Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Use of GIS for analysis of community health worker patient registries from Chongwe district, a rural low-resource setting, in Lusaka Province, Zambia
(USC Thesis Other)
Use of GIS for analysis of community health worker patient registries from Chongwe district, a rural low-resource setting, in Lusaka Province, Zambia
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
USE OF GIS FOR ANALYSIS OF COMMUNITY HEALTH WORKER PATIENT
REGISTRIES FROM CHONGWE DISTRICT, A RURAL LOW-RESOURCE
SETTING, IN LUSAKA PROVINCE, ZAMBIA
by
Mine Metitiri, MPH, CPH
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 2014
Copyright 2014 Mine Metitiri
TABLE OF CONTENTS
DEDICATION........................................................................................................................... v
ACKNOWLEDGEMENTS .................................................................................................... vi
LIST OF TABLES .................................................................................................................. vii
LIST OF FIGURES ............................................................................................................... viii
LIST OF ABBREVIATIONS ................................................................................................. ix
ABSTRACT .............................................................................................................................. xi
CHAPTER ONE: INTRODUCTION ..................................................................................... 1
Motivation .......................................................................................................................... 3
CHAPTER TWO: BACKGROUND....................................................................................... 8
2.1 Geography .................................................................................................................... 8
2.1.1 Physical .................................................................................................................. 8
2.1.2 Climate ................................................................................................................... 9
2.2 Population ................................................................................................................... 10
2.3 Disease Burdens ......................................................................................................... 12
2.3.1 Malaria ................................................................................................................. 13
2.3.2 Pneumonia Infections........................................................................................... 19
2.3.3 Diarrheal Diseases ............................................................................................... 20
2.4 Mobile Health (mHealth) Applications .................................................................... 21
2.5 GCC mHealth Study in Chongwe District, Zambia ............................................... 28
2.5.1 Overview .............................................................................................................. 28
2.5.2 Study Area ........................................................................................................... 29
2.5.3 Population ............................................................................................................ 30
2.5.4 Community Health Volunteers ............................................................................ 30
2.5.5 Data Collection .................................................................................................... 31
2.5.6 Outcomes ............................................................................................................. 32
CHAPTER THREE: METHODOLOGY ............................................................................ 33
3.1 Datasets ....................................................................................................................... 33
3.1.1 GCC Dataset ........................................................................................................ 33
3.1.2 CSO Administrative Area Shapefiles .................................................................. 39
3.1.3 CSO Population Datasets ..................................................................................... 41
3.1.4 District Health Facility Dataset ............................................................................ 42
3.2 Variables ..................................................................................................................... 42
3.3 Analysis ....................................................................................................................... 49
3.3.1 ArcGIS Data Preparation ..................................................................................... 49
3.3.2 Population Density Distributions ......................................................................... 50
3.3.3 Illness Case Count Distributions .......................................................................... 50
3.3.4 Incidence Rates .................................................................................................... 51
3.3.5 SaTScan v9.3 Cluster Analysis ............................................................................ 51
CHAPTER FOUR: RESULTS .............................................................................................. 55
4.1 Spatial Epidemiologic Maps ..................................................................................... 55
4.1.1 Population Density Maps (Ward and SEA) ......................................................... 55
4.1.2 Illness Case Count Distributions .......................................................................... 57
4.1.3 Incidence Rates .................................................................................................... 58
4.2 Cluster Analysis Results ............................................................................................ 61
4.2.1 Fever .................................................................................................................... 61
4.2.2 Malaria ................................................................................................................. 62
4.2.3 Diarrheal Illnesses ................................................................................................ 65
4.2.4 Pneumonia............................................................................................................ 68
4.3 Limitations.................................................................................................................. 71
CHAPTER FIVE: DISCUSSION .......................................................................................... 73
CHAPTER SIX: CONCLUSIONS ........................................................................................ 84
6. 1 Policy Implications.................................................................................................... 84
6.2 Future Research ......................................................................................................... 86
REFERENCES ........................................................................................................................ 88
APPENDICES ......................................................................................................................... 94
Appendix A: Population Density Maps ......................................................................... 95
Appendix B: Disease/Illness Count Proportional Symbols Maps ............................... 98
Appendix C: Disease/Illness Incidence Maps .............................................................. 108
Appendix D: SaTScan v9.3 Cluster Analysis Maps .................................................... 118
v
DEDICATION
This is dedicated to the Community Health Workers and Volunteers serving on the frontlines
of healthcare delivery in Zambia and developing regions globally.
"Everybody can be great, because anybody can serve. You don't have to have a college degree
to serve. You don't have to make your subject and verb agree to serve. You only need a heart
full of grace. A soul generated by love."
- Dr. Martin Luther King Jr.
vi
ACKNOWLEDGEMENTS
I would like to acknowledge my committee members Dr. Darren Ruddell and Dr. Daniel
Warshawsky. Many thanks to my mentors Dr. Travis Longcore for his support and guidance
on this project and Dr. Pascalina Chanda-Kapata for her guidance also and whose work in
Chongwe district served as the inspiration for this thesis.
Thank you to my family and friends for their constant support and encouragement throughout
this process, without whom I would not have gotten this far, most especially William Ngosa,
Jessie Pinchoff, and Colin Mattison for sharing their data analysis and GIS expertise when
answering my questions and Dr. Dela Kusi-Appouh for reading all of my drafts. I am forever
grateful.
Thank you also to the Zambian Ministry of Health staff at all government levels, the staff at
the Central Statistical Office in Lusaka, the Surveyor General’s Office in Lusaka, and PATH’s
Lusaka offices for sharing their datasets and providing information.
vii
LIST OF TABLES
Table 2.1 Leading Causes of Death in Zambia for Non-communicable and Communicable
Diseases..................................................................................................................................... 13
Table 2.2 Epidemiology of prevalent malaria species in Zambia .......................................... 14
Table 3.1 GCC Health facilities included in GCC geospatial dataset .................................... 38
Table 3.2 GCC Geodataset with census data and village locations identified using PATH
database. .................................................................................................................................... 43
viii
LIST OF FIGURES
Figure 1.1 Zambia is a sub-Saharan country located in the Southern Hemisphere of Africa .. 1
Figure 2.1 Lusaka Province (green) is the location of Chongwe district is located (orange).*9
Figure 3.1 PATH log in an Access database containing latitude and longitude coordinate
locations for Chongwe villages ................................................................................................. 36
Figure 3.2 PATH log in Access database containing the geocoded village names used
matching village names............................................................................................................. 37
Figure 3.3 GCC spatial dataset Excel spreadsheet with completed health facility and village
coordinates ................................................................................................................................ 39
Figure 3.4 Zambia administrative boundary areas in descending hierarchal order ............... 40
Figure 4.1 Chongwe district population densities for Ward administrative areas ................. 56
Figure 4.2 Chongwe population densities for SEA administrative boundary areas .............. 57
Figure 4.3 Proportional Symbol Map of case counts ............................................................ 59
Figure 4.4 Chongwe disease/illness incident rate estimates .................................................. 60
Figure 4.5 Fever Incidence Cluster Areas .............................................................................. 62
Figure 4.6 Malaria Incidence Cluster Areas .......................................................................... 63
Figure 4.7 Malaria with Fever Incidence Cluster Areas ........................................................ 64
Figure 4.8 Diarrhea Incidence Cluster Areas ......................................................................... 66
Figure 4.9 Vomiting Incidence Cluster Areas ....................................................................... 67
Figure 4.10 Severe Diarrhea Incidence Cluster Areas ............................................................ 68
Figure 4.11 Pneumonia Incidence Cluster Areas .................................................................... 69
Figure 4.12 Severe Pneumonia Incidence Cluster Areas ........................................................ 70
Figure 5.1 Population densities for Chongwe SEA areas and RHC locations. ..................... 77
Figure 5.2 Exploratory map of disease/illness counts........................................................... 78
ix
LIST OF ABBREVIATIONS
ACT Artemisinin-base Combination Therapy
CD Communicable Diseases
CHW Community Health Worker
CSO Central Statistical Office
DHIS2 District Health Information System version 2
eHealth Electronic Health
GDP Gross Domestic Product
GIS Geographic Information System
GIST Geographic Information Systems and Technology
ICCM Integrated Childhood Case Management
ICT Information and Communication Technologies
IRS Indoors Residual Spray
ITNs Insecticide Treated Nets
LLINs Long-lasting Insecticide Nets
LMICs Low-Middle Income Countries
LR Likelihood Ration
mHealth Mobile Health
MOH Zambia Ministry of Health
NCD Non-communicable Diseases
NGO Non-Governmental Organization
NMCP National Malaria Control Program
ORS Oral Rehydration Solution
PDA Personal Digital Assistant
RDT Rapid Diagnostic Test
x
RHC Rural Health Center
RS Remote Sensing
SEA Standard Enumeration Area
UN United Nations
UNICEF United Nations Children’s Fund
WHO World Health Organization
xi
ABSTRACT
The growing accessibility of mobile phones in developing countries has led to increased innovation and
utilization of handheld technology in managing health outcomes. Mobile health (mHealth) technologies
enabled significant gains in localized data collection methods and increased timeliness in disease
surveillance and control programs. Mobile technology has become an important tool for point of care
productivity and effective task shifting for Community Health Workers (CHWs) in many developing
countries. Concurrently, GIS technology has increasingly been utilized in public health research,
planning, monitoring, and surveillance within many developing countries and low-resource settings.
This has resulted in opportunities for better understanding of spatial variation of diseases and the
correlations with environmental factors.
To better understand community needs and burden of illnesses managed by CHWs, a geospatial
analysis at the sub-district level was performed on CHW catchment area health data registries. Risk
assessments and cluster analyses were conducted to identify high incidences of fever related illnesses
for malaria, diarrhea, and pneumonia in community areas within the rural district area of Chongwe,
Zambia. Seventy CHWs recorded 7,674 cases over a time-period of ten months, of which 3,130 cases
were geocoded for geospatial analyses. One hundred forty-one village areas within 15 rural health
center catchment areas were geocoded and mapped. Results were used to create thematic maps
illustrating disease distribution and risks for malaria, pneumonia, and diarrheal illnesses for each sub-
district village area manage by CHWs. The use of mobile technology integrated with GIS to manage
community health data and the application of GIS to analyze community level data may provide further
insight into local area disease distribution, variability, and community needs than systems focused
solely on district level data analysis and lacking GIS integration.
1
CHAPTER ONE: INTRODUCTION
In Zambia, like in many other developing countries, health care workers are in short supply.
Community Health Workers (CHW) compensate for shortages in health care workers and serve
as the backbone of health care service delivery (Braun et al. 2013; Ministry of Health NHSP
2011-2015). Zambia is a sub-Saharan low-income country located in the Southern hemisphere
of the African continent Figure 1.1.
Figure 1.1 Zambia is a sub-Saharan country located in the Southern Hemisphere of Africa
Advances in information and communication technologies (ICT) have provided
opportunities for improving community health services delivery both in Zambia and across
developing regions. The introduction of ICT devices such as computers and mobile phones, and
2
training of CHW to use such technologies, has strengthened CHW capacity. It has also opened
up opportunities for improving the quality of health service delivery at a community level.
Zambian CHW responsibilities include community health data management and reporting of
disease burdens to local health centers. The introduction of mobile phones as a tool for
community-based case management has led to timely reporting to health facilities and patient
feedback (Kamanga et al. 2010, Blaya et al. 2010). The use of mobile devices has been most
beneficial for CHW working in rural areas and within communities located far from the district
health centers. ICTs are an example of how technology can be used to improve service delivery
in rural parts of Zambia.
In Chongwe district, CHWs were provided ICT devices during a pilot program. The aims
of the program were to determine the feasibility of usage in a rural setting and to evaluate the
effects of mobile devices on CHW activity and retention, health services, and health status of the
community. Information collected by each CHW was aggregated and reported on a weekly basis
to district health centers. In this reporting scheme, health information is aggregated via a bottom
up hierarchy converging at the national level to provide evidence-based support for new health
policies and eventual policy dissemination of national health strategic plans back down to the
districts. The challenge with this hierarchal reporting system is that information often travels
slowly between various levels of government, and is often non-responsive to the individual
community level needs. Additionally, community level information provided by CHWs is
attenuated as a result of aggregation of sub-district level data to district and provincial centers
(Freifeld et al. 2010). Furthermore, the current reporting system is labored and time intensive
during outbreak investigations (Kamanga et al. 2010). It is often difficult to evaluate the overall
CHW contribution towards disease identification, treatment of illnesses, and in reducing disease
burdens.
3
The use of ICT devices for management of CHW registries containing disaggregated
community health data and GIS to analyze these datasets may provide insights into local spatial
variations in disease distributions and treatments. It would also potentially provide feedback on
individual CHW performance along with individual needs associated with illness burden and
availability of health resources. Furthermore, it would allow for evaluations of CHW
effectiveness in addressing community health needs. An exploration of CHW registries using
GIS coupled with the existing ICT technology would help provide evidence for local health
managers to be more responsive to CHW needs in the future. It may also provide evidence to
support policies that allow for more targeted resource allocation of limited health commodities
by identifying areas of high need (Kamanga et al. 2010).
Motivation
Zambia is a malaria endemic country. Malaria is fever related illness that is both
preventable and treatable, and wherever environmental controls fail, the need for rapid response
and effective treatment is imperative. Malaria annually affects more than 4 million Zambians and
accounts for 30% of outpatient visits. It is the primary cause of nearly 8,000 deaths annually
(UNICEF Zambia 2014). The most vulnerable populations susceptible to malaria are those
located in remote or rural areas, pregnant women, persons who are immuno-compromised, and
children under the age of 5. These groups are at an increased risk of death attributable to malaria,
with 20% of deaths occurring among pregnant women and 35–50% percent of deaths occurring
in children under 5 (UNICEF Zambia 2014). Zambia’s National Malaria Strategic Health Plan
for 2011–2015 cites the scale up of malaria interventions, including prevention and treatment
services, as a main focus area.
Zambian CHW serve in the frontlines of health service delivery, which includes malaria
treatment and management at the community level, especially in rural settings. CHW are
4
equipped with the knowledge and skills to treat malaria at its various stages either by referring
severe cases to the nearest health facility or by treating less severe cases with anti-malarial drugs
(Chanda et al. 2011). Treatment starts with the early recognition of symptoms and immediate
action upon the onset of illness. Having access to CHW for appropriate diagnosis and treatment
reduces the likelihood of delayed action and intervention in rural communities.
The introduction of rapid diagnostic test (RDT) in Zambia in the last 5 years has led to
more efficient and accurate confirmation of malaria cases. This has culminated into timely and
more cost effective malaria case management by CHWs, especially among those living within
more remote areas (Chanda et al. 2011). Yet despite the availability of RDTs, low rates of
routine usage and adherence to diagnostic guidelines for malaria confirmation have been
documented (Chipwaza et al. 2014; Kamanga et al. 2010). These findings may be a result of
inadequate training of CHWs or stock outs of RDTs within an area as a result of lapses in the
health system infrastructure. Stock outs are sudden shortages in availability of healthcare
commodities such as medicines, diagnostic tools, and equipment. This could be attributed to a
number of factors including poor logistical planning, weak health system frameworks,
manufacturer shortages or limits on production, and expiration of products. Stock outs of health
commodities in Africa are a common occurrence and in areas where RDT stock outs frequently
occur, CHW capacity is reduced to reliance on symptomatic diagnosis for malaria using the
onset of fever to determine probable malaria cases (Mayando et al. 2014; Derua et al. 2011).
Symptomatic diagnosis for malaria has often led to inaccurate diagnosis and over-
treatment for malaria. Furthermore this has resulted in unnecessary costs and side effects
associated with the use of anti-malaria drugs and possible exaggeration of local malaria disease
burden (Derua et al. 2011). Equally, non-febrile malaria cases may be under-diagnosed often
leading to progression into more severe non-malaria cases, which would then lead to delayed
5
treatment and death. Most important, incorrectly attributing fever symptoms to malaria increases
the risks of CHW missing diagnosis and treatment of other potentially life threating causes of
febrile illness, such as bacterial infection, cholera, typhoid, and pneumonia (Chipwaza et al.
