Analyst
Generate Service Areas tool was not feasible due to a lack of a transportation network in the
country since its civil wars. The most common form of travel in the DRC is walking. Thus, a
service area around existing facilities of no more than three-hour walking distance was
determined to be the most feasible for a pregnant woman. At the fastest walking speed of 3 km
per hour (Blanford et al. 2011), this would be a maximum distance of 9 km. Using this logic, a
medical facility located at a distance greater than a three-hour walking distance during a
woman’s pregnancy or labor is deemed inaccessible.
The Path Distance tool was used to determine the area of the country that is beyond a
three-hour walk from any medical facility. The Path Distance tool determines the minimum
cumulative travel cost from a source, in this case a medical facility, to each cell location on a
raster while accounting for impedance by land use and slope. The reclassified land use raster
defines the impedance or cost to move planimetrically through each cell. The value at each cell
location represents the cost-per-unit distance for moving through the cell. Each cell location
value is multiplied by the cell resolution while also compensating for distance added by crossing
cells diagonally (where direction of travel is indicated by the tool’s pre-calculated backlink
raster) and moving across a slope (as determined by the tool’s pre-calculated slope, calculated
from the user supplied elevation raster), to obtain the total cost (i.e. time) of passing through the
cell. By incorporating the elevation raster, the Path Distance tool also takes into account the
actual surface distance that must be traveled as well as the difficulty of walking up or down a

31

slope by utilizing values from the Tobler’s Hiking Function which determines the difficulty of
walking on slopes. By setting river impedance at the highest speed (hence lowest time to cross a
river cell), this tool optimizes for river travel when available, which is a primary mode of
transportation within the DRC, secondary to walking.
In the absence of local knowledge, a number of assumptions were made regarding how
people move across this landscape. Boat travel on rivers was determined to be the fastest means
of travel as most rivers are navigable. Bridges were not taken into account in determining the
cost distance given the badly degraded condition of the transportation network. Any functional
bridges are most likely to be small foot bridges whose locations quickly change. It was also
assumed that crossing larger rivers by canoe would be relatively easy to accomplish within the
1 km resolution of a grid cell. Finally, based on the low impedance value given to rivers and the
prevalence of river travel in the DRC, the tool will choose river routes to minimize travel time by
balancing time on rivers against time spent walking overland.
The output path distance raster was Reclassified to either 1 (served) or 0 (unserved) to
indicate areas within a three-hour walking distance from the medical facility locations and areas
beyond a three-hour walking distance. The reclassified raster output of the path distance tool was
then converted to polygons using the Raster to Polygon tool. Next, the polygons with attribute 1
(served) were selected using Select By Attributes and then the selection was deleted. This new
layer was then restricted to the boundary of the DRC using the Clip tool. This result is the
‘Unserved Areas of the DRC’ and displays the extent of the country that is beyond a three hour
walk from any medical facility.

32

3.2.3. Generate Tessellation and Calculate Population
Next, the Generate Tessellation tool was used to create a grid of hexagons to cover the
extent of the Unserved Areas of the DRC. The tessellation grid size chosen was 210.44 sq
kilometers, which produces hexagons with a circumcircle radius of 9km. At a 3km average
walking speed, if any new medical facility were placed at the center of the hexagon it would be a
3 hour walk to the extent of the hexagon. Each hexagon serves as an individual potential service
area for any new medical facility. When the grid was generated, each hexagon was assigned a
GRID_ID with letters starting at AA being the columns and the rows were numbers.
The Clip tool was then used on the hexagons using the Unserved Areas of the DRC
polygon to limit the hexagon coverage to only the unserved areas. In doing this, the hexagons
that extend beyond the DRC boundary or cover the existing service areas would be omitted or
partially clipped resulting in 9,911 full or partial hexagons in the Unserved Areas.
Next, to determine an estimate of the adult female population within each hexagon the
Tabulate Intersection tool was used. The 9,911 hexagons that cover the Unserved Area of the
DRC were used as the Input Zone Features. The health zones that were joined with the adult
female population data were used as the Input Class features. The Adult Female Population was
designated as a sum field. The result is a generated table of the hexagons that list the estimated
adult female population count based on the percentage of area the hexagon occupies in a health
zone.
The 9,911 hexagons that cover the Unserved Area were joined with the generated table
based on the hexagon GRID_ID. Table 3 displays a sample of the adult female population data
by hexagon. The result is the original 9,911 hexagons but it now includes an attribute that is the
adult female population for the percent of the health zone the hexagon occupies.