2014; Mayando et al. 2014). It has been documented that non-malarial febrile illnesses result in
higher childhood mortality across malaria-endemic countries than malaria (Black et al. 2010).
The increased likelihood of population level drug resistance over time, which is associated with
over prescription of anti-malarial drug treatments, is also a consideration.
These scenarios are more likely to occur, within remote malaria endemic areas where
malaria control interventions are poor, RDT use is inadequate or consistently unavailable, and
the existence of overlapping febrile illness outcomes is common. It is for these reasons that
routine disease surveillance and programmatic evaluation should be performed at the community
level for all febrile illnesses and malaria management. The use of GIS to conduct a geospatial
assessment of non-malarial disease/illness burdens in malaria endemic areas at the community
level could help stakeholders to better understand better the local variations in overall disease
burdens.
Additionally, the use of GIS on health data would lead to adequate management of non-
malaria fevers and further exploration into causes beyond the generalized assumption of malaria
as the primary source of febrile illness in endemic areas (Acestor et al. 2012). Evaluation of
CHW-generated data for community health management has the potential to provide greater
insight into local spatial variations in disease burdens and illness case management at a local
level. This would allow for further exploration into disease/illness incidence for evidence-base
support for future policies encouraging targeted distribution of healthcare commodities and
resources. Future research could then explore the feasibility and costs associated with such a
targeted approach as well as provide evidence to amend existing policies and provide a
6
framework for implementation strategies. Moreover, areas where strengthening of interventions
are needed or where additional CHW trainings are needed for management of non-malarial
febrile illnesses could be identified. Resulting to better implementation of programs and
additional understanding of factors associated with increased risk of disease burdens and
illnesses at the community level.
This thesis describes the application of GIS and spatial scan statistics to assess illness
distribution and CHW illness case burdens at the sub-district level in Chongwe, Zambia. This
geospatial analysis of CHW registries explored community level illness distribution to identify
clusters of three fever related illnesses, malaria, pneumonia, and diarrhea. The analyses served to
illustrate the benefits of sub-district area level data analyses in order to provide a better
understanding of the extent to which febrile illness cases exists throughout the district. The use
of GIS to perform spatial analyses of malaria and other febrile illnesses at the community level
opens up opportunities for in-depth exploration into areas where case management can be
improved, and diagnostic commodities would have the greatest impact on malaria surveillance
and control efforts within this malaria endemic district.
The following thesis chapters will discuss the application of GIS on CHW health
registries to seek answers to the following questions:
1. Are CHW health registries an underused source of information about sub-district level
standards of care and health outcomes?
2. Can existing CHW health registries be used to create disaggregated, sub-district level,
low-resolution geo-datasets suitable for geospatial health data analyses?
3. Can the use of GIS on low-resolution health geo-datasets provide adequate spatial
insights into CHW needs and community health outcomes at a sub-district level?
7
The hypothesis is that CHW registries are an underutilized source of community health
information and can be used to identify areas of local disease/illness distribution that may be
masked by current data aggregation methods. A secondary hypothesis is that if GIS were to be
integrated in mHealth initiatives at the community level, it would have a great impact on disease
and illness surveillance at the community level. Timely monitoring of disease/illness variation
could occur which would allow districts to be more efficient in outbreak response and strategic
in allocation of resources using this technology.
The ensuing chapters are organized as follows. Chapter 2 provides background
information on the geographic area and population distribution of Zambia. It also elaborates on
the research questions and hypotheses that guide this study. Chapter 3 examines the primary
data sources and secondary methods of analysis. Chapter 4 presents findings of geospatial data
visualization and spatial cluster analyses. It also discusses the limitations of the findings.
Chapter 5 provides an in-depth discussion of the findings of spatial cluster analyses. Chapter 6
provides a discussion on the impact of research findings on policy development. It also focuses
on final conclusion and suggestions for future studies.
8
CHAPTER TWO: BACKGROUND
2.1 Geography
2.1.1 Physical
Zambia is a country located between the latitudes of 10° to 18° South and longitudes of
20° to 30° East in the southern region of the continent of Africa. It is a large landlocked country
with a land area of 752,618 square kilometers that is bordered by the countries of Malawi,
Zimbabwe, Angola, Tanzania, Botswana, Democratic Republic of Congo, and Mozambique.
Zambia’s terrain is mostly high plateau broken up by small hills and isolated mountain ridges
(CIA World Factbook 2013). Most of the land is flat, with an average elevation ranging between
3500ft to 4500ft above sea level. The lowest elevation is in the Southern region. The capital city
is Lusaka and has an elevation of 4,265ft above sea level (CIA World Factbook 2013; Zambia
Tourism 2014).
Wetlands account for approximately 5% of Zambian land cover. Lakes Bangweulu,
Mweru, and Tanganyika are the three most important natural lakes of the country. The total
water area for the country is 9,220 square kilometers. Zambia lies on the watershed between the
Democratic Republic of Congo and Zambezi River systems. The upper part of the Zambezi
River and its major tributaries, of which the Kafue and Luangwa Rivers are the largest, divide
the plateau into several large valleys (Zambia Tourism 2014). Communities located on or near
these water areas have higher relative risks for vector-borne illnesses such as malaria,
onchocerciasis, trachoma, and for water born bacterial or parasitic illnesses like cryptosporidium,
giardiasis, and cholera illness outcomes.
Zambia is administratively divided into ten provincial areas, namely: Central, Copperbelt,
Eastern, Luapula, Lusaka, Muchinga, Northern, North Western, Southern, and Western
provinces (Figure 2.1). At the time of the 2010 census data collection, there were 74 district
9
areas, 150 constituencies, and 1,430 wards. The government is comprised of Central and Local
Governments with Lusaka city serving as the central level of government. Lusaka province,
highlighted in green, is where Chongwe district is located and can be seen in the orange area in
Figure 2.1.
Figure 2.1 Lusaka Province (green) is the location of Chongwe district is located (orange).*
*Muchinga Province is the 10
th
most newly created Province and is not represented in the inset map.
2.1.2 Climate
Zambia has a tropical climate, with most of country classified as humid subtropical or
tropical wet and dry, which is modified by elevation. There are two main seasons, the rainy
season and dry season. The rainy season typically begins between November/December and
10
extends through to March/April, corresponding with the summer season. The dry season
typically occurs between the months of May and August, corresponding with the winter season.
The dry season is divided into hot and cold dry seasons, with the hot season occurring between
the months of September and October/November, and the cold season occurring between May
and August. The average monthly temperature throughout the country is 68°F, with higher
temperatures occurring in lower elevation areas and valleys (Zambia Tourism 2014). Changes in
climate and temperature have been known to be associated with increased incidence of malaria,
cholera, respiratory infections, dysentery, and lymphatic filariasis in Zambia (Kasali 2008; Slater
and Michael 2012). For many vector-borne diseases, climate is a key factor for disease
prevalence. Changes in the raining season have resulted in more incident cases of malaria
illnesses in many district areas.
2.2 Population
The total population estimate for Zambia is 13,092,666 and 2,513,768 estimated
households. The country is sparsely populated with a population density of 17.4 persons per
square kilometer. The most densely populated area is Lusaka province were there are 100
persons per square kilometer. The population is distributed disproportionately between rural and
urban areas with approximately 60% living in rural areas. The estimated number of people living
in the rural areas is 7,919,216 and in the urban areas it is estimated to be 5,173,450. The
population by gender is estimated to be 6,454, 647 for males and 6,638,019 for females (CSO
Census Report 2010).
Agriculture is the most common source of income and livelihood, which constitute 85%
of the labor force. The mining and industry sectors make up 6% of the labor force and services
industry approximately 10%. Sixty-five percent of the population lives below the poverty line,
earning less than $1.25USD a day, much of which occurs in the rural areas. The level of poverty
11
in the rural areas is three times higher than in urban areas, which are estimated at 77.9% and
27.5% respectively. (CSO Census Report 2010).
The number of physicians available per patient is .07 physicians for every 1,000 patients.
Physicians are medical doctors, specialists, and general practitioners. The World Health
Organization (WHO) estimates that fewer than 2.5 healthcare workers, which include physicians,
nurses, and midwives per 1,000, would be insufficient to adequately meet primary healthcare
needs. The estimated percentage of health expenditures is 6.5% of Zambian Gross Domestic
Product (GDP). In 2010, the Zambian health system had a total of 1,883 health facilities, six
tertiary level hospitals, 21 general hospitals, 85 district hospitals, 1,495 urban and rural health
centers, and 275 health posts (CIA World Factbook 2013; CSO Census Report 2010; WHO
2012).
Like most of the African continent where 62% of the African population is under the age
of 25, Zambia has a young population, with 45% of the population under the age of 15. The
median age in Zambia is 16.9 years with the average life expectancy of 52 years. The estimated
population under the age of five is 2,252,748 with a mortality rate of 138 deaths per 1,000 live
births. The crude birth rate is 35 per 1000 people present at mid-year. The infant mortality rate in
Zambia is 76 deaths per 1,000 live births and the child mortality rate is 62 deaths per 1,000 live
births. The major causes of child mortality are infectious diseases like pneumonia, diarrhea,
malaria, and measles and are common in Zambia. The mean age of a mother’s first birth is 19
years old. The population of women of childbearing age (15-49 years) is estimated to be
2,822,635. The national maternal mortality rate is 86.1 deaths per 100,000 live births (CSO
Census Report 2010).
12
2.3 Disease Burdens
Seventy-five percent of all reported deaths in Zambia are attributed to illnesses and
disease. Additionally, the major cause of disability in Zambia is also attributed to disease.
According to 2012 estimates, 12.7% of adults aged 15 to 49 are living with HIV/AIDS
(UNAIDS 2012). Zambia, like most African countries, has a generalized HIV/AIDS epidemic
and has made great progress in effective treatment and distribution of antiretroviral therapies.
This has resulted in a steady decline in AIDS related deaths in the last decade. This steady
decline has led to a shift in focus to address other high burden diseases and illnesses in Zambia.
Malaria, Pneumonia, and Diarrhea are among the leading causes of death in sub-Saharan Africa,
especially for children under the age of 5.
The overall disease burden in Zambia can be divided into two classes, Non-
communicable (NCD) and Communicable diseases (CD). Non-communicable diseases are
defined as diseases or conditions that are non-infectious and non-transmissible among people.
These may be chronic diseases that have a long duration and slow progression or conditions that
are short duration and rapid progression. NCDs include many types of cancers, heart disease, and
diabetes, etc. For CDs, these are any condition that is transmitted to a person either directly or
indirectly from an infected person or animal. This also includes transmission through an external
agency such as intermediate animal, host, vector, or inanimate environment (Annual Health
Bulletin 2011). Table 2.1 lists the leading NCD and CD burdens for all age groups in Zambia.
13
Table 2.1 Leading Causes of Death in Zambia for Non-communicable and C
Communicable Diseases
Non-communicable Diseases Communicable Diseases
Hypertension Malaria
Asthma Respiratory Infection (RI): (Non-Pneumonia)
Epilepsy Diarrhea: non-bloody
Herpes Zoster
Muscular skeletal and connective tissue non-
trauma
Cardio Vascular Disease Trauma other injuries wounds
Diabetes Digestive System Non-Infectious
Karposi Sarcomas Respiratory Infection (RI): Pneumonia
Cervical Cancers Eye Diseases Infections
Breast Cancers Skin Disease Non-Infectious
Dental Carries
Source: 2011 Annual Health Statistics Bulletin, HMIS Dataset
2.3.1 Malaria
Epidemiology
Globally it is estimated that 3.4 billion people are at risk of malaria. In 2013, there were
104 countries and territories in which malaria was considered endemic (World Malaria Report
2013). Pregnant women and children under the age of 5 are the population groups with an
increased risk of developing severe malaria outcomes and death. Estimates by the WHO state
that approximately 207 million cases of malaria and nearly 627,000 deaths occurred globally in
2012. The majority of these cases (80%) and deaths (90%) occurred on the continent of Africa,
with most deaths (77%) occurring among children under 5 years of age (World Malaria Report
2013).
Malaria is caused is caused by five species of parasite that affect humans. Of the five, the
species Plasmodium falciparum and Plasmodium vivax are the two most important and P.
falciparum is responsible for the deadliest form of malaria. P. falciparum is the predominant
species in Africa, however P. vivax has a wider distribution. P. vivax has a wider distribution as a
result of its ability to develop in the Anopheles mosquito at lower temperatures, survive at higher
14
altitudes, and exist in cooler climates (World Malaria Report 2013). P. vivax also has a dormant
liver stage, which allows it to survive during winter periods and when the Anopheles mosquitos
are not present to carry out transmission. In areas outside of Africa, infections with P.vivax are
more common than infections due to P. falciparum.
Table 2.2 Epidemiology of prevalent malaria species in Zambia
Life Cycle P. falciparum P. vivax
Minimum temp needed for
maturation in the mosquito
Lowest temp 60.8°F
Lowest temp for cycle to be
complete 59°F; Lowest temp for
parasite survival, 50°F for two
days
Dormant liver stage No Yes
Gametocytes Appear after asexual blood stage is
established
Appear at time of asexual blood
stage often before clinical
symptoms
Disease
Severity 5% of cases develop into severe
illness; responsible for majority of
deaths
Risk of severe disease not firmly
established
Relapse No
Yes
Asymptomatic
Outcomes
Common Very Common
Diagnosis
Blood film, RDTs and PCR for blood
stage
Blood film, RDTs and PCR for
blood stage
No test for dormant
Liver stage
Treatment
Blood stage Artemisinin combination treatment
(ACT) recommended
Chloroquine still efficacious in
most areas
Gametocytes Need single dose Primaquine,
Artemesinins have some effect
Sensitive to blood stage treatment
Liver stage 14 days of Primaquine
Source: WHO World Malaria Report, 2013
Malaria Control and Prevention
In endemic countries, where the risk of exposure to malaria is high, great efforts have
been made to develop effective malaria control policies and strategies. This has involved the
15
establishment of well-funded National Malaria Control Programs (NMCP), tailored national and
regional strategies, applied and operational research initiatives, and collaboration among
multinational partners within the malaria and development communities.
The World Health Organization is a multinational collaborator and primary facilitator of
many global malaria control standards and initiatives. The WHO has developed malaria control
guidelines and recommends a multi-faceted strategy for malaria control and elimination, which
includes prevention therapies, diagnostic testing, Artemisinin-based combination therapies,
strong malaria surveillance, and vector control (World Malaria Report 2013).
Vector control has been an important strategy of many malaria control programs. The
main objectives of vector control strategies are to lower the intensity of malaria transmission by
shortening the lifespans of local mosquito populations; and to reduce, or prevent altogether, the
interaction between mosquitos carrying malaria causing parasites and humans. Nearly all vector
control strategies involve the use of pesticides and chemicals designed to achieve one or more of
the main objectives for vector control strategies. Of the different strategies available for vector
control, the most successful are indoor residual spraying (IRS) and insecticide-treated nets
(ITNs), including long-lasting insecticide treated nets (LLINs).
Malaria prevalence in Zambia
Malaria is the leading cause of morbidity for all age groups in Zambia affecting more
than 4 million people annually. Zambia has made great efforts in improving the extent of malaria
control services offered throughout the country. As a result, the government has identified
malaria control interventions, both treatment and prevention, as a major public health priority
area for the National Health Strategic Plans (National Malaria Strategic Plan 2011–2015).
Between the years of 2009–2011, Eastern Province had the highest incidence of malaria.