33

Table 5 A sample of the calculated adult female population counts within each hexagon by
GRID_ID
GRID_ID Adult Female Pop
AA-70 36,224
AA-71 3,844
AA-72 1,409
AA-73 3,238
AA-74 2,893
AA-75 1,973
AA-76 1,315
AA-77 1,165
AA-78 1,717
AA-79 1,812

3.2.4. Determine Distance
To determine the ranking of hexagon locations that are not accessible medical facilities,
the Near tool was used to calculate the distance between the edge of each hexagon and the
closest medical facility point location. The Near tool added the nearest medical facility’s
ObjectID and the distance to the nearest medical facility to the hexagon grid attribute table.
3.2.5. Create Need-Based Ranking
Next, to assign ranks to individual hexagons, the Sort tool was used on the hexagon grid
layer attribute table to sort by distance from the nearest medical facility in ascending order. A
new field called Distance Rank was added to the hexagon grid attribute table, and the Field
Calculator was used to generate a sequential number based on the sorted near distance column.
The closest hexagon has a value of 1 and the furthest hexagon from a medical facility has the
highest number.
The Sort tool was used again but now to sort the population attribute in ascending order.
A field was again added and called Population Rank. The Field Calculator was used based on the

34

ascending sort order of the population to generate sequential numbers in the population rank
field. The lowest population count within a grid hexagon has the rank of 1 and the highest
population count has the highest rank. Thus, all hexagons are ranked by both distance and
population.
To create a final needs-based rank, it was necessary to decide what metric should be used
to combine these two rankings. There are many ways this could be done, many of them very
complex computationally, but for the demonstration purpose of this study, it was decided that a
mean of the two ranks would provide a value that balances the criteria evenly. Thus, to create the
final needs-based rank of the hexagons, a field was added to the layer’s attribute table called
Mean. The Field Calculator was used to calculate the mean of the Population Rank and Distance
Rank for each hexagon. The Sort tool was used on the Mean field, this time in descending order.
Since the previous sorts gave the highest population and the furthest distances the highest values,
these high values produce a high mean and these should be ranked highest (closer to 1) because
it means the hexagon would likely be far from medical facilities and have a higher population. A
final field was added to the attribute table called Needs-Based Rank and a sequential number was
populated in the Needs-Based Rank field based on the descending order of the Areas in Needs
field.
From the results, the top (lowest rank) location will be the most ideal for a new medical
facility. If ten, twenty, or fifty new medical facilities were to be added, the service area rank and
coverage should ideally be recalculated after the addition of each new medical facility to
optimize coverage. Within the selected hexagons, a facility can be constructed at the center point
as the center point is 9 kilometers, or a 3 hour walk, from the edge. Each hexagon is its own
service area. However, once constructed in one location, all adjacent hexagons would have a

35

much closer distance to an existing facility and in the combined ranking their ranks would drop
(i.e. grow larger). All the ranks throughout the unserved areas would adjust.

36

Chapter 4 Results
The intent of this study was to develop a methodology for identifying the optimal locations to
create new medical facilities or birth centers in order to reduce maternal mortality. The ideal
location to implement new medical facilities is based on population that could be served within a
three-hour walking distance service area and distance from existing medical facilities. This
chapter presents and discusses the results of each step in the methodology.
4.1. Intermediate Results
There were several stages in the methodology, and each produced an intermediate result.
These are shown in the following sub-sections.
4.1.1. Determine the Unserved Areas of the DRC
In order to identify the Unserved Areas of the DRC, identifying and eliminating the
service areas of the existing medical facilities needed to occur first. To identify the three-hour
walking distance service areas of the medical facilities the Path Distance tool was used. A cost
distance raster was prepared and reclassified based on walking speeds through various terrains
for use in the Path Distance tool. Figure 12 displays the results of the reclassification of the land
use based on travel speed. From the results, it is observed that most of the country is shown to
have either a 3km or 1km travel speed. The lack of variation in land use travel speed likely did
not alter the results in determining the 3-hour walking distance service area much more than a
simple Euclidean Distance calculation might have provided. However, this study is on the
development of a methodology using the DRC as the study area. The same methodology could
be used in an alternate country that has more variation in their land use travel speed.

37

Figure 12 The land cover raster reclassified based on travel speed.
The result of the Path Distance tool was a raster that was then reclassified to two
categories: within a three-hour walking distance and beyond a three-hour walking distance and
converted to polygons. This result is the ‘Unserved Areas of the DRC’ (Figure 13). This makes it
very clear the vast extent of the country that is beyond a three-hour walk from any medical
facility.

38

Figure 13 The results of the path distance tool display the areas of the DRC that are beyond a 3-
hour walk from a medical facility and are considered Unserved Areas in the DRC
4.1.2. Generate Tessellation and Calculate Population
The Generate Tessellation tool created a grid of 9,911 hexagons to cover the extent of the
Unserved Areas of the DRC (Figure 14).

39

Figure 14 The 9,911 generated hexagons covering the Unserved Area of the DRC.
After the hexagon grid was established for the Unserved Area of the DRC, the population
that could potentially be served within each hexagon was determined using the Tabulate
Intersection tool. After joining the intersection output table to the hexagon grid based on
GRID_ID, the results of the Adult Female population per hexagon can be seen in Figure 15.

40

Figure 15 Distribution of the estimated adult female population by hexagon.
The sum of the adult female population in the DRC is approximately 17.6 million while
the sum of the adult female population within the grids of the unserved area is 12.5 million.
These sums show that only about 5.1 million adult females in the DRC are within a three hour
walking distance of a medical facility, which is less than 1/3 of the country being served. Table 5
displays a sample of the adult female population data by grid. While a portion of the population
is located near medical facilities on the east side of the country, a large number of the adult
female population are located in remote rural areas of the south central.

41

4.1.3. Create Need-Based Ranking
Figure 16 is a visualization of the Adult Female Population Rank. A high rank in this
map indicates high population numbers. The hexagons with the higher population counts are in
yellow.

Figure 16 Visualization of the Adult Female Population Rank. Recall that rank closer to 1 in this
map indicates hexagons with the lowest population counts. The hexagons with the highest
population counts are in yellow.
4.1.4. Determine Distance
To determine the ranking of hexagon locations that are not accessible to medical
facilities, the Near tool was used to calculate the distance between the edge of each hexagon and
the closest medical facility point location. Figure 17 displays a visualization of the ranked
distance of hexagons from the existing medical facilities. The closest hexagon to a medical

42

facility has a rank of 1 and the furthest hexagon from an existing medical facility has a rank of
9,911. The hexagons that are the furthest distance from medical facilities are in yellow.

Figure 17 Visualization of the ranked distance of the 9,911 hexagons from the medical facilities.
A further distance is a higher rank. The hexagons with the furthest distance from medical
facilities are in yellow.
4.2. Final Needs-Based Ranking
The results of the overall ranking of the hexagons based on estimated adult female
population and the distance from existing medical facilities can be seen in Figure 18. Low ranks
mean a combination of high distance and high population. The results show that the areas
displayed in yellow, primarily the west and south-central portion of the country, are in the most
need for medical facilities.

43

Figure 18 Visualization of the hexagons based on need for a medical facility. The result is the
mean of the population rank and the distance rank.
Recall from Figure 15 that a large adult female population is present in the south-central
portion of the DRC. This area also appears as high ranking (low value rank) for need of a facility
in the resulting map in Figure 18 and would, thus, seem to be the optimal first place to
implement a new medical facility. Additional facilities might be evenly distributed within the
yellow areas of the need-based map in Figure 18.
As a way to evaluate the validity of the mean rank index used, histograms of the hexagon
distance and population data as well as a scatter plot of the relationship between the two were
evaluated. As would be expected, that the data was not evenly or normally distributed due
outliers in the data at the high end of both distributions. Figure 19 displays the distribution of the

44

distances of the hexagons from existing medical facilities. The majority of the hexagons are
between 20 and 125 km away and only a few are at much greater distances.

Figure 19 Histogram displaying the distance in kilometers of each hexagon from existing
medical facility.
The histogram of the distribution of the adult female population can be seen in Figure 20.
The data is not normally distributed due to outliers at the extreme high end where some hexagons
are located in very urban areas with high populations and a large number of hexagons with a
population estimate of under 1,000 due to large rural areas of the country.