During this same period, there was a reduction in malaria incidence from 821 per 1,000 persons
16
in 2010 to 772 per 1,000 persons in 2011 (2011 Annual Health Statistical Bulletin). According to
the 2012 Zambia National Malaria Indicator Survey, malaria parasite prevalence was 14.9% with
more parasite prevalence occurring in children (22%) living in rural areas. Children in rural areas
were more likely to suffer from fevers than those in urban areas (29.3% and 13.2%,
respectively). Over 4.7 million cases of malaria were treated at health facilities, with 6.1 million
rapid diagnostic test kits distributed to health facilities in 2012.
Malaria Diagnosis and Treatment
The introduction of rapid antigen-detection diagnostic tests (RDTs) into many national
malaria control programs, has made malaria case detection more reliable, and has improved the
timeliness of appropriate treatment interventions. Prior to the introduction of RDTs malaria
diagnosis was clinical (symptom-based) and was proven to be quite unspecific as to what
symptoms were reliable for diagnosis (Msellem at al. 2009). The presence of fever was the most
commonly used symptom for malaria clinical diagnosis and treatment was given for malaria.
Additionally, children with fever who were not confirmed to have malaria were also given
antimalarial medications, as precautionary measures for malaria illness. This practice referred to
as presumptive treatment has become less appropriate given the increased availability of low cost
RDTs and lower transmission rates of malaria.
Fever is often an overlapping clinical symptom for malaria, pneumonia, and diarrheal
diseases, which are all common illnesses in resource poor settings or developing regions. In
malaria endemic areas, fevers in children under the age of five are often attributed to malaria
illness (Kazembe et al. 2007, Amexo et al. 2004, d’Acremont et al. 2010). Children under the
age of five account for ninety-percent of all malaria deaths. In malaria endemic areas, the onset
of fever in children is often attributed to malaria, regardless of clinical confirmation, and
treatment is therefore given for malaria (Rao, Shellenberg, and Ghani 2013). This method of
17
symptomatic diagnosis has led to misdiagnosis and overtreatment of malaria in many endemic
areas (Amexo et al. 2004). Misdiagnosis and over treatment of drug therapies has been known to
reduce the effectiveness of drug treatment and is a major cause of drug resistance.
For children who are misdiagnosed as malaria, the true cause of fever remains untreated
resulting in adverse outcomes, potential drug resistance, prolonged and more severe illness, or
death (Amexo et al. 2004, Hume et al. 2008). The development of RDTs was an important
innovation for malaria case detection. It created opportunities in malaria endemic areas and
resource poor or remote settings to have access to parasitological confirmation of potential
malaria cases (Hamer et al. 2007).
The WHO has taken steps to acknowledge the mismanagement and under-diagnosis of
non-malarial febrile illnesses, with high mortality rates, among children. The World Health
Organization and the United Nations Children’s Fund (UNICEF) released a joint statement in
2006 encouraging the practice of integrated case management for febrile illnesses among
children less than 5 years of age. The encouragement by the World Health Organization and the
United Nations Children’s Fund reflected a shift from previous standards in patient management
that was focused solely on febrile malaria cases. The new standard for patient management was a
shift towards more concurrent approaches that focused on other common non-malaria febrile
illnesses like pneumonia and severe diarrhea among children less than 5 years of age (Chanda et
al. 2009; Young et al. 2012; World Malaria Report 2013).
According to a review conducted by the Child Health and Epidemiology Reference
group, it is estimated that a 70% reduction in mortality from pneumonia in children could result
due to community efforts to manage all cases of childhood pneumonia (Young et al. 2012). The
review also estimated the use of oral rehydration salts and zinc, at the community level in
treatment for diarrheal illnesses, could prevent 70-90% of deaths caused by acute watery diarrhea
18
(Young et al. 2012). These findings illustrate the impact of addressing non-malarial febrile
illnesses. The use of RDT kits has led to more accurate diagnosis of malaria, however for cases
of febrile illness confirmed not to be malaria, the cause of fever is difficult to establish (Chanda
et al. 2009; White et al. 2012).
The use of RDT also improved drug treatment management and reduced the early onset
for drug resistance. Early drug treatment involved use of monotherapies either with Chloroquine
or Sulfadoxine-pyrimethamine. As a result of common usage, misdiagnosis and overtreatment,
drug resistance quickly emerged for malaria treatment using Chloroquine and Sulfadoxine-
pyrimethamine. This emerging resistance prompted many African countries to change treatment
policies to Artemisinin-base combination therapy (ACT) as first line treatment for non-severe
malaria cases (Msellem et al. 2009). The combination of RDT and ACT was an important
strategic shift towards reducing morbidity and mortality associated with malaria in endemic
regions. In Zambia, the malaria control program focuses on the use of Coartum as first-line drug
therapy. Sulfadoxine-pyrimethamine is the primary drug treatment for non-severe malaria in
children, and Quinine is prescribed to the most severe malaria cases.
ACTs have become a mainstay in malarial drug treatment therapies and involve the
concurrent use of two drugs, one of which is Artemisinin based, to slow down the evolution of
parasite resistance. Reports of ACT resistance have been published and are cited in Hastings
(2011). These findings have caused some concern in the malaria research community and
spurred research into the causes of resistance. Hastings (2011) comments, that combining various
drugs with Artemisinin has delayed progression into more severe drug resistance that would have
occurred with monotherapies. Hastings (2011) suggests that emphasis now should be placed on
good data capture accompanied by efforts to improve diagnosis and drug compliance to maintain
the efficacy malaria treatments.
19
2.3.2 Pneumonia Infections
Pneumonia is the single largest cause of death in children under 5 years of age and kills
an estimated 1.1 million children annually, which is more than malaria and tuberculosis deaths
combined. Pneumonia is an acute respiratory infection that affects the lungs and can be caused
by viruses, bacteria, or fungi. It is a preventable disease through immunization, adequate
nutrition and control of and protection from environmental factors (WHO “Pneumonia 331”
2013).
Pneumonia can be spread through the air and through blood contact. For viral and
bacterial pneumonia, clinical symptoms are similar, with viral pneumonia having more explicit
symptoms than bacterial pneumonia. In children under the age of 5, who have a cough and/or
difficulty breathing, with or without fever, pneumonia is diagnosed by the presence of either fast
breathing or constricted chest movements during breathing inhalation. Children with weakened
immune systems or who are malnourished are at increased risk for developing pneumonia
infection (WHO “Pneumonia 331” 2013).
Most cases of pneumonia can be treated with antibiotics, with most requiring oral
antibiotics, which are often prescribed at a health center. In Zambia, acute respiratory infection is
among the top five causes of morbidity and hospital admissions for both adults and children
(Annual Health Bulletin 2011). Between the years of 2009 to 2011, the pneumonia incidence rate
steadily increased from 29.3 to 35.9, respectively for number of new cases per 1,000 persons.
Eastern province had the highest incidence rate during that same time period and Lusaka
province had the lowest incidence rate (Annual Health Bulletin 2011).
20
2.3.3 Diarrheal Diseases
Similar to pneumonia, diarrheal diseases are both preventable and treatable. Diarrheal
diseases are the second leading cause of death in children under five years of age, accounting for
nearly 760,000 deaths annually. Diarrhea is a leading cause of malnutrition among children
(WHO “Diarrhea 330” 2013). Annually, there are nearly 1.7 billion cases of diarrheal diseases
globally and diarrhea is a leading cause of malnutrition in children under five years. Many of
these cases can be attributed to poor sanitation and hygiene in addition to a lack in access to safe
drinking water (WHO “Diarrhea 330” 2013).
The clinical definition of diarrhea is the passage of three or more loose or liquid stools
per day or more than normal frequency of passage for an individual. There are three clinical
types of diarrhea, acute watery diarrhea, acute bloody diarrhea, and persistent diarrhea. Diarrhea
is usually a symptom of infection caused by virus, bacteria, or parasitic organisms in the
intestinal tract. Infection is spread from person to person due to poor hygiene and may also be
spread through contaminated food and drinking water (WHO “Diarrhea 330” 2013). Diarrhea
due to infection is common in many developing regions. The most severe symptom of diarrhea is
extreme dehydration where water and essential electrolytes (sodium, chloride, bicarbonate and
potassium) are lost and not replaced as a result of watery stools, vomit, sweat, urine, and fast
breathing. Death can follow severe dehydration if fluids and electrolytes are not replaced.
The prevention of diarrheal outcomes involves interventions that include accesses to safe
drinking water, improved sanitation and hygiene systems and practices, and improved hand
washing practices with soap. Treatment for diarrhea can be performed through consumption of
an oral rehydration solution (ORS) packet containing clean water, sugar and salt, and with zinc
tablets. Globally, 780 million individuals lack access to clean drinking water and 2.5-billion lack
21
access to improved sanitation conditions (WHO “Diarrhea 330” 2013).
In Zambia, diarrhea is among the top ten major causes of morbidity and dehydration.
During the time period of 2009-2011 diarrhea incidence were 72 cases and 86 cases per 1,000
persons respectively, reflective of a steady increase for diarrheal outcomes over time (Annual
Health Bulletin 2011). All provincial areas had a steady increases in diarrheal outcomes during
this time with Southern and Northwestern provinces recorded the highest incidence rates. Lusaka
and Northern provinces had the lowest incidence in 2011 with 74 cases and 68 cases per 1,000
persons, respectively.
2.4 Mobile Health (mHealth) Applications
Mobile health innovations have helped to improve the overall quality of care in countries
like Zambia and other Sub-Saharan African countries. Mobile health (mHealth) is a subset of
information communication technology (ICT) applications, which relies on the use of mobile
phones, personal digital assistants (PDAs), tablets, smart phones, and other mobile devices to
manage and share health information and data.
Kamanga et al. (2010) used mobile phones and GIS to detect potential outbreaks of
malaria to identify locations where parasite reservoirs are likely to occur for targeted
interventions. The study occurred in Southern Province of Zambia in Choma and Namwala
districts. This study area had reliable mobile network reception and nearly all residents and at 12
out of 14 rural health centers (RHC) had access to the network.
The CHW entered routine health data into the health center registry daily. Once a week,
data for positive malaria diagnosis and total number of RDT used per week by SMS to
researchers at the Malaria Institute in Macha from each health center. Each SMS transmission
contained information that included the RHC name, name of transmitting nurse, number of RDT
used during the week and number of positive diagnosis with RDT. The data were then entered
22
into a spreadsheet and sent to District Health hospitals, National Malaria Control Center
(NMCC) and John Hopkins University in Baltimore, Maryland. Each weekly spreadsheet
consisted of data from 12 of the 14 RHC, two RHC were omitted from the study due to poor
phone access. The data was stratified by RHC into two separate ecological zones, flood plain
area and an area of increasing aridity and elevation. Comparing SMS texts with actual registry
data at each RHC validated data and any discrepancies were investigated and ameliorated. The
participants were compensated for the use of their mobile phones and provided a modest stipend.
GPS coordinates were also collected for each RHC and plotted on a Landsat map that
included elevation data and other features. The weekly malaria incidence rate was calculated for
each RHC catchment area. These methods illustrated the plausibility of mobile phone usage for
routine surveillance activities and that the timeliness and efficiency in resource allocation for
malaria control operations could be improved in doing so. The use of GIS also served to identify
areas where patterns of RDT usage affected the overall cost effectiveness of diagnostic testing
for malaria.
The authors concluded that through this case study, there is evidence that use of mHealth
approaches for reporting critical data would provide opportunities rapid deployment of specific
interventions and drugs to prevent malaria transmission. The network of CHWs recruited into the
program had been receptive and highly cooperative, and regular feedback of data and
information was conducted. The use of mobile phone system was effective and inexpensive and
broader implementation would allow for more timely communication between RHC and central
health authorities. All expenses and reimbursements were handled through the SMS system.
In Tanzania, a similar study evaluating SMS and mobile phone usage was conducted
(DeRenzi et al. 2012). In this study, researchers evaluated the impact of SMS reminders to
improve the timeliness of routine CHW visits to homes of clients located in their communities.
23
In Tanzania, CHW success has been tied to routine home visits, which can be difficult to
maintain as a result of shortages in available equipment, supplies, and reliable transportation.
DeRenzi et al. (2012) introduced mobile phones to CHW as a tool for community health
data management. They initially began with a pilot study in Dodoma, Tanzania for 9 weeks, to
evaluate the impact of SMS reminders to improve CHW performance. This was then followed by
two larger studies implemented in Dar es Salaam, Tanzania over a 9-month period with 87
CHWs. The authors introduced a reminder system that augmented existing supervision structures
within CHW programs. This system sent proactive reminders to CHWs the day before and the
day of scheduled routine visits.
Based on results and feedback from the pilot study, the SMS reminder system was
modified in two ways. First, in the pilot study, the first SMS reminder was sent after working
hours, in the evening of the day that the CHW’s visit was due, which was too late for an on time
visit. This was addressed in the Dar es Salaam studies with the introduction of proactive
reminders, with the earliest SMS reminder sent the day before a visit is due. Secondly during the
pilot study, if a referral was not closed during the pilot, the system continued to send SMS
messages but had no capabilities of determining why the CHW was not reporting a follow-up
visit. This led to the addition of the SMS reminder escalation to the supervisor stage for
intervention, implemented in the two larger studies in Dar es Salaam.
The strengths of this study can be found in the use of a control and intervention group, a
pilot phase that informed a larger study, and comparison of two methods illustrating the impact
of structured supervision. The limitations of the study can be found in the accuracy of the
findings with actual follow-ups and ground trothing. Unlike, Kamanga et al. (2010), which
verified mHealth data with data available at each RHC, DeRenzi et al. (2012) did not verify if
CHWs actually visited clients. Logistically this was not feasible, however upon Supervisor
24
escalation, CHW would call or SMS an automated number and provide reasons why to
supervisors, which provides some evidence that visits actually occurred. The authors note that
future applications should make use of GPS and lower cost mobile phone technology to address
this limitation.
Braun et al. (2013) provide a systematic meta-analysis review of peer-reviewed literature
focused on evaluating the impact of mobile technology on community health workers’ utilization
in healthcare management. They reviewed 25 research articles published between January 1,
2000 and June 30, 2012 to identify opportunities and challenges for strengthening health systems
in resource-constrained settings. They analyzed and coded each article for various topics then
compiled findings into tables according to key topics. The 25 articles reviewed involved 28
unique studies and most reported on projects in developing countries. There were more programs
operating in rural areas than in urban areas, with a few operating in both. A broad range of health
issues was addressed with the most common topic areas being sexual, reproductive, and maternal
and child health or which more than half focused specifically on HIV/AIDS. Other key topic
areas were tuberculosis and malaria. In the articles reviewed, mHealth technology was most
commonly used for data collection, decision support, alerts and reminders, information on
demand, and as a facilitation tool for CHW activities associated with field based research and
direct provision of medical care.
The authors concluded that more analyses evaluating the cost effectiveness of mHealth
utilization by CHW would be helpful for program staff and policy makers. Future research
should also focus on utilization of qualitative data collection methods in order to better
understand how mHealth tools may be improved and adapted to local CHW contexts and
performance. They also note limitations of their review, citing the number of articles reviewed
unlikely reflects the scope of mHealth projects being implemented. They go on to state that most
25
projects were focused on deployment of mobile tools to enhance service delivery rather than
scientific interests, it is likely that many mHealth projects are not reported in published literature.
In another meta-analysis review conducted by Nhavoto and Gronlund (2014), the authors,
unlike Braun et al. (2013), sought to identify applications of mHealth initiatives that integrate
mobile technologies and GIS together in concurrent application. Through their analyses they
identified how mobile technologies and GIS applications have been used independently as well
as in combination in healthcare information infrastructure as a means to provide a basis for data
analysis and decision-making support.