Figure 20 Histogram displaying the estimated adult female population per hexagon

45

Figure 21 is a revaluation of the adult female population histogram with the adult female
population data normalized with a logarithmic transformation to account for the outliers at the
high end. The logarithmic transformation squeezes together the larger values in the data set and
stretches out the smaller values. Logarithmic transformations are used to make highly skewed
distributions less skewed. This can be valuable both for making patterns in the data more
interpretable and for helping to meet the assumptions of inferential statistics.

Figure 21 Histogram displaying the estimated adult female population per hexagon with the adult
female population data normalized
Scatter plots display the relationship between two variables. Figure 22 displays the
relationship between estimated adult female population and distance from existing medical
facilities. Due to the outliers at the high end of the data, the scatter plot is condensed and
practically unreadable. Figure 23 displayed the scatter plot relationship with one axis, the adult
female population, normalized using the logarithmic transformation so that the distribution could
be more easily viewed and understood.

46

Figure 22 Scatter plot displaying the relationship between the estimated adult female population
count within a hexagon and the distance a hexagon is from existing medical facilities. Colors
indicate the hexagons’ final ranks as displayed in the maps above.

Figure 23 Scatter plot displaying the relationship between the estimated adult female population
count within a hexagon and the distance a hexagon is from existing medical facilities. The Y axis
(adult female population) has been log transformed. Colors indicate the hexagons’ final ranks as
displayed in the maps above.
The last scatter plot in Figure 24 displays the relationship between the two variables with
both axes having a logarithmic transformation. The logarithmic transformation makes it evident
that the mean of the two sets of data produced the intended result. As both the distance and the
population increase (moving to the upper right), hexagons are correctly categorized in the highest
rank, being most in need of medical facilities. Areas that are closer with a smaller population are
in the bottom left in dark blue. The colors of the scatter plots reflect the final ranks of hexagons

47

displayed in Figure 18 where yellow is the most desirable location to build medical facilities and
the dark blue is the least desirable.

Figure 24 Scatter plot displaying the relationship between the estimated adult female population
count within a hexagon and the distance a hexagon is from existing medical facilities. Both axes
have been log transformed.

48

Chapter 5 Discussion and Conclusion
The objective of this study was to develop a simple methodology that could be used to identify
ideal locations for new birth facilities where large populations are the furthest distance from
existing medical facilities. This methodology was implemented in the study area of the
Democratic Republic of the Congo but could be used in any similar rural, largely undeveloped
country needing to identify areas in need of medical facilities.
The methodology developed was successful in identifying the areas of the DRC that are
currently unserved by existing medical facilities. There were a few reasons the path distance tool
was used to first identify the service areas of the existing facilities instead of just omitting the
hexagons that had an existing facility within it. First, the existing facilities would not fall
perfectly in the center of a hexagon and its service area may encompass parts of two or three
hexagons. Second, this was a methodology developed that could be used in other countries with
different parameters. If another country had a 5-hour walking distance service area for already
present medical facilities, it could be easily identified without altering the hexagon grid size used
here. Further, the study was successful in identifying areas that have the highest populations and
areas that are the furthest distance from existing medical facilities. Most importantly, the study
was successful in combining both distance from medical facilities and estimated population to
determine areas most in need of medical facilities.
The locations were identified through the generation of a tessellation grid over the areas
of the DRC with low walking accessibility to medical facilities. For each tessellation grid cell,
the distance to the nearest medical facility and the population within the cell were calculated.
Based on this combination of population size and distance from a medical facility, a rank of
locations for new facilities was created. Facilities built in these highest ranked locations would

49

have the maximum impact by supporting the highest population that is the furthest distance from
medical facilities. The resulting increase in medical accessibility could greatly decrease birth
complications and preventable death.
5.1. Lessons Learned
When I began this study, I initially chose the Democratic Republic of the Congo as my
study area due to the fact that the maternal mortality rate was so high. In hindsight, if I had done
more discovery into the country prior to committing so much time and effort into the research, I
would have discovered earlier that the entire country lacked a usable road network which
drastically altered my original plan of utilizing the Service Areas tool. Had I been able to use this
tool to identify the service network of a medical facility in say a state in the United States where
personally owned vehicles are a primary means of travel then the service areas would have been
more accurate and a lot easier to find. I made my research a lot more difficult by committing to
the country that chose as a study area.
Through the data discovery phase, it was revealed that the existing medical facilities do
not adequately service the population of the DRC nor do they provide any significant amount of
service area coverage. In a country that has been unable to reconstruct their road infrastructure in
the course of three decades since their civil wars and where mothers are routinely dying from
preventable causes, it would be expected that their current medical support is extremely and
critically limited. Such a detailed analysis was perhaps not needed for the DRC given the
enormity of need.

50

5.2. Improvements for Future Applications of the Methodology
A different approach could have been taken with the population data. The readily
available source population data was aggregated by the health zones and the population of each
hexagon was estimated based on the proportion of a health zone the hexagon occupied. There is
also a dasymetric-based population estimate available from Oak Ridge National Laboratory, the
LandScan population grid, which comes in 1 km spatial resolution and represents an ambient
population (average over 24 hours) distribution. This scale would have been smaller for a more
precise estimate, and since the dasymetric estimation accounts for areas that are unlikely to have
residents (such as lakes or preserved areas) it would have likely produced a more accurate
hexagon population estimate than the DRC health care areas. In using the population data at the
health zone level and estimating the population within the hexagons based on the percentage of
the health zone the hexagon occupies, no hexagons had zero population. This likely is not the
case as there are large areas of rivers and open greenspace that are not occupied. However, for
the proof of concept that was the goal of this study, the health zone population data was
sufficient and the more detailed categories in the health zones data allowed the analysis to focus
only on adult females, the target of maternal mortality reduction efforts.
Additionally, the data utilized for the medical facility location data was obtained
primarily from crowd sourcing. While this data goes through multiple verification processes, the
data at times is incomplete, could potentially be inaccurate, and the update rate of the data is not
regulated so a facility could close or open and it could be unreported. Another source of
validated medical facility data in Sub-Saharan Africa is the spatial database developed by Maina
et al. (2019). This comprehensive spatial database was assembled from government sources such
as Ministries of Health or government-developed Master Facility Lists. However, only public,

51

government-owned facilities are included in the Maina et al. database, unlike the healthsite.io
data that include private facilities. What type of data is more important to a particular study
should dictate which data is best to use. I chose to use the healthsites.io data, and while it might
not be the most accurate, it did provide the largest list of potential facilities. Since my goal was
to identify unserved areas of any medical facility, including private facilities and small clinics
was more important than only referencing government-owned public services.
The bridges and boat crossings for rivers were not factored in at any different travel time
because it was assumed that most bridges and boat crossings are not permanent locations in the
DRC. Most bridges are temporary structures made from natural materials. If this study were
conducted in a country with permanent bridge locations and dedicated ferry docking points, then
this could be factored into the cost surface when identifying service areas.
All of the major rivers were included in the classified land use raster. These were
assumed to be navigable at the same speed either when crossing or going up or down stream. To
improve the accuracy of the served areas, a horizontal factor that could account for the variations
in travel costs between the three modes of travel on the river might be added.
While the rivers were accounted for in the classified land use and river pixels were
assigned a fast travel speed, they are represented as disconnected pixels. To further improve the
analysis, a polygon layer of rivers could be used that has the accurate width of the rivers
accounted for. This could be merged with the land use raster so the pixels that intersect the river
polygon would obtain the river cost. This was not done in this methodology as only lines of the
DRC rivers were available, so depending on the land use raster to categorize the pixels on wide
rivers like the Congo was considered more reliable.

52

This methodology could be used for more than just identifying new locations for medical
centers. This could be used to identify new locations for any new facility such as a school, a
gym, or a retail store. The service area of existing locations would need to be identified and
removed from the study area to find the unserved areas. Then the targeted demographic could be
identified. The target demographic that has the largest population the furthest from any new type
of facility would be the new ideal location.
5.3. Value of this Research
From the results, the top location we be the most ideal for a new medical facility. If ten,
twenty, or fifty new medical facilities were to be added the service area rank and coverage
should ideally be recalculated after the addition of each new medical facility to optimize
coverage. Within the selected hexagons, a facility can be constructed at the center point as the
center point is 9 kilometers, or a 3 hour walk, from the edge. Each hexagon is its own service
area. However, once constructed in one location, all adjacent hexagons would have a much
closer distance to an existing facility and in the combined ranking their ranks would drop (i.e.
grow larger). All the ranks throughout the unserved areas would adjust.
High maternal mortality rates are caused in large part by preventable complications that
could be circumvented by increasing geographic access to medical facilities for pregnant
mothers. Walking long distances during pregnancy or while in labor and especially at night is a
significant barrier for women seeking medical care. Improving the conditions of roads alone
would not increase access to care as most of the population does not own a personal vehicle.
Establishing more healthcare facilities in the rural areas to increase access to care would be the
greatest benefit to women seeking care during pregnancy and labor. Establishing optimized
medical facility coverage through implementing new facilities based on need in a country like