The authors selected a total of 271 articles centered on use of mobile technologies and
GIS in improving healthcare, 220 focused on mobile technologies and 51 GIS. Most of the
articles reviewed involved studies implemented in developed countries, most notably the United
States. The articles were then sorted into six predominant themes: treatment and disease
management, data collection and disease surveillance, health support systems, health promotion
and disease prevention, communication to or between health care providers, and medical
education. Applications of GIS technology were sorted into four predominant themes: disease
surveillance, health support systems, health promotion and disease prevention, and
communication to or between health care providers.
Although most of the articles reported positive results, Nhavoto and Gronlund (2014)
identified research gaps. First, the overwhelming majority of publications reported positive
results, suggesting that unsuccessful applications were under reported. Secondly, evaluated
publications focused on effects and often failed to discuss implementation efforts, creating
barriers for scale-up into broader applications and adaptation for various settings. The authors
cite this as an important research gap and an indication of a more serious problem. Despite all of
the successive outcomes, many of the interventions were performed as pilot projects or small-
26
scale projects, with very few scaling up to larger scales. A possible explanation suggested by the
authors was the challenge of aligning many more actors and stakeholders, standardizing data,
upfront investment in digitization of data, and legal, economical, and practical constraints
regarding communication in larger scale implementation. Research into implementation
feasibility may need to be conducted in these areas before large-scale interventions occur.
Finally, the findings of this review also concluded that little integration between GIS and mobile
technologies occurs in many of the mHealth interventions reviewed. Nhavoto and Gronlund
(2014) conclude that in order for mHealth information processes to be most effective they must
integrate different kinds of existing technologies. The increasing development of mobile apps
provides additional opportunities for integration of mHealth health application and management
tools.
Although Nhavoto and Gronlund’s (2014) criticisms are valid, there are often numerous
barriers to GIS integration. The use of GIS in itself can be quite costly and integration with other
technologies may be even more expensive. In addition to this, the training costs both in time and
funding, are added barriers to GIS utilization in many developing regions and health systems.
Mobile phones with GPS or GIS components are also costly despite the increasing availability of
smartphones in developing countries. Many smartphone applications require reliable Internet
network access and this can prove difficult in remote areas. Numerous open sourced GIS
software and platforms have emerged, such as QGIS and R, reducing and even eliminating some
of these barriers.
In the public health sector, SaTScan v9.3 is an open-sourced software developed by
Martin Kulldorff, that analyzes datasets for spatial, temporal, and space-time clustering and has
been widely used to perform hot spot analysis on health data. It was designed for geographical
surveillance of diseases through spatial or space-time disease cluster detection and assesses
27
statistical significance of identified clusters. SaTScan v9.3 can be used on discrete or continuous
data for spatial scan statistics through a number of scanning options for various data types.
Epidemiologists in developed and developing regions have used SaTScan v9.3 to detect disease
clusters for cancers, malaria, and TB among a number of other diseases (Coleman et al. 2009;
Zhao et al. 2013, Sherman et al. 2014). The low cost and ease of use, allow for SaTScan v9.3 to
be an alternative option to complicated propriety software for GIS analyses, and provides an
opportunity for resource strapped health systems to conduct geospatial analyses on their health
data.
Nhavoto and Gronlund’s (2014) claimed that mHealth information processes would be
most effective if they integrate different kinds of existing technologies. For mHealth programs in
Zambia, given the availability of low cost mHealth and GPS devices, what would be the impact
of integrating existing mHealth devices with GPS capabilities and programming? Would the
integration of GPS and mHealth devices be more advantageous for CHW community
disease/illness case management and treatment? Given the positive outcomes of previous
mHealth programs and initiatives in Zambia on CHW case management and performance, it is
predicted that the integration of mHealth devices with GPS would prove to be more
advantageous. It would allow for data coupling with GIS for in-depth spatial data analyses at the
sub-district level. Community areas could be explored for non-random disease variability and
clustering. Identified hot spots or clusters would benefit from more targeted interventions. This
thesis will aim to substantiate this hypothesis by exploring the potential benefits of integrating
existing mobile technologies used by CHW with GPS capabilities for routine health data
management and activities. The proceeding methods and analyses describe the use of GIS on
CHW community health registries to assess illness distribution at the sub-district level in
Chongwe, Zambia.
28
To explore this potential impact, a geodataset will be created from paper-based CHW
registries to model the expected output of a GPS enabled mHealth device. Spatial analyses using
SaTScan v9.3 on geo-referenced CHW health registry will seek to identify clusters of febrile
illness related malaria, diarrhea, and pneumonia at the community level. The assumption is that
these registries can highlight areas where diagnostic commodities and scarce health care
resources would have the greatest impact on disease/illness surveillance, control, and treatment
efforts within the district.
This thesis will build upon a previous mHealth program for the proceeding secondary
data analyses and will explore health registries managed by CHWs in Chongwe during the Grand
Challenges Canada (GCC) mHealth feasibility study. SaTScan v9.3 will be used for spatial
statistics on a sub-district administrative area to evaluate and identify non-random illness
distribution in Chongwe.
2.5 GCC mHealth Study in Chongwe District, Zambia
2.5.1 Overview
The GCC mHealth study was a Zambian IRB approved community-based project
conducted between the months of January and October 2013 (IRB# 00006464;
IORG#00005376). It was a pilot study that evaluated the feasibility of mobile phone usage
among CHWs for community health data management in the rural district area of Chongwe,
Zambia.
The program was a feasibility study to assess the effect of ICT devices on community
health performance in Zambia (Chanda et al. 2009; Chanda-Kapata et al. 2011). Furthermore, the
study aimed to explore the effects of mobile health and ICT devices on CHW retention and on
the quality of health services provided at the community level. The researchers primary
objectives sought to answer the following questions:
29
1. What would be the impact of ICT devices on community health services and CHW
motivation and retention in rural Zambia?
2. What would be the impact of ICT use for community health services on the health
status of communities in Zambia?
3. What is role would ICT devices have on the capabilities for health center staff to
monitor CHWs?
A mixed methods approach of both qualitative and quantitative data collection methods
was used. Seventeen health facilities participated in the study and were randomly assigned into
a control or intervention group to determine the effects of ICTs on CHW retention rates. Eight
health facilities were randomly selected and placed in the control group and 9 in the
intervention group. The health facility staff in each facility included primary nurses,
environmental health technicians, and clinical officers who acted as supervisors to CHWs.
A preliminary survey was conducted to determine the perceptions of ICT device
utilization among community and professional health workers. Successful utilization by CHWs,
promised to improve information management, incentivize CHW retention, and improve
community education and services.
2.5.2 Study Area
The location of the study area took place in an area called Chongwe, a rural district are
located approximately 35 Km from the capital city of Lusaka in Lusaka Province. It is an area
with high malaria prevalence within the communities. Agriculture and farming are the main land
uses and source of income by residents. Chongwe’s climate is arid and it is nestled in valley
region surrounded by numerous hills. This district area was selected as the study location
because of previous work with CHWs on malaria case management (Chanda et al. 2011). It is
also a sentinel site for malaria surveillance for the region.
30
The district is divided into rural health facility catchment areas, which are further sub-
divided into CHW zones. The health facilities that fell within the catchment area zones of each
participating CHW were also included in the study. Each CHW reported to one health facility
within their catchment area, so multiple CHWs may report to one health facility. Health facilities
that did not have CHWs were excluded from the study.
2.5.3 Population
The population estimates for Chongwe district were 192, 303 (2010) and for each CHW
zone the population ranged from 200–1000 individuals. Each CHW zone had a number of
farming villages located within their respective catchment areas. The district had 40 health
facilities with a total of 377 health workers, of which 245 were health professionals, 116 were
daily employees, and 16 administrators. The administrative staff was based in the District Health
Office. All residents who lived within the catchment areas of participating CHWs were included
in the study. Resident participants were any individuals who sought health services from the
CHWs, both in the intervention and control areas, during the time period of January to October
2013. The primary health services provided were mainly for malaria, pneumonia, diarrhea, and
health education.
2.5.4 Community Health Volunteers
The district had a total of 88 CHWs and CHWs who resided within their respective zone
and were currently not involved in any other research program were included into the GCC
study. The CHWs who participated had been previously involved in earlier health research
studies in the area and therefore had an existing relationship with researchers (Chanda et al.
2011). The professional health staff for each health facility served as supervisors for each CHW.
In the study each community health worker training status was evaluated through a
questionnaire along with their currently existing responsibilities. After the baseline assessment,
31
training began for those selected in the intervention arm on how to use a mobile phone for
routine health data collection for program activities. The training involved an overview of the
project, ICT devices and their application to health, reporting format and frequency, and how to
generally care for and security of the devices (mobile phones and solar chargers). The training
for the CHW in the control group was a refresher course for Integrated Community Case
Management (ICCM).
The CHW were provided with conditional airtime to add to their mobile phones for the
purposes of transmitting weekly SMS summary reports to the central district reporting system.
The summaries were stored electronically on the District Health Information System 2 (DHIS2)
platform. Additionally, CHWs in both the intervention and control groups kept handwritten
registries for patient specific epidemiological information management. These registries served
as mobile data backups for each CHW involved in the intervention arm of the program.
2.5.5 Data Collection
The patient level variables collected included patient ID, sex, age, location name, medical
complaint, and treatment outcome. The participant information was completed for each visit. On
a weekly basis, CHWs composed aggregated summary reports, which were then sent by SMS to
the DHIS2 server at the nearest health facility. Health facility staff and central level program
staff reviewed this information using password-protected access on the DHIS2 server routinely.
The server had restricted access preventing health facilities from viewing health data provided by
other health facilities. Monthly monitoring was conducted by review and comparison of copies
of the CHW registries to DHIS2 databases.
For the control sites the patient information was collected and entered into the CHW
registries only and manually transferred to the nearest RHC for data entry into the DHIS2 server
monthly. Health facilities in the control group operated per the usual guidelines.
32
2.5.6 Outcomes
The study indicated that it was feasible to implement ICT interventions using mHealth
devices in rural areas. All of the CHWs and health facility staff in the intervention group were
able to use the reporting system effectively. Epidemiological data was available for all levels of
care in a timely manner at the intervention facilities. The ICT devices had positive effects on the
community health services and retention of CHWs. Perceptions among community members,
CHWs, and health facility staff were positive, creating an opportunity for broader
implementation of mHealth application for community health initiatives.
Overall, mobile phones as a tool for CHW case management was beneficial. However,
there were some issues in regards to the model of the phone and the capacity of the district to
respond to the reports received from the various sub-district reporting areas. Additionally this
initial study did not integrate the use of mobile phones with other technologies like GIS. This is
in line with similar studies noted in Nhavoto and Gronlund (2014). However, unlike most
mHealth programs, spatial thinking was involved in the development of the program. One of the
secondary objectives was to “develop real-time epidemiological maps of notifiable diseases.” In
order to achieve this secondary objective, the geospatial data analyst assigned to this study
performed the following methods.
33
CHAPTER THREE: METHODOLOGY
3.1 Datasets
3.1.1 GCC Dataset
In Zambia, spatially explicit data is not routinely collected for research purposes.
Geospatial thinking does not occur during program or study development, resulting in
retrospective GPS coordinate data collection activities or the omission of any spatial evaluation
altogether due. An added challenge was the lack advanced road network and address systems.
Street names, addresses, and zip codes, typically used for geocoding algorithms and programs in
more developed regions like the United States, do not extensively exist in Zambia, most
especially in rural areas. Without advanced road network and address systems, traditional
geocoding methods are often more difficult to perform, and are at present great barriers towards
geospatial applications in developing countries. Some well-funded non-governmental
organizations (NGO) have overcome this barrier by using GPS devices to obtain coordinate
locations. This approach to geolocation data collection can often be expensive and time-
consuming for everyone to do. For researchers, the optimal method for geospatial data collection
would be to integrate GPS data collection methods, into routine data acquisition activities, during
ongoing research programs.
To spatially examine the CHW registries with GIS, the existing GCC database needed to
be geocoded, but unfortunately investigators had not collected any GPS latitude or longitude
coordinate points for patient or facility locations. Due to limitations in research funding and
human resources, it was not feasible to use a GPS unit to retrospectively collect coordinate data
points for patients’ household locations in Chongwe. As the only member of the data analysis
team with GIS knowledge assigned to this study, it was proposed that use of the village names
recorded in the registries, may be useful for manual geocoding methods for village coordinate
34
locations without the use of a GPS device. The feasibility of this alternative approach to
coordinate data collection was unknown and had not been documented in any currently available
literature.
A de-identified CHW dataset was obtained for this project. The registries were entered
into an Excel spreadsheet, the variable names for latitude and longitude were added to the sheet
next to the village and health facility variable in the spreadsheet. The name of the village was
used as a patient’s location and residential address. There were two methods that were attempted
to determine the feasibility of manually geocoding villages by name. The first involved the use
of Google maps and Google Earth software to enter the village names directly into the search bar
to determine the geo-coordinates for each village location. This was unsuccessful due to the
duplicate naming of certain villages and cities in Zambia. Some village areas had the same name
as other geographic locations in other provinces, resulting in the wrong coordinate locations
showing up in searches. It was found that this would not be a feasible method for coordinate
identification as a result duplicate location nomenclature.
The second method required collaboration with other researchers and organizations that
had done work in Chongwe or near these village areas. Chongwe is a malaria endemic area and
is a sentinel site for a number of NGO and governmental malaria research programs and
surveillance initiatives. PATH, an American based NGO focused on malaria research and health
systems development programs, had conducted numerous studies in the area in previous years.
Many of PATH’s research activities involved the use of mobile GPS devices and GPS enabled
PDAs to collect coordinate locations of local landmark areas and villages for their malaria
disease mapping and modeling initiatives. In 2010, researchers from PATH had conducted a
study that required the collection of various geolocations within Chongwe district at the
community level. The data was later stored in an Access database and included the village names
35
and corresponding latitudes and longitudes for numerous village locations in Chongwe (Figures
3.1 & 3.2).
A request was made to officials at PATH to have access to this dataset. PATH officials
were gracious to share their village location geodataset and it was used as a reference log to
identify the village names in Chongwe for the GCC dataset. For each data entry in the PATH
log, a GPS coordinate was recorded and a note was included as to where the coordinate points
were collected and an indication of if the location was a village or landmark.
After sorting for all of the records in the GCC dataset that had a village name recorded,
each record with a village name in the GCC dataset was copied. Using the search function of the
Access dataset, village names were manually searched for in search for a matching location in
the PATH log. If there was a match, then the geographic coordinates recorded in the PATH log
were copied and pasted in the columns for Latitude and Longitude respectively, for the
corresponding village in the GCC dataset. This was done for each village location recorded in the
GCC dataset. Additionally, a second dataset was obtained from the Central Statistical Office
(CSO) that also contained a number of village coordinate locations in Chongwe. The locations
were obtained during 2010 census activities, however the dataset was not entirely complete with
many of the GPS coordinates missing the naming reference for each coordinate location.
In total of the 7,673 patient records, 2,529 did not have a village name recorded for the
patient, and of the 5,144 that had a village name recorded, 3,130 were geocoded using both the
CSO and PATH dataset log. A total of 141matching village locations that were identified with
corresponding geo-coordinates in the reference datasets.
36
Figure 3.1 PATH log in an Access database containing latitude and longitude coordinate
locations for Chongwe villages
37
Figure 3.2 PATH log in Access database containing the geocoded village names used
matching village names
In addition to geocoding village names and locations, health facility locations (n=17)
were also geocoded. This task was made easier by the availability of datasets with health center
names and coordinates managed by the CSO and the Zambian Ministry of Health (MOH). After
a formal request was made to CSO, the health facility dataset was obtained, along with
administrative area shapefiles. Each health center was identified along with the corresponding
38
coordinates, which were then entered into the corresponding coordinate variable name for each
health facility. Of the 17 RHC that participated in the GCC study, only 15 health facilities were
geocoded, due to no village location records for the two missing health facilities (Table 3.1).
Table 3.1 GCC Health facilities included in GCC geospatial dataset
GCC mHealth Feasibility Study Dataset:
Health Facility Included
GCC Geospatial Evaluation
Dataset: Health Facility Included
(Yes/No)
Katoba RHC Yes
Mpanshya Hospital/HAHC Yes
Rufunsa RHC Yes
Chinyunyu RHC Yes
Lwiimba RHC Yes
Kasisi RHC Yes
Nyangwena RHC No*
Chainda RHC No*
Chalimbana RHC Yes
Kanakantapa RHC Yes
Waterfalls RHC Yes
Kankumba RHC Yes
Lukwipa RHC Yes
Kampekete RHC Yes
Mpango RHC Yes
Mwalumina RHC Yes
Shikabeta RHC Yes
*Location not included in geo-dataset due to missing village names for patients in CHW registries
After the coordinate locations were identified and entered into the Excel spreadsheet for
both the health facilities and village names, verification on the accuracy of these coordinates was
performed using the administrative area shapefiles obtained for Chongwe district from CSO, in a
GIS (Figure 3.3). The health facility and village points were input into ArcGIS 10.2 as ‘X,Y’
coordinate points and were then projected onto the district area shapefiles. The points were
validated through visual inspection and all were accurately placed within the Chongwe district
area boundary. For every village coordinate location identified (n=141), each location
represented multiple patient cases within the village.
39
Figure 3.3 GCC spatial dataset Excel spreadsheet with completed health facility and village
coordinates
3.1.2 CSO Administrative Area Shapefiles
The next steps required that the points be assigned to lower level administrative areas.
The data collected by CHWs for each health facility and village location were collected at the
lowest area administrative unit called Standard Enumeration Areas (SEA); however, to
determine which SEA code corresponded to each village and health facility location, it was
necessary to work first from higher level administrative areas down to the SEA levels. Shapefiles
40
were requested from CSO for four hierarchal administrative area boundaries for Chongwe, the
district, constituency, ward, and standard enumeration boundaries. These were then uploaded
into ArcGIS. In looking at the attribute tables, the FIDs were identified for each area and
incorporated into the GCC spatial dataset for future joins and relations of tables. The SEAs were
available only as FID codes (ex. 501073061031) requiring that to accurately identify each village
area within the right administrative boundaries and to the accurate codes, the GeoIDs for the
higher levels would have to be obtained prior to SEA assignment (Figure 3.4). Once the GeoIDs
were identified for each administrative area level and entered into the GCC Excel dataset for
each village location; the file was then uploaded into ArcGIS. Using the GCC dataset, each
shapefile was spatially joined to the attribute tables for the district and sub district administrative
areas based on the matching FID and GeoIDs.
Figure 3.4 Zambia administrative boundary areas in descending hierarchal order
National Area
Provincial Areas
District Areas
Constituency Areas
Ward Areas
Standard
Enumeration
Areas
41
The spatial join created an attribute table that included all of the information from the
GCC dataset for each boundary area and the spatial attributes from each administrative area
boundary shapefile. The GCC dataset was now ready to preliminarily assess the spatial
distribution of patient health outcomes (Figure 5.2).
3.1.3 CSO Population Datasets
Population data was also added to the dataset for each SEA administrative area prior to
ArcGIS input. The Central Statistical Office provided shapefiles that were used to identify and
manually enter population data for the village areas. The shapefile for the SEAs included the
2010 population estimates and the number of households for each SEA area. This information
was spatially joined with the GCC attribute table. The SEA boundary population areas served as
a proxy for CHW catchment areas. CHW catchment areas are artificially derived, unofficial
geographic boundary areas that not available for mapping.
In addition to the SEA populations, the ward level populations were included as well.
This population and administrative area was best for analyses regarding the RHC health
outcomes. Each health facility operated in a catchment area that was at the ward level and would
provide insights into the total number of cases per the total population in the area. This data was
unfortunately unavailable electronically but was available as a hard copy booklet at CSO. A
request for a copy of the booklet was granted and the population tables for Chongwe district and
wards were used to identify the population distributions, both household and total population, for
each ward area. This information was manually entered for the respective wards for each patient
village location in the GCC dataset prior to upload into ArcGIS. There were 17 Ward areas
within Chongwe district and both Ward level and SEA level population information were
recorded manually into an Excel GCC dataset, which served as a backup to the ArcGIS dataset.
42
3.1.4 District Health Facility Dataset
In addition to the overall population distributions for the sub-district areas, the actual
RHC catchment populations were also included in the GCC dataset. A request was made to the
District Health Office for RHC catchment area populations. This information was then added to
the GCC dataset for each RHC location, prior to the upload into ArcGIS.
3.2 Variables
The complete GCC spatial dataset included coordinate locations for villages and RHC,
administrative area information, and health outcomes. The following table (Table 3.2) is an in-
depth summary of each variable included in the GCC dataset and uploaded in ArcGIS after
village location matching and census data were included. If the data methods had been
performed by CHWs using a GPS enabled mHealth device for community health data
management, this is an example of the type of data output that would be expected for use with a
GIS.
43
Table 3.2 GCC Geodataset with census data and village locations identified using
PATH database.
Variable Name Description Source
Provincial
Code
This is a first order subnational administrative
boundary area. The country of Zambia is divided into
10 provincial locations. The Central Statistics Office
has assigned geoIDs to each provincial area available
shapefiles. The geoID for Lusaka Province is 05 and is
the ID used to spatially join the GCC dataset with
provincial level CSO shapefiles.
CSO, 2010
Census
Provincial
Name
This refers to the corresponding Provincial name for
the geoID allocated to each provincial area. The geoID
05 is for the Lusaka Provincial administrative boundary
area.
CSO, 2010
Census
District Code
This is a second order subnational administrative
boundary area. At the time of the 2010 census, the
country of Zambia was divided into 74 district level
locations. The Central Statistics Office has assigned
geoID to each district area in available shapefiles. The
geoID for Chongwe district is 501 and is the ID used to
spatially join the GCC dataset with district level CSO
shapefiles.
CSO, 2010
Census
District Name
This refers to the corresponding District name for the
geoID allocated to each district area. The geoID 501 is
for the Chongwe district administrative boundary area.
CSO, 2010
Census
44
Constituency
Code
This is a third order subnational administrative
boundary area. At the time of the 2010 census, the
country of Zambia was divided into 150 constituency
area level locations. The Central Statistics Office has
assigned geoIDs to each constituency area in available
shapefiles. The geoIDs for the constituencies included
in the GCC study, conducted in Chongwe district, are
501073 & 501074 and are the IDs used to spatially join
the GCC dataset with constituency level CSO
shapefiles.
CSO, 2010
Census
Constituency
Name
This refers to the corresponding Constituency names
for the geoIDs allocated to each constituency area in
Chongwe district. The geoIDs 501073 & 501073 are
for Chongwe and Rufunsa constituency administrative
boundary areas respectively.
CSO, 2010
Census
Ward Code
This is a fourth order subnational administrative
boundary area. At the time of the 2010 census, the
country of Zambia was divided into 1,430 ward area
level locations. The Central Statistics Office has
assigned geoIDs to each ward area in available
shapefiles. The geoIDs for the ward areas included in
the GCC study conducted in Chongwe district, are
50107301, 50107302, 50107303, 50107304, 50107305,
50107306, 50107307, 50107308, 50107309, 50107310,
50107412, 50107413, 50107414, 50107415, 50107416,
& 50107417 and are the IDs used to spatially join the
GCC dataset with ward area level CSO shapefiles.
CSO, 2010
Census
Ward Name
This refers to the corresponding Ward names for the
geoIDs allocated to each constituency area within
Chongwe district. The geoIDs 50107301, 50107302,
50107303, 50107304, 50107305, 50107306, 50107307,
50107308, 50107309, 50107310, 50107412, 50107413,
50107414, 50107415, 50107416, & 50107417 are for
Kapwayambale, Chinkuli, Ntandabale, Chongwe,
Kanakantapa, Chalimbana, Nakatindi, Lukoshi,
Manyika, Lwimba, Nyangwena, Bunda Bunda,
Nyamanongo, Rufunsa, Mankanda & Shikabeta ward
administrative boundary areas respectively.
CSO, 2010
Census
45
Ward
Household
Population
This refers to household composition at the ward area
level, which classifies all households according to the
relationships among the people in them, and whether
there is a family nucleus present or not. These people
may or may not be related by blood, marriage, or
adoption but make common provision for food or other
essentials for living and have one person whom is
regarded as head. A household can also have one
member.
CSO, 2010
Census
Ward
Population
This refers to total population for Ward areas included
in the GCC study. Population estimates were De Jure
meaning that the usual household members present and
usual household members temporarily absent at the time
of census were counted as members of their usual
households. The de jure counts is considered the most
accurate population count for the country.
CSO, 2010
Census
SEA Code
These refer to geographic statistical unit numbers
created in the Census of Population and Housing (CPH),
which contain a certain number of households. SEAs are
well-defined boundaries with area codes that are
recorded on census maps. There were 334 SEA code
areas for Chongwe district.
CSO, 2010
Census
SEA Population
An SEA is usually a group of small villages or a
village, or a part of a large village in the rural areas. The
SEA population is the total number of individuals living
in an SEA area.
CSO, 2010
Census
SEA Household
Population
An SEA is usually a group of small villages or a village,
or a part of a large village in the rural areas. Household
population is the number of households located within
villages or number of families within the villages of an
SEA area. .
CSO, 2010
Census
46
Date
The date the CHW saw the patient and health
information was recorded into registry.
GCC CHW
Registries,
2013
Rural Health
Center Name
The name of the Rural Health Center that each CHW
reported to in the GCC study within Chongwe district.
GCC CHW
Registries,
2013
RHC Latitude
GPS latitude coordinates for Rural Health Center
locations.
MOH 2012,
CSO 2012
RHC Longitude
GPS longitude coordinates for Rural Health Center
locations.
MOH 2012,
CSO 2012
RHC Catchment
Area Population
This refers the total community population that is
served by a RHC facility.
Chongwe
District
Health Office
Registries
2013
Patient Address
This is the village name of the patient that was recorded
during the patient visit by the CHW into the registry.
GCC CHW
Registries,
2013
Patient Address
Latitude
This is the latitude coordinate for each village area
identified using CSO census shapefiles and PATH
program registries.
CSO Census
2010, PATH
Patient Address
Longitude
This is the latitude coordinate for each village area
identified using CSO census shapefiles and PATH
program registries.
CSO Census
2010, PATH
Male Patients identified as 'Male' gender in CHW registries
GCC CHW
Registries,
2013
Female
Patients identified as 'Female' gender in CHW
registries
GCC CHW
Registries,
2013
Age Patients Age
GCC CHW
Registries,
2013
47
Fever
Patients were diagnosed with a fever by CHW during
visit
GCC CHW
Registries,
2013
History of Fever Patients who had a fever in the past month
GCC CHW
Registries,
2013
Headache
Patients who complained of a headache during CHW
visit
GCC CHW
Registries,
2013
Cough Patients who complained of a cough during CHW visit
GCC CHW
Registries,
2013
Diarrhea
Patients who complained of diarrhea or watery stool
during CHW visit
GCC CHW
Registries,
2013
Vomiting
Patients who complained of vomiting during CHW
visit
GCC CHW
Registries,
2013
Problems
Breathing
Patients who complained of problems breathing during
CHW visit
GCC CHW
Registries,
2013
Chest pain
Patients who complained of chest pain during CHW
visit
GCC CHW
Registries,
2013
No Symptoms
Patients who did not have or complain of any
symptoms during CHW visit
GCC CHW
Registries,
2013
Other Symptoms
Patients who complained of other symptoms during the
CHW visit
GCC CHW
Registries,
2013
ITNS Patients who had Insecticide Treated Bed-nets
GCC CHW
Registries,
2013
Slept Under
ITNS
Patients who slept under an ITN the previous night
GCC CHW
Registries,
2013
IRS
Patients who had Indoor Residual Sprays done in the
home in the last 6 months
GCC CHW
Registries,
2013
IPT
Intermittent Preventive Therapy for malaria provided to
pregnant women using Sulfadoxine and Pyrimethamine
drug therapies.
GCC CHW
Registries,
2013
48
Positive Rapid
Diagnostic Test
Patients who were given a RDT resulting in a positive
confirmation for malaria
GCC CHW
Registries,
2013
Negative Rapid
Diagnostic Test
Patients who were given a RDT resulting in a negative
confirmation for malaria
GCC CHW
Registries,
2013
Rapid Diagnostic
Test Not Done
Patients who were not given a RDT to confirm malaria
GCC CHW
Registries,
2013
Coartem
This is the first line ACT drug treatment for malaria in
Zambia
GCC CHW
Registries,
2013
Fansidar
This is the first line IPT drug treatment for malaria in
Zambia, it is provided as prophylaxis against malaria
for pregnant women.
GCC CHW
Registries,
2013
Amoxyl The antibiotic Amoxicillin prescribed for infections.
GCC CHW
Registries,
2013
Zinc
Zinc supplements provided for dehydration related to
diarrhea. Often combined with ORS for rehydration
therapy.
GCC CHW
Registries,
2013
ORS
Oral Rehydration Salts (ORS) used to treat severe
dehydration related to diarrhea in combination with
Zinc supplements.
GCC CHW
Registries,
2013
Panedol
Common pain killer that is prescribed directly by CHW
to patients in the community
GCC CHW
Registries,
2013
Other Drugs Other drugs that are prescribed for community illnesses
GCC CHW
Registries,
2013
Referred
Patients who were referred to nearest RHC for drug
prescriptions and further follow-up.
GCC CHW
Registries,
2013
49
3.3 Analysis
3.3.1 ArcGIS Data Preparation
Once all variables from each CHW registry were incorporated into the GCC geodataset,
the dataset was imported into SPSS 15.0 statistical analysis software to perform descriptive
statistical analysis. Cross-tabulations and frequencies were performed for each disease and
illness outcome of interests for each village location area. The total number of fever, fever and
malaria, malaria, fever with cough and problems breathing, fever and diarrhea, fever and
vomiting, pneumonia, and severe diarrhea cases were tabulated for each village location. This
information was added to a secondary geodataset in an Excel spreadsheet for each village
location. Afterwards, this geodataset was imported into ArcGIS 10.2 and spatially joined to the
existing GCC shapefile. Prior to adding any data into ArcGIS, the data frame was set to a
projected coordinate of Arc 1950 UTM Zone 36S. All shapefiles, including village and RHC
coordinate points were set to the same projection in preparation for analysis and mapping.
Finally, to determine the corresponding SEA for each village coordinate location, a
shapefile of SEA areas was used and village coordinates were overlaid onto the SEA shapefile. A
manual search was done using the ‘identify tool’ in ArcMap for each village location point
placement to identify the corresponding SEA in which each village point fell. For each village
location that fell within an SEA area, the SEA number was identified and recorded into a new
field in a third dataset for each village location in the attribute table. After all SEA numbers were
found for each of the patients and village locations, the dataset was spatially joined to the GCC
shapefile. To disease distribution at the Ward level, a spatial join was also done using the Ward
codes and Ward area level shapefiles. Information on corresponding village locations was
available in the CSO report and was also recorded during the same time population estimates
were determined for the Wards.
50
3.3.2 Population Density Distributions
To determine the population densities for both the Ward and SEA areas, the area of the
various geographic boundaries were calculated first. The geographic areas were determined by
creating a new field for each SEA and Ward level geography areas in ArcMap. In ArcMap the
new field, ‘Area’, was added to the attribute tables and areas were calculated in square
kilometers for each location and administrative level using the ‘Calculate Geometry’ function.
After areas were determined, the population densities were calculated into an additional field
labeled, ‘Pop_desit’, in the attribute tables using the ‘Field Calculator’ function to divide the
total populations by the total square kilometer areas. The result was used to map the population
densities for both of the respective SEA and Ward area levels. No population counts were
missing and a few areas had a total population estimate of zero. Chloropleth maps showing the
population densities using quintiles were created to evaluate the population distributions and case
count distributions.