53

the DRC requires a through consideration of all factors ranging from infrastructure, to the
location of the target population, and the current facilities.
According to historian Dan Snow, the DRC has the potential to be one of the richest
countries on earth, but colonialism, slavery and corruption have turned it into one of the poorest
(Snow 2013). As recounted in the Related Work chapter, the women in DRC face many perils
related to the high maternal mortality rate, with the level of poverty in the country being the
immediate cause of medical problems endangering maternal lives

(Ellison 2017). Expanding
medical services to the interior of the country to cater to the rural population would aid the adult
female population that lacks accessibility to the current medical facilities due to the poor
infrastructure.

54

References
Becker, Charles. 2016. “A Spatial Analysis of Veteran Healthcare Accessibility.” Thesis, Spatial
Sciences Institute. University of Southern California. https://spatial.usc.edu/wp-
content/uploads/2016/07/Becker-Charles.pdf.
Blanford, Justine I., Supriya Kumar, Wei Luo, and Alan M. Maceachren. 2012. “It’s a Long,
Long Walk: Accessibility to Hospitals, Maternity and Integrated Health Centers in
Niger.” International Journal of Health Geographics, 24, 11 (1): 1–15.
doi:10.1186/1476-072x-11-24.
Chong, Shanley, Michael Nelson, Roy Byun, Liz Harris, John Eastwood, and Bin Jalaludin.
2013. “Geospatial Analyses to Identify Clusters of Adverse Antenatal Factors for
Targeted Interventions.” International Journal of Health Geographics, 46, 12 (1): 1–10.
doi:10.1186/1476-072x-12-46.
Ebener, Steeve, Maria Guerra-Arias, James Campbell, Andrew J. Tatem, Allisyn C. Moran, Fiifi
Amoako Johnson, Helga Fogstad, Karin Stenberg, Sarah Neal, Particia Bailey, Reid
Porter, and Zoe Matthews. 2015. “The Geography of Maternal and Newborn Health: the
State of the Art.” International Journal of Health Geographics, 19, 14 (1): 1–10.
doi:10.1186/s12942-015-0012-x.
Ellison, Katherine, and Nina Martin. 2017. “Severe Complications for Women During
Childbirth Are Skyrocketing - and Could Often Be Prevented.” ProPublica. December 22.
https://www.propublica.org/article/severe-complications-for-women-during-childbirth-
are-skyrocketing-and-could-often-be-prevented.
Gabrysch, Sabine, Simon Cousens, Jonathan Cox, and Oona M. R. Campbell. 2011. “The
Influence of Distance and Level of Care on Delivery Place in Rural Zambia: A Study of
Linked National Data in a Geographic Information System.” PLoS Medicine 8 (1).
doi:10.1371/journal.pmed.1000394.
HealthSites. 2019. “About healthsites.io.” https://www.healthsites.io/#about.
Howard, Jacqueline. 2017. “Childbirth Is Killing Black Women, and Here's Why.” CNN. Cable
News Network. November 15. https://www.cnn.com/2017/11/15/health/black-women-
maternal-mortality/index.html.
Khan, Khalid S., Daniel Wojdyla, Lale Say, A. Metin Gülmezoglu, and Paul Fa Van Look. 2006.
“WHO Analysis of Causes of Maternal Death: a Systematic Review.” The Lancet 367
(9516): 1066–74. doi:10.1016/s0140-6736(06)68397-9.
Maina, Joseph, Paul O. Ouma, Peter M. Macharia, Victor A. Alegana, Benard Mitto, Ibrahim
Soce Fall, Abdisalan M. Noor, Robert W. Snow and Emelda A. Okiro. 2019. “A Spatial
Database of Health Facilities Managed by the Public Health Sector in sub Saharan
Africa.” Scientific Data 6 (1): 134.

55

Massey, Philip. 2011. “Reducing Maternal Mortality in Senegal: Using GIS to Identify Priority
Regions for the Expansion of Human Resources for Health.” World Health & Population
13 (2). doi:10.12927/whp.2011.22633.
Molla, Yordanos B., Barbara Rawlins, Prestige Tatenda Makanga, Marc Cunningham, Juan
Eugenio Hernández Ávila, Corrine Warren Ruktanonchai, Kavita Singh, Sylvia Alford,
Mira Thompson, Vika Dwivedi, Allisyn C. Moran, and Zoe Matthews. 2017.
“Geographic Information System for Improving Maternal and Newborn Health:
Recommendations for Policy and Programs.” BMC Pregnancy and Childbirth, 26, 17: 1–
7. doi:10.1186/s12884-016-1199-y.
Munn, Richard. 2006. “How Far Can We Go?” Canadian Canoe Routes - Wilderness Canoe
Association, July. https://www.myccr.com/articles/how-far-can-we-go.
Nour, Nawal M. 2008. “An Introduction to Maternal Mortality.” Reviews in Obstetrics and
Gynecology 1 (2): 77–81. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2505173/#.
Ntembwa, Hyppolite Kalambay, and Wim Van Lerberghe. 2015. “Democratic Republic of the
Congo: Improving Aid Coordination in the Health Sector.” World Health Organization.
https://apps.who.int/iris/bitstream/handle/10665/186673/WHO_HIS_HGF_CaseStudy_15
.4_eng.pdf;jsessionid=A85ED662A12F68FD5D5B8842F0D0B4D1?sequence=1.
Nwakeze, Ngozi M., and Ngianga-Bakwin Kandala. 2011. “The Spatial Distribution of Health
Establishments in Nigeria.” African Population Studies 25 (2): 680–96.
http://www.bioline.org.br/pdf?ep11047.
OCHA. n.d. “Quality Assurance Framework” Humanitarian Data Exchange.
https://centre.humdata.org/wp-
content/uploads/HDX_Quality_Assurance_Framework_Draft.pdf.
Okolocha, C., J. Chiwuzie, S. Braimoh, J. Unuigbe, and P. Olumeko. 1998. “Socio-Cultural
Factors in Maternal Morbidity and Mortality: a Study of a Semi-Urban Community in
Southern Nigeria.” Journal of Epidemiology & Community Health 52 (5): 293–97.
Ronsmans, C., J. F. Etard, G. Walraven, L. Hoj, A. Dumont, L. Bernis, and B. Kodio. 2003.
“Maternal Mortality and Access to Obstetric Services in West Africa.” Tropical Medicine
and International Health 8 (10): 940–48.
Ronsmans, Carine, and Wendy Graham. 2006. “Maternal Mortality: Who, When, Where, and
Why.” The Lancet 368 (9542): 1189–1200. doi:https://doi.org/10.1016/S0140-
6736(06)69380-X.
Ruiz, Ivan Zahinos. 2010. “Effectiveness Assessment of Maternity Waiting Homes in Increasing
Coverage of Institutional Deliveries Using Geographical Information Systems in Six
Districts of Cabo Delgado Province (Mozambique).” Masters of Public Health.
University of South Africa.
http://uir.unisa.ac.za/bitstream/handle/10500/4843/dissertation_ruiz_iz.pdf?sequence=1

56

Say, Lale, Doris Chou, Alison Gemmill, Özge Tunçalp, Ann-Beth Moller, Jane Daniels,
A. Metin Gülmezoglu, Marleen Temmerman, and Leontine Alkema. 2014. “Global
Causes of Maternal Death: a WHO Systematic Analysis.” The Lancet Global Health 2
(6): e323–e333. doi:10.1016/s2214-109x(14)70227-x.
Snow, Dan. 2013. “DR Congo: Cursed by Its Natural Wealth,” October 9.
https://www.bbc.com/news/magazine-24396390.
The Guardian. 2012. “Congo’s Maternal Mortality Rate Exacerbated by Poverty and Ignorance.”
https://www.theguardian.com/global-development/2012/may/11/congo-maternal-
mortality-rate-poverty.
United Nations. 2019. “World Population Prospects.” Population Division.
https://population.un.org/wpp/.
United Nations Population Fund. 2016. “Maternal Health.” https://www.unfpa.org/maternal-
health.
The World Bank. 2015. “Maternal Mortality Ratio (Modeled Estimate, per 100,000 Live
Births).” https://data.worldbank.org/indicator/SH.STA.MMRT?year_high_desc=true.