3.3.3 Illness Case Count Distributions
Illnesses were cross-tabulated with patient addresses to determine the locations and case
counts using SPSS 15.0. More complex illnesses were cross-tabulated concordantly with two or
more symptoms to determine these illness outcomes. For example, the illnesses ‘Severe
diarrhea’ and ‘Pneumonia’ were defined as cases where both vomiting and diarrhea were
present and where cough and problems breathing were present, respectively. Malaria cases were
defined as occurrences where RDT results were positive and where RDT results were positive
and fever was also present. Other febrile illness cases were tabulated and recorded as the total
number of cases for each location into geodataset. Cases were determined for both the SEA and
Ward area levels. Maps of proportional case counts for each Ward level were produce for
exploratory data analysis and descriptive epidemiology of the general distribution of illnesses.
51
3.3.4 Incidence Rates
Once the total numbers of case counts were determined for each location, the incidence
rates were calculated in Excel using the formula cases/total population * 10^2 annually. 10^2
was used as the per person number at risk value due to the small areas and low population
distributions associated with rural districts. The incidence rate for each location was calculated
for all symptoms and illnesses of interest and recorded into the geodataset. Illness incidence
rates were calculated for SEA administrative area level only to provide better spatial insights into
symptom and illness distributions, at the community level. Chloropleth maps showing the
incidence rates per 100 persons using natural breaks for descriptive spatial epidemiology. The
incidence maps created and opportunity for understanding of the spatial distribution of diseases
and for comparison with future spatial analysis results.
3.3.5 SaTScan v9.3 Cluster Analysis
SaTScan v9.3 Data Preparation
SaTScan v9.3 requires multiple input datasets to analyze spatial data for clustering or hot
spots. Dataset inputs for SaTScan v9.3 analysis were prepared using Excel, ArcGIS 10.2, and
SPSS 15.0. SaTScan v9.3 datasets consisted of a case file, population file, coordinate file, and an
adjustment file for each disease/illness evaluated. All village level patient cases were aggregated
using SPSS 15.0 to the SEA area level for each illness case file as total counts.
The total population for each SEA area was used for the population file. Missing
population data or zero population areas were identified and entered into a separate adjustment
file used for spatial adjustments made by the software during analysis. Geographic coordinate
locations for each SEA were required for the coordinate file input for SaTScan v9.3 analysis.
Coordinate locations for each SEA area were determined using ArcGIS to identify the centroid
location of each polygon feature in the SEA shapefile. The resulting output provided the latitude
52
and longitude coordinates for each SEA location. This file was extracted into Excel and a
‘vlookup’ formula using the SEA codes as the matching variable was used to match the
geodataset with all the SEAs coordinate locations. This would allow for identification of SEA
areas identified to be cluster areas in the SaTScan v9.3 analysis using ArcGIS.
Missing data for locations and cases were included in the adjustment file for areas with no data.
The adjustment file also included risk estimates for areas with missing population information
for every disease and illness. A relative risk of 0 was determined for these locations in the
adjustment file, used during analysis.
SaTScan v9.3 Methods
SaTScan v9.3 was used to run a standard purely spatial scan statistic for each illness and
disease outcome of interest. Spatial cluster analyses were performed to test whether
disease/illness cases were distributed randomly over space and, if not, to identify areas of spatial
disease clusters for statistical significance. Spatial scan statistics for cluster detection was applied
to test the null hypothesis that the relative risk (RR) for disease/illness cases were the same
between SEAs or groups of SEAs.
The SaTScan v9.3 methods imposed a circular window on the map, which moved over
the areas and centered on the centroid of each SEA, allowing the maximum window for cluster
size to be set to any value less than 50% of the total population. The default setting was 50% of
the total population at risk and this was maintained for all spatial analyses conducted (Kulldorff
2014). This allowed SaTScan v9.3 to evaluate very small and very large clusters, and everything
in between. For each window of varying population and size, the software tested the risk of
illness inside and outside the window. Using the null hypothesis of spatial randomness, the
expected number of cases in each window was proportional to the combined population of SEAs
whose centroid is inside the circle. This allowed for SaTScan v9.3 adjusted for uneven
53
population distributions. The datasets were scanned for only clusters with high rates of
disease/illness, equivalent to a one-sided statistical test (Kulldorff 2014).
Clusters were identified by a comparison of the expected and observed number of cases
within and outside a scanning window that has a varying radius and center. This comparison is
called the Likelihood Ratio (LR), and it determined how likely a cluster exists due to chance.
The higher the LR, the more likely the cluster exists or the maximum likelihood, due to more
than chance alone. The scanning window with the maximum likelihood ratio is flagged as the
most likely or primary cluster and LR is reported. The primary cluster and significant secondary
clusters (ordered in descending LR values) are also included in the output. P-values were
assigned to each cluster to show statistical significance of cluster findings, with small p-values
indicating clusters of significance. The p-values were obtained through Monte-Carlo simulations
that randomly generated 999 replications of the dataset under the null hypothesis. Monte-Carlo
simulations create random simulations of the data, then report how many of those simulations
resulted in higher likelihood ratios than what the actual data found (Kulldorff 2014).
Clusters that were identified for each disease or illness of interest were mapped in
ArcGIS to visualize the location of significant primary and secondary clusters. Areas identified
to be clusters area were shaded in pink for secondary clusters and a darker shade of red for
primary clusters for each disease/illness outcome evaluated. Clusters with p-values less than .05
were considered significantly valid clusters. The SaTScan v9.3 scanning window buffer areas for
each cluster group was also included in the maps at the 50% risk population level.
A relative risk and its p-value were also given for each cluster output, indicating the risk
of developing a disease/illness within cluster or geographic areas relative to areas outside of
those locations. In epidemiology, a relative risk of 1 would indicate equal risk for disease
incidence given exposure. A relative risk above or below 1 would indicate an elevated or
54
decreased risk, respectively, for disease incidence given exposure. Relative risks that are below 1
are said to have a protective effect, meaning that exposure may prevent or delay disease or
illness incidence. In SaTScan v9.3, a relative risk above 1 in a cluster indicated how much more
common a disease/illness occurred in those locations compared to the baseline (Kulldorff 2014).
The corresponding p-values for relative risks determine the probability that these could have
happened by chance alone.
55
CHAPTER FOUR: RESULTS
4.1 Spatial Epidemiologic Maps
4.1.1 Population Density Maps (Ward and SEA)
The population density of the Ward area level was mapped to visualize the population
distribution throughout the district (Figure 4.1). Darker areas of blue indicate higher population
densities for the Ward areas. The distribution of village and Rural Health Center locations can
also be observed. The Ward area level maps provide insights at the sub-district level for Rural
Health Centers. Rural Health Centers in the Western part of the district are located in wards with
the highest population densities. Rural Health Centers in the Eastern part of the district were less
in number as a result of lower population densities. The village locations were spread throughout
the district and were greater in the Western part of the district and along the major roadway, the
T4, moving from West to East away from Lusaka city.
56
Figure 4.1 Chongwe district population densities for Ward administrative areas with
village and RHC locations
The population density for Chongwe SEAs was also mapped to evaluate the population
distributions at the community sub-district level.
The population distribution at the SEA level was similar to the Ward areas, however the
population was most dense in the center of the Western section of the district area (Figure 4.2.).
Village locations are represented in pink dots were overlaid onto a population density map. The
most densely populated SEA areas also had the smallest land area in the district. The population
densities and numbers of RHC locations also coincided at this administrative area level, with
areas with higher densities having more RHC facilities than lower densely populated areas.
57
Figure 4.2 Chongwe population densities for SEA administrative boundary areas and
RHC locations
4.1.2 Illness Case Count Distributions
Case counts were determined for each disease/illness at the Ward and SEA area level.
The counts were shown using proportional symbols to illustrate the volume of cases for each
ward area in comparison to other ward areas. These proportional symbol maps helped to visually
evaluate different disease/illness distributions for various locations. Larger circles indicate more
cases or counts and smaller ones proportionately fewer counts. For some maps, multiple
variables were overlaid to identify areas were symptoms occurred together or to evaluate number
cases versus interventions. Three proportional symbol variables were overlaid together on the
58
same map (Figure 4.3); malaria, fever, and RDT usage to identify areas where fever counts were
high and malaria cases were low, and where fever cases were high and RDT counts use was low
to see possible areas of concern for effective case management. Areas were identified where
RDT use was proportionately less than in areas with high fever and malaria counts, indicating a
need for more RDT implementation in these areas.
The proportions of cases in the Wards indicated areas where possible clusters of illness
may exists and were potential areas for further interventions. There were areas where malaria
cases were low, but fevers were high and RDT usage was also high to confirm truly confirm
malaria (Figure 4.4). These areas would be priority areas to explore further for other causes of
fevers and to develop interventions to address these unknown causes of fever. Additionally, there
were areas (Eastern part of district) where malaria cases were higher in proportion to RDT usage,
and these areas are of equal concern because they may be potential areas for targeted RDT
resource allocation.
4.1.3 Incidence Rates
The SEA administrative level was used for community level calculations and all
geostatistical analyses of malaria, pneumonia, and severe diarrhea illness outcomes and to
map the incidence rates for all diseases/illnesses. This administrative level was the closest
proxy to CHW catchment areas and community level assessments. Incidence rates were
calculated for each illness recorded in the CHW registries and were rendered in ArcGIS in
chloropleth maps (Figure 4.4).
59
Figure 4.3 Proportional Symbol Map of case counts for malaria, fevers, and RDT usage within Chongwe Ward areas.
60
Figure 4.4 Chongwe disease/illness incident rate estimates per 100 persons annually for
SEA administrative boundary areas: 1.) Incidence of Diarrhea & Fever; 2.)Incidence of
Problems Breathing, Chest Pain, Cough and Fever; 3.) Incidence of Fevers with Fever
History; 4.) Malaria Incidence; 5.) Incidence of Fever; 6.) Incidence of Fever, Cough, &
Problem Breathing; 7.) Incidence of Severe Diarrhea (Diarrhea & Vomiting) not referred;
8.) Incidence of Malaria with Fever; 9.) Incidence of Cough & Problems Breathing
(Pneumonia)
An evaluation of the incidence maps provided additional insights into the spatial
distribution of disease/illness cases. There were variations in incidence rates for malaria, severe
diarrhea, and pneumonia throughout the district. For a number of diseases/illness, the highest
61
incidence rates occurred in remote and less densely populated areas (Figure 4.4. Map #:
1,4,5,7,8). For malaria incidence, areas known to have the greatest risk for malaria outcomes
coincided with map areas that had the highest incidence rates.
4.2 Cluster Analysis Results
Sub-district areas identified on the maps to have the highest incidence rates were
locations that required further investigation. A spatial scan cluster analysis was performed to
provide more in-depth insights into these high incidence areas. The analysis focused on fever,
malaria, diarrhea, vomiting, severe diarrhea, and pneumonia disease/illness burdens, which were
identified by CHWs to be the most common in the district. Areas identified to be statistically
significant clusters were represented with pink shading. Primary cluster areas where represented
with a darker shade of pink in all maps. Secondary clusters continued to be represented with a
lighter shade of pink for all maps.
If areas of high disease/illness incidence were identified to be statistically significant hot
spots, then it would provide evidence for stakeholders and CHW to make better-informed
decisions inform where allocation of scarce resources would be most effective for treatment and
case management by CHWs.
4.2.1 Fever
The cluster analysis of fever incidence identified seven primary and secondary cluster
areas (Figure 4.5). The scan identified seven cluster areas. The primary most likely cluster area
consisted of 15 rural SEA areas near Rufunsa RHC and Mpanshya Mission Hospital RHC,
indicated in darker pink in the figure. This group of SEAs was determined to be the most likely
area to have a cluster of fever incidence with a Monte Carlo rank of 1/1000 (p < 0.05). The RR
for the primary cluster area was 8.97 (p < 0.05), indicating that the risk of fever within this area
group was greater than locations outside this area group. There was one SEA area, shown in the
62
inset map, located in a high population density area that had an excess risk of fever incidence
(RR= 41.2; p>.05). This area was determined to be a secondary cluster area.
Figure 4.5 Fever Incidence Cluster Areas
4.2.2 Malaria
The cluster analysis of malaria incidence identified five primary and secondary cluster
areas. The scan identified a primary cluster consisting of 2 rural SEA areas near Lukwipa RHC
(Figure 4.6). This highlighted group in the darker shade of red was the group of SEAs was
determined to be the most likely area to have a cluster of malaria incidence with a Monte Carlo
rank of 1/1000 (p < 0.05). The RR for the primary cluster area was 57.4 (p < 0.05), indicating
that the risk of malaria within this area group was more than locations outside this area group.
There were two SEA areas, located in a high population density areas that had an excess risk of
63
fever incidence (RR= 58.2; p>.05; RR= 44.0; p>.05), however there were determined to be a
secondary cluster areas and were not identified to be a most likely cluster area. Although the RR
were high in these areas, it was not unusual given the malaria endemicity and spatial variability
in malaria outcomes.
Figure 4.6 Malaria Incidence Cluster Areas
A secondary analysis of recorded malaria cases was done, but with the inclusion of the
presence of a fever symptom (Figure 4.7). This analysis of the incidence of malaria outcomes
with fever identified six primary and secondary cluster areas. The scan identified a primary
cluster consisting of 23 rural SEA areas near Lukwipa RHC, Shikabeta RHC, Rufunsa RHC, and
Mpanshya Mission Hospital RHC in the Eastern part of the district and Kasisi RHC and
Waterfalls RHC in the Western part of the district. In addition to this cluster group, four SEA
64
areas in the more densely populated Western part of the district were also included as primary
cluster areas. This group and location of SEAs was determined to be the most likely areas to
have a cluster of malaria with fever incidence with a Monte Carlo rank of 1/1000 (p < 0.05). The
RR for the primary cluster areas was 20.3 (p < 0.05), indicating that the risk of malaria with
fever within these group and locations was more than locations outside this area group and
locations.
Figure 4.7 Malaria with Fever Incidence Cluster Areas
65
4.2.3 Diarrheal Illnesses
The case definitions for diarrheal illnesses were vomiting, diarrhea, and for severe
diarrhea outcomes, the presence of both diarrhea and vomiting symptoms together. The cluster
analyses evaluated each of these diarrheal illness outcomes to identify sub-district level hot spot
areas in Chongwe district (Figures 4.8 and 4.9).
4.2.3.1 Diarrhea
The cluster analysis of diarrhea illness incidence identified three primary and secondary
cluster areas. The scan identified a primary cluster consisting of 39 SEA areas within a densely
populated area in the Western part of the district, near Katoba RHC, Lwiimba RHC, Mwalumina
RHC, Chalimbana RHC, and Kampekete RHC. This group of SEAs was determined to be the
most likely area to have a cluster of diarrhea incidence with a Monte Carlo rank of 1/1000 (p <
0.05). The RR for the primary cluster area was 4.76 (p < 0.05), indicating that the risk of diarrhea
incidence within this area group was more than locations outside this area group.
66
Figure 4.8 Diarrhea Incidence Cluster Areas
4.2.3.2 Vomiting
The cluster analysis of only vomiting illness incidence identified five primary and
secondary cluster areas (Figure 4.9). The scan identified a primary cluster in a single SEA area
within a densely populated area in the Western part of the district near Kasisi RHC and
Waterfalls RHC. This SEA area, which is shown in the inset map, was determined to be the most
likely area to have a cluster vomit incidence with a Monte Carlo rank of 1/1000 (p < 0.05). The
RR for this primary cluster area was 51.71 (p < 0.05), indicating that the risk of vomiting illness
incidence within this SEA area was more than any other location outside of this area.
67
Figure 4.9 Vomiting Incidence Cluster Areas
4.2.3.3 Severe Diarrhea
The cluster analysis of severe diarrhea illness incidence identified four primary and
secondary cluster areas (Figure 4.10). The scan identified a primary cluster consisting of 11 SEA
areas within a densely populated area in the Western part of the district, near Chalimbana RHC
and Kampekete RHC. This group of SEAs was determined to be the most likely area to have a
cluster severe diarrhea incidence with a Monte Carlo rank of 1/1000 (p < 0.05). The RR for the
primary cluster area was 10.61 (p < 0.05), indicating that the risk of severe diarrhea incidence
within this area group was more than locations outside this area group.
68
Figure 4.10 Severe Diarrhea Incidence Cluster Areas
4.2.4 Pneumonia
The case definition for pneumonia illnesses were cough and problems breathing and for
severe pneumonia, the definition was the presence of fever, cough and problems breathing
symptoms occurring together. The cluster analyses evaluated both pneumonia and severe
pneumonia illness outcomes to identify sub-district level hot spot areas in Chongwe district.
69
4.2.4.1 Pneumonia
The cluster analysis of pneumonia incidence identified three primary and secondary
cluster areas (Figure 4.11). The scan identified a primary cluster consisting of 10 SEA areas in
the Western part of the district, near Katoba RHC. This group of SEAs was determined to be the
most likely area to have a cluster of diarrhea incidence with a Monte Carlo rank of 1/1000 (p <
0.05). The RR for the primary cluster area was 23.75 (p < 0.05), indicating that the risk of
pneumonia incidence within this area group was more than locations outside this area group.
Figure 4.11 Pneumonia Incidence Cluster Areas
70
4.2.4.2 Severe Pneumonia
The cluster analysis of severe pneumonia illness incidence identified five primary and
secondary cluster areas (Figure 4.12). The scan identified a primary cluster consisting of 10
SEA areas, which were the same areas that were identified in the cluster analysis for pneumonia.
Two additional secondary clusters were also identified in this analysis. The Monte Carlo rank
and RR were also the same as for pneumonia, indicating that the risk of severe pneumonia
incidence is the same for pneumonia outcomes regardless of severity.
Figure 4.12 SaTScan v9.3 Severe Pneumonia Incidence Cluster Areas
71
4.3 Limitations
A number of limitations may affect the reliability of these results. Any final conclusions
made should be made with caution and reservation. One of the primary limitations that should be
noted is the quality of the dataset used to evaluate cluster areas. Geocoding methods were
applied only to areas where village locations had been recorded by CHWs. If a CHW entered
village names for a quarter of all patients in the area, this would lead to inaccurate assumptions
about the incidence rate and RR calculations for that area. Secondly, validation of the matched
village locations and ground truthing was not performed, with the possibility that villages may
have been inaccurately matched to with the village coordinates or location area. It was assumed
that village names that matched in both datasets were inherently the same village areas, given the
same village nomenclature, indicated in the reference datasets. It is possible for villages to have
the same name but be located in two entirely different locations. This would have an influence
on the disease incidence in the SEAs. The locations that were identified serve as proxies for the
true village location and should be considered to be the general location of a village and not the
true location for all the villages in the dataset. As a result of using secondary data collection
methods for the geolocations and not having complete census data for these locations, it was not
possible to verify the accuracy for each village location coordinates with official government
census. This is a primary limitation of secondary data analysis and should be noted for this study.
Third, the analyses were limited to purely spatial analysis for disease/illness outcomes
without the consideration of time or age adjustments. A spatial-temporal analysis would provide
insights into cluster variation of disease occurrence over time. A provision was made for these
adjustments with the optional adjustment dataset file. The adjustment file only included
adjustments for zero population areas, and even so, did not account for low population densities
72
and heterogeneity in risk was not accounted for. The Although CHW registries had a column for
recording patients’ age and dates of visits, too many inconsistencies and numerous missing data,
were sufficient reasons to exclude these variables from the analyses. Future studies should
address the importance of good record keeping and data completeness during CHW trainings.
Additional information on other covariates, such as gender and rainfall, would also have
an influence on cluster outcomes and risk estimates. The outputs generated with very high RR
should be further explored and tested using adjustments for known confounding and risk factors
to adjust the relative risk rates to more accurate estimates for reliable inference and decision-
making.
Another limitation was the issue of measurement error and misclassification bias. The
CHWs data were not validated for accuracy or cross-referenced with follow-up visits, which
could lead to measurement errors for cases of disease and illness. Also patients who had multiple
visits recorded multiple times could lead to over reporting of cases. It was difficult to know how
many times a patient was recorded with the same symptoms over multiple visits. However,
because many of the illnesses that were observed are endemic and chronic, each occurrence of
cases was treated independently regardless of the number of visits from the patient.
Lastly, it should also be noted that low and zero cases did not have an effect on the
statistical stability of the risk estimates. SaTScan v9.3 analyses do not depend on the
geographical resolution of the entered data, but rather on the population size of the circles
constructed by SaTScan v9.3. It is a primary reason SaTScan v9.3 is used for health surveillance
data, because it avoids arbitrary geographical aggregation of the data, allowing the scan statistic
to consider varying sized aggregations through a moving window (Kulldorff 2014). High rates
could then be attributed to confounding factors or covariates that would require further
exploration.
73
CHAPTER FIVE: DISCUSSION
It was hypothesized that CHW health registries were an under utilized source of
community health information, and that if health information were managed using mHealth
devices quipped with GPS capabilities, then community health data could be analyzed for local
area spatial variations using a GIS. It is thought that this would allow for more efficient health
systems strengthening and would be advantageous to policy-makers, stakeholders, and
communities. The remainder of this thesis will respond to the objective questions. Secondly an
in-depth discussion on the findings of the study, which suggests the hypotheses to be valid, will
be provided. The discussion on the primary objectives will be presented in the order that they
were addressed in the research.
Can existing CHW health registries be used to create disaggregated sub-district level, low-
resolution, geo-datasets suitable for geospatial health data analyses?
Yes, existing CHW health registries can be used to create sub-district level geo-datasets
for Chongwe, Zambia. A number of methods for how to manually create a geodataset were
explored to determine the best approach. A method that was explored required entering the
names of the villages into Google Earth to identify the geolocations to determine the village
locations manually. However, this approach yielded very little to no results. This was due to the
occurrence of multiple spellings of village names in the CHW registry records, sometimes
resulting in none of the spellings being truly accurate. So many village locations could not be
located using Google Earth. Secondly, names of villages recorded may not have been the official
names of the villages, often times the name of the headman of chief of a village was used as the
village name and not the name listed for the official census, again resulting in the inability to find
village location on Google Earth. Lastly, names of villages that were identified by Google Earth
were locations for urban areas that had identical nomenclature. A village named “Kafue” was
74
identified as the city or region of Kafue in Google Earth and not the village itself. Another
method that was explored for a manually generated geodataset relied on obtaining village
coordinate locations from the Census Statistics Office or the Office of the Surveyor General to
gain access to geodatasets containing coordinate information of village locations. This was done
and it required an official written request from the Ministry of Health to the various directors of
each agency. A request was made and after a month access was granted, however what was
accessible was limited to what was readily available and processed by the agencies. Additionally,
access to those sub-district level files was not available for public access and use. Unfortunately,
there were costs associated with data acquisition as well. These costs were very expensive and in
the end the decision was made to attempt other means. Fortunately, a few district level
administrative area shapefiles obtained from CSO had village names and locations in the
attribute tables. These attribute tables were used as a codebook for matching the village locations
to the CHW registries.
Lastly, a third approach that was explored to identify village geo-locations was the use of
a Gazetteer that could be downloaded freely from the Internet. This method, similar to a Google
search was unsuccessful due to the same limitations in the variations of village names and the
same name being used in multiple places. Also noted, were the poor spatial temporality of the
Gazetteer, which was very outdated and did not include any new village areas created within the
last 10 years.
Despite these challenges, an alternative approach was performed involving the use of
secondary data obtained from existing NGO geodatabases and geodatasets. These data sources
were used as reference geodatasets to identify the coordinate locations for each village, by
matching village names to the reference geodatasets.
75
Village names recorded in the health registries were used in the geocoding methods for
identifying patient village locations. The use of secondary geodatasets for use as georeferencing
codebooks was the ideal method used to determine coordinate locations for each village. Once
the locations were finalized, the dataset was entered into ArcGIS to confirm that the locations
made sense in relation to where each CHW worked and the corresponding patient records.
The quality of this dataset depended on the completeness of the CHW registries. Ideally,
CHWs who consistently recorded all the patient information would yield a higher quality
geodataset. However in this case, there were a number of CHWs who had inconsistencies in their
data, either in spelling the patient’s address, or in recording the patient village name at all. In
total, of the 7,673 patient records written in the registries, 2,529 did not have a village name
recorded for the patients, and of the 5,144 that had a village name recorded, 3,130 were
successfully geocoded by matching village names with secondary geodatabases as codebooks.
The final dataset was merged with administrative boundary area shapefiles and was
suitable for further geospatial exploration and analyses for health outcomes at the sub-district
levels at both the Ward and SEA levels.
Can the use of GIS on low-resolution health geo-datasets provide adequate spatial-temporal
insights into CHW needs and community health outcomes at a sub-district level?
Yes, the final geodataset created was a complete digitization of paper-based CHW health
registries managed at the sub-district community level. It served as a model for the type of output
that could be created if mHealth devices were enabled with GPS capabilities for routine CHW
activities.
The health registry geodataset had census data was added and population densities, case
counts, and incidence rates were calculated. A visual exploration of the data was then performed.
Through this visual exploration, the geographic distribution of villages and various incidence
76
cases were observed. This visual evaluation process was informative and provided insight on the
village locations and community health burdens and the conditions in which CHWs worked.
Looking at the preliminary maps it was clear that most health facilities were located in densely
populated areas. It could be inferred that CHWs in these areas may have a larger volume of
incident cases and may potentially experience frequent shortages in supplies and resources. For
the CHWs who worked in remote areas, it was evident that the nearest RHC could be a great
distance away, which could be a potential barrier for health access and service delivery. Many of
the villages were located in densely populated areas near Lusaka city in the Western part of the
district and along the main roadway, travelling Eastward, away from the Lusaka city. There was
an area in Figure 5.1 located in the far Eastern part of the district, that had a high population
density, indicated by the darker blue shade, located far from a nearby RHC. This would be an
optimal area for policy-makers to explore for to establish new health facilities in the district.
77
Figure 5.1 Population densities for Chongwe SEA areas and RHC locations.
Preliminary exploration of the disease/illness distribution maps also provided insights
prior to analysis (Figure 5.2), on potential hot spot or cluster areas. A number of areas seemed to
have a clustering in the number of illness cases. Some of the high numbers of cases identified on
the map were in locations where a RHC was not nearby. Areas identified to have high case
counts were noted as areas for additional exploration during the analyses. Additionally it was
noted that there were areas with high fever counts but RDT usage was low, indicating potential
areas of concern for misdiagnosis and mismanagement of malaria illness cases.
78
Figure 5.2 Exploratory map of disease/illness counts for each village location in relation to
RHC
After visual exploration of the proportional density maps was complete, the next step was
to determine if areas with high case counts were significant illness hot spots within the district.
SaTScan v9.3 was used to conduct a cluster analysis. ArcGIS 10.2 had been suitable for visual
explorations but was less ideal for performing a cluster analysis. The reason being, the geocoding
methods used and the type of records available in the CHW health registries. Ideally if the cases
had been individual case counts with unique geolocations, and not an aggregation into one
village location point, ArcGIS could be used to perform a cluster analysis using Morans I for
global district level cluster detection or Getis Ord Gi* for local sub-district cluster detection.
However, the dataset consisted of binary data (yes and no) records for symptom case
79
management. This data type was not suitable for spatial analysis using the available ArcGIS
tools and methods for hot spot and cluster analysis. There were a number of assumptions needed
to run ArcGIS hot spot and cluster analyses that the GCC dataset failed to meet.
An alternative option for cluster detection analysis was the open source software
SaTScan v9.3. The software uses population size and a moving scanning circular window of
varying size to determine hot spots or clusters areas. A location that has a low number of cases is
not influenced by the size of the underlying geographical area. An added benefit of the software
is the adjustment file, which accounted for low or missing populations along with any covariates,
and known risk ratios. The minimum number of spatial locations was also less restrictive than
ArcGIS, the minimum number of locations needed to run a purely spatial scan was two locations,
compared to 30 features for ArcGIS. Although for two locations a chi-square statistic could be
used instead, but ideally for SaTScan v9.3, the more geographical locations the better.
To run the spatial analyses, SaTScan v9.3 required at minimum a case file for each
disease/illness, a population file consisting of total population numbers for each geographic
location, and a coordinate file with the latitude and longitude for each geographic location.
Missing data could be adjusted for, including areas with zero population estimates, using an
adjustment file.
For areas that were not included in the study area, resulting in no case counts, SaTScan
v9.3 interpreted these areas as zero case counts, and these areas were included in the cluster
analyses. This inclusion would most likely influence the RR calculated within the circular
windows. Because the number of cases and the RR for those locations was unknown, any
conclusions made from the analyses should be made with caution. High risks could be a result of
low or missing case data or confounding effects not adjusted for. Additionally, the
inconsistencies in recording village names by CHWs, also increases the level of caution to any
80
inferential conclusions about the risks. This is another limitation of this study, which can be
accounted for if more CHW registries from more locations were included in the geocoding
process. Inclusion of more CHW registries from more geographic locations would allow for a
clearer assessment of the disease/illness distribution and risk estimates at the sub-district level.
Despite these limitations, areas were identified as clusters. The incidence and cluster
analysis maps were shown to a panel of district health experts from the district, including the
Provincial Medical Officer and the District Medical Officers. The maps were internally validated
for reliability to confirm results matched what would be expected for identified cluster areas. The
district medical officer and the principal investigators provided key informant information on the
plausibility of cluster locations occurring in certain areas. Feedback provided by key informants
revealed that pneumonia and fever clusters occurring in high incidence areas, were located in
areas where CHW training for fever integrated case management had been inadequate,
potentially leading to misdiagnosis or mismanagement of fevers in those areas. RDT usage was
lowest in rural areas where malaria with fever was highest and RHC were few in the Eastern part
of the district. It was noted that the Eastern part of the district was also an area that had been
provided the fewest interventions to combat malaria and febrile illness. In the cluster areas for
severe diarrhea, primary cluster areas occurred around Kampekete health center, a newly formed
RHC in the district. This may have been an area in need of a health facility and the results
support the need for one in the area. The development of a new health clinic may reduce severe
diarrhea incidence in the area over time and this map could serve as a base line for future
assessments. For malaria with fever incidence and clustering, there were a few locations
identified as primary cluster sites in densely populated rural areas in the Western part of the
district. Health experts cited the locations of two nearby rivers in the area, providing an
opportune vector area for mosquitoes and increased risk for malaria. Further exploration in
81
environmental controls in the area could lead to decreases in incidence and elimination of the
cluster areas. Overall, the identified cluster areas detected by SaTScan v9.3 coincided with
expectations provided by health experts and key informants interviewed who were familiar with
the area.
Are CHW community health registries an under-used source of information for insights into
sub-district level standards of care and health outcomes?
Based on the findings of this thesis, CHW health registries are an underused data source
capable of providing beneficial health information at the sub-district and community levels.
Coincidently, many mHealth programs do not include routine CHW case management registry
indicators into mHealth device programming. Instead most mHealth programs are designed to
address or explore a specific case management area like malaria, HIV, and maternal child health
indicators, etc alone. The beneficial insights that can be gained by evaluating illnesses at the sub-
district level, using mHealth methods, would be advantageous towards understanding the
epidemiology of diseases and illness within specific areas. Additionally, mHealth programs that
are coupled with GIS would strengthen surveillance and monitoring of diseases/illnesses, CHW
performance and impacts, and the efficacy of implemented health programs within communities.
Current data management processes attenuate the underlying epidemiology of many
disease/illnesses occurring within a district in Zambia. Data collected by CHWs is aggregated to
the nearest RHC, then from the RHCs the data is aggregated further to the district health centers,
once at the district health centers the data is sent to the provincial health centers who then report
data for the district to the Ministry of Health located at the central level. This process is time
consuming and potentially masks any hot spots or cluster areas occurring in a sub-district area.
For Chongwe, 334 SEA areas where located within the district and 102 were analyzed for in the
82
analyses. Performing an analysis on approximately one-third of the SEA areas identified cluster
areas and provided other useful and beneficial information, suggesting that data at the sub-
district level should be incorporated into health systems analyses and surveillance for more
timely and accurate assessments.
Community health worker health registries are a primary source for community data and
would be ideal for sub-district community level surveillance. The emergence of mHealth and
existing GIS technology create an ideal opportunity to begin using non-aggregated CHW
generated health data for evidence-based decision making addressing monitoring and treatment
of disease outcomes. Doing so would increase efficacy and timeliness of outbreak response and
other interventions. Community health worker registries are indeed an overlooked data source
and an under utilized asset.
Significance for Other Low-resource Settings and Regions
For other low-resource settings similar to Zambia, this project illustrates not only the
importance of geospatial thinking in disease surveillance and disease management, but also that
barriers associated with GIS technology costs can be mitigated. The methods used for the
retrospective analysis of the CHW patient registries can be applied to any registry or database
that had a geographic place or name that can be identified on a map. Other areas in which these
methods can be applied are Road Traffic Accident studies. Accident locations recorded in paper-
based police registries (ex. Kafue Rd 2 miles from Chongwe village) could be mapped using
similar methods and techniques presented in this paper. By working in collaboration with local
agencies that routinely collect geospatial data, manual geocoding methods can be achieved, as
was done with the Central Statistic Office or PATH. Additionally it provides an opportunity for
more remote settings to be mapped and databases generated. The geocoding process may take
time, but the benefits outweigh the cost in time or in not doing so at all.
83
Community-based participatory mapping is another way that coordinate locations could
be gathered at a low cost. This low-tech approach involves creating large paper-based maps with
latitude and longitude degree grid markings of a community area. Community members are then
asked to place stickers of varying shapes, sizes, and colors on top of the map at locations that
they know, this can include households, schools, landmarks, etc. This coordinate data can then
be entered into a GIS or Google Map for visualization and further analysis. In this project,
barriers for geocoding were overcome through partnerships with NGOs and other government
agencies, similar approaches can be done in other settings.
Secondly, the use of an open-sourced software also illustrates way in which GIS cost
barriers can be overcome. There have been a number of open-sourced GIS programs that are
available for data analysis and mapping. One that is most frequently used is QGIS, which is
completely free and is capable of performing the same data analysis of commonly used
proprietary software. There are a number of free tutorials and online resources to learn how to
use the software.
Lastly, the main take-away from this research is that geospatial thinking and
implementation can be achieved without the need of expensive technology or software. In many
developing regions, paper-based methods are still the primary form for data management. Even
with these practices and norms, geospatial methods can be applied as long as a geography
component exists in the data. The availability of open-sourced software also opens up more
opportunities to overcome financial barriers. Geospatial thinking enhances the depth of
knowledge in which decisions are made and if time or resources are limited, this is a good way to
prioritize decisions on policies and actions.
84
CHAPTER SIX: CONCLUSIONS
In conclusion, the findings of this study support the concept of applying GIS to CHW registries.
The model dataset yielded beneficial health information and provided geospatial insight into the
distribution of disease/illness at the community level. It also provided support for mHealth
initiatives to incorporate both routine CHW indicators into mHealth device programming for use
with a GIS. By coupling mHealth initiatives with GIS, it provides opportunities to explore
disease incidence over space and time, further encompassing the epidemiological tenets of
person, place, and time. Thus resulting in more efficient and accurate epidemiological
assessments within the district.
6. 1 Policy Implications
The use of mHealth and GIS on CHW health registries creates an opportunity for more
informed decision-making, through evidence-based support for community interventions and
policy development. Analysis of sub-district level disease/illness distributions would have
immediate impact on health policies at the community level. Using GIS to identify cluster areas
would provide evidence for more targeted interventions and strategic roll out of implemented
programs. Additionally, stakeholders and community developers could make better-informed
decisions on where new RHC would be most beneficial using incidence and density maps.
Routine surveillance and maps of sub-district areas would also inform stakeholders and
policy makers on programmatic impact and efficacy. In the case of severe diarrhea, the cases
were clustered around a new health facility. With routine community level surveillance and
mapping, changes in the cluster area could be monitored after additional CHW trainings and
health interventions are employed. Changes in size and presence of the cluster altogether would
inform if the interventions were efficacious. In areas where malaria and fever clustered, policy
makers could make informed decisions on where limited RDT resources should be allocated.
85
Also, monitoring and impact evaluations could occur after RDT resources are allocated and
malaria interventions are implemented in these areas.
The immediate impact these findings have on policies in Zambia is that they provide
some evidence to support the claim that the current system for reporting health data at district
and provincial level is insufficient. By aggregating community level data, many local variations
in disease outcomes, especially in rural areas, are inadequately reported reducing the surveillance
potential for rapid response to any disease/illness outbreaks. The primary policy
recommendation and next steps, based on these findings, would be to integrate GIS into existing
health systems for more accurate sub-district level data acquisition and real-time surveillance.
The recent introduction of the District Health Information System version 2 (DHIS2)
provides an ideal opportunity to implement this recommendation immediately at a national level.
The DHIS2 is a health information system used to manage data collected at the district levels
from all health facilities in a district. This second generation system has mapping and geospatial
data management capabilities integrated into the framework. However current implementation of
the system has yet to install or activate this feature in many of the districts, based on key
informant interviews. The system is able to visualize geospatial data down to the health facility
level for all health facilities in a district for real-time disease surveillance. The ability to acquire
and visualize data at a sub-district level opens up opportunities to merge mHealth initiatives
using CHW community level data directly into the DHIS2 system. This would optimize the
performance of the system and allow for user-generated data to be monitored immediately for
surveillance purposes, especially in rural areas and areas hard to travel to in the districts.
Additionally, data could be downloaded for further geospatial analysis by facility level, or
community level if merged with mHealth data, providing evidence for more informed decision-
making and policy development relevant to individual communities. Until the DHIS2 mapping
86
capabilities are fully activated to include sub-district data from individual health facilities, the
system will continue to be under utilized and disease surveillance inadequate for many
communities in Zambia. The findings of this research illustrate the advantages of looking at
diseases and illness through a geospatial lens and the benefits of geospatial analysis for decision-
making and policy development.
6.2 Future Research
Future research should evaluate cluster areas for variations in age, gender, and over time.
By stratifying analyses by age or gender, additional sub-district variations could be identified for
vulnerable populations such as children under-five and pregnant women. Due to inconsistencies
and missing data, a cluster analyses by age and gender could not be done with the existing CHW
registry dataset. Space-time cluster analyses could not be performed as a result of incomplete and
inconsistent data entry. This is often the case with paper-based data collection methods and
secondary data analyses.
Future mHealth research should also explore coupling with GPS and integrating CHW
registry indicators into mHealth device programming. By incorporating the use of mHealth
technology for CHW community health data management, internal quality assurance features
could be implemented to prevent incomplete or inconsistent data collection by CHWs. This
would also provide insights into individual CHW impact and would identify areas for additional
CHW trainings. It would also result in higher quality datasets for more types of analyses, such as
a geographic weighted regression, to identify spatial relationships related to elevation, rainfall,
temperature, and landuse, etc.
A cost benefit analysis should also be explored to identify the benefits and costs of broad
implementation of mHealth programs. In addition to this, explorations into various types of
mHealth devices should also be explored. As smart phone technology becomes more available
87
globally and cloud based data management becomes more prevalent, insights into performance
and outcomes related to the types of devices used would be interesting to know and would help
guide the types of devices that are best for mHealth utilization in resource poor settings.
Finally, validation of the findings of this initial assessment should be done using primary
data collection methods of village locations and additional CHW registry datasets throughout the
district.
88
REFERENCES
Abrams, A.M., M. Kleinman, and M. Kulldorff. 2014. Gumbel based p-value approximations for
spatial scan statistics. International Journal of Health Geographics 9: 61. http://www.ij-
healthgeographics.com/content/9/1/61 (last accessed 29 August 2014).
Acestor, N., R. Cooksey, P. N. Newton, D. Ménard, P. J. Guerin, J. Nakagawa, E. Christophel, I.
J. González, and D. Bell. 2012. Mapping the aetiology of non-malarial febrile illness in
Southeast Asia through a systematic review--terra incognita impairing treatment policies.
PLoS ONE 7 (9): e44269.
Amexo, M., R. Tolhurst, G. Barnish, and I. Bates. 2004. Malaria misdiagnosis: effects on the
poor and vulnerable. Lancet 364 (9448): 1896–8.
Bell, B.S., R.E. Hoskins, L.W. Pickle, and D. Wartenberg. 2006. Current practices in spatial
analysis of cancer data: mapping health statistics to inform policymakers and the public.
International Journal of Health Geographics 5: 49. http://www.ij-
healthgeographics.com/content/5/1/49 (last accessed 29 August 2014).
Black, R.E., S. Cousens, H.L. Johnson, J.E. Lawn, I. Rudan, D.G. Bassani, P. Jha, H. Campbell,
C.F. Walker, R. Cibulskis, T. Eisele, L. Liu, and C. Mathers. 2010. Global, regional, and
national causes of child mortality in 2008: a systematic analysis. Lancet 375 (9730):
1969–87.
Blaya, J., H. Fraser, and B. Holt. 2010. E-health technologies show promise in developing
countries. Health Affairs 2 (2): 244–251.
Braun, R., C. Catalani, J. Wimbush, and D. Israelski. 2013. Community health workers and
mobile technology: a systematic review of the literature. PLoS ONE 8 (6): e65772.
Ceccato, P., S.J. Connor, I. Jeanne, and M.C. Thomson. 2005. Application of Geographical
Information Systems and Remote Sensing technologies for assessing and monitoring
malaria risk. Parassitologia 47 (1): 81–96.
Central Statistical Office. 2012a. 2010 Census of Population and Housing National Descriptive
Tables.
———. 2012b. 2010 Census of Population and Housing National Descriptive Tables Volume
11.
———. 2012c. 2010 Census of Population Summary Report.
Chanda, E., and V. Mukonka. 2012. Using a geographical-information-system-based decision
support to enhance malaria vector control in Zambia. Journal of Tropical Medicine 2012:
363520
89
Chanda, P., B. Hamainza, H.B. Moonga, V. Chalwe, and F. Pagnoni. 2011. Community case
management of malaria using ACT and RDT in two districts in Zambia: achieving high
adherence to test results using community health workers. Malaria Journal 10: 158.
http://www.malariajournal.com/content/pdf/1475-2875-10-158.pdf (last accessed 29
August 2014).
Chanda, P., B. Hamainza, and S. Mulenga. 2009. Malaria control and implications for the
management of fever in under-five children at a peripheral health facility: a case study of
Chongwe rural health centre in. Malaria Journal 8: 49.
http://www.malariajournal.com/content/8/1/49 (last accessed 1 September 2014).
Chipwaza, B., J.P. Mugasa, I. Mayumana, M. Amuri, C. Makungu, and P.S. Gwakisa. 2014.
Community knowledge and attitudes and health workers’ practices regarding non-malaria
febrile illnesses in eastern Tanzania. PLoS Neglected Tropical Diseases 8 (5): e2896.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=4031176 (last accessed 30
August 2014).
Clemmer, G. 2010. The GIS 20 essential skills 1rst ed. Redlands, California: ESRI Press.
Cromley, E.K. 2003. GIS and Disease. Annual Review of Public Health 24: 7–24.
Cromley, E.K., and S.L. McLafferty. 2012. GIS and public health 2nd ed. New York, New York,
USA: The Guilford Press.
d’Acremont V., A. Malila, N. Swai, R. Tillya, J. Kahama-Maro, C. Lengeler, and B. Genton.
2010. Withholding antimalarials in febrile children who have a negative result for a rapid
diagnostic test. Clinical Infectious Diseases: An Official Publication of the Infectious
Diseases Society of America 51 (5): 506–11.
DeRenzi, B., B. Birnbaum, L. Findlater, J. Mangilima, J. Payne, T. Parikh, G. Borriello, and N.
Lesh. 2012. Improving community health worker performance through automated SMS.
Proceedings of the Fifth International Conference on Information and Communication
Technologies and Development - ICTD ’12: 25.
http://dl.acm.org/citation.cfm?doid=2160673.2160677.
Derua, Y.A., D.R. Ishengoma, R.T. Rwegoshora, F. Tenu, J.J. Massaga, L.E. Mboera, and S.M.
Magesa. 2011. Users’ and health service providers' perception on quality of laboratory
malaria diagnosis in Tanzania. Malaria Journal 10: 78.
http://www.malariajournal.com/content/10/1/78 (last accessed 30 August 2014).
English, M., H. Reyburn, C. Goodman, and R.W. Snow. 2009. Abandoning presumptive
antimalarial treatment for febrile children aged less than five years--a case of running
before we can walk? PLoS medicine 6 (1): e1000015.
http://www.plosmedicine.org/article/info:doi/10.1371/journal.pmed.1000015 (last
accessed 25 November 2013).
90
Fisher, R.P., and B.A. Myers. 2011. Free and simple GIS as appropriate for health mapping in a
low resource setting: a case study in eastern Indonesia. International Journal of Health
Geographics 10: 15. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3051879/pdf/1476-
072X-10-15.pdf (last accessed 11 November 2012).
Freifeld, C.C., R. Chunara, S.R. Mekaru, E.H. Chan, T. Kass-Hout, A. Ayala Iacucci, and J.S.
Brownstein. 2010. Participatory epidemiology: use of mobile phones for community-
based health reporting. PLoS medicine 7 (12): e1000376.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2998443 (last accessed 21
August 2014).
Gosling, R.D., C.J. Drakeley, A. Mwita, and D. Chandramohan. 2008. Presumptive treatment of
fever cases as malaria: help or hindrance for malaria control? Malaria Journal 7: 132.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2488354 (last accessed 26
November 2013).
Hamer, D.H., M. Ndhlovu, D. Zurovac, M. Fox, K. Yeboah-Antwi, P. Chanda, N. Sipilinyambe,
J.L. Simon, and R.W. Snow. 2007. Improved diagnostic testing and malaria treatment
practices in Zambia. JAMA : The Journal of the American Medical Association 297 (20):
2227–31.
Hastings, I. 2011. How artemisinin-containing combination therapies slow the spread of
antimalarial drug resistance. Trends in Parasitology 27 (2): 67–72.
Hume, J.C.C., G. Barnish, T. Mangal, L. Armázio, E. Streat, and I. Bates. 2008. Household cost
of malaria overdiagnosis in rural Mozambique. Malaria Journal 7: 33.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2279141/pdf/1475-2875-7-33.pdf (last
accessed 26 November 2013).
Kalyango, J.N., T. Alfven, S. Peterson, K. Mugenyi, C. Karamagi, and E. Rutebemberwa. 2013.
Integrated community case management of malaria and pneumonia increases prompt and
appropriate treatment for pneumonia symptoms in children under five years in Eastern
Uganda. Malaria Journal 12: 340. http://www.malariajournal.com/content/12/1/340 (last
accessed 13 October 2013).
Kamanga, A., P. Moono, G. Stresman, S. Mharakurwa, and C. Shiff. 2010. Rural health centres,
communities and malaria case detection in Zambia using mobile telephones: a means to
detect potential reservoirs of infection in unstable transmission conditions. Malaria
Journal 9: 96. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=2861692 (last
accessed 16 October 2013).
Kandala, N.B., M.A. Magadi, and