57

Appendix - Model Builder Diagram of the Complete Methodology

Abstract (if available)

Abstract The Democratic Republic of the Congo (DRC) is the fourth most populated country in Africa, with approximately 87 million people, of which 44 million are female. Unfortunately, it also has the 10th highest maternal mortality rate of any country in the world at 693 deaths per 100,000 births in 2015. High maternal mortality in the DRC is due in large part to pregnant mothers being remotely located from medical facilities and routinely dying from preventable complications. Cars are not prevalent in the DRC, and the most common means of travel is by foot due to the destruction of the infrastructure caused by the First and Second Congo Wars in the late 1990s. Walking long distances during pregnancy or while in labor and especially at night is a significant barrier for women seeking medical care. This study’s objective was to develop a simple methodology that could be used to identify ideal locations for new birth facilities where large populations are the furthest distance from existing facilities. The locations were identified through the generation of a tessellation grid over the areas of the DRC with low walking accessibility to medical facilities. For each tessellation grid cell, the distance to the nearest medical facility and the population within the cell were calculated. Based on a combination of population size and distance from a medical facility, a rank of locations for new facilities was created. Facilities built in these highest ranked locations would have the maximum impact by supporting the largest population that is the furthest distance from medical facilities. The resulting increase in medical accessibility could greatly decrease birth complications and preventable death.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A model for placement of modular pump storage hydroelectricity systems

PDF

An accessibility analysis of the homeless populations' potential access to healthcare facilities in the Los Angeles Continuum of Care

PDF

Analysis of future land use conflict with volcanic hazard zones: Mount Rainier, Washington

PDF

The geography of voter power in the U.S. electoral college from 1900-2012

PDF

Increase in surface temperature and deep layer nitrate in the California Current: a spatiotemporal analysis of four-dimensional hydrographic data

PDF

Spatial analysis of veteran access to healthcare in Los Angeles County

PDF

Providing a new low-cost primary care facility for under-served communities: a site suitability analysis for Service Planning Area 6 in Los Angeles County, California

PDF

A model for emergency logistical resource requirements: supporting socially vulnerable populations affected by the (M) 7.8 San Andreas earthquake scenario in Los Angeles County, California

PDF

Spatiotemporal visualization and analysis as a policy support tool: a case study of the economic geography of tobacco farming in the Philippines

PDF

Identifying areas of high risk for avian mortality by performing a least accumulated-cost analysis

PDF

Walkability study for school accessibility: case study of the San Juan, Puerto Rico elementary schools

PDF

Suitability analysis for wave energy farms off the coast of Southern California: an integrated site selection methodology

PDF

Different geographical representations of middle America

PDF

Explorations of American churchscape diversity

PDF

Modeling nitrate contamination of groundwater in Mountain Home, Idaho using the DRASTIC method

PDF

Designing an early warning system web mapping application for the Atlanta Metropolitan Area before a flooding event

PDF

Selection of bridge location over the Merrimack River in southern New Hampshire: a comparison of site suitability assessments

PDF

Out-of-school suspensions by home neighborhood: a spatial analysis of student suspensions in the San Bernardino City Unified School District

PDF

Preparing for earthquakes in Dallas-Fort Worth: applying HAZUS and network analysis to assess shelter accessibility

PDF

A site suitability analysis for an inland port to service the ports of Los Angeles and Long Beach

Asset Metadata

Creator Aazami, Lindsay Lynette (author)

Core Title Reducing maternal mortality by improving medical facility accessibility: a methodology demonstrated for the Democratic Republic of the Congo

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/18/2019

Defense Date 08/28/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag DRC,GIS,health,maternal mortality,OAI-PMH Harvest

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen K. (committee chair), Vos, Robert O. (committee member), Wilson, John P. (committee member)

Creator Email LAazami@usc.edu,LindsayAazami@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-217816

Unique identifier UC11673731

Identifier etd-AazamiLind-7821.pdf (filename),usctheses-c89-217816 (legacy record id)

Legacy Identifier etd-AazamiLind-7821.pdf

Dmrecord 217816

Document Type Thesis

Rights Aazami, Lindsay Lynette

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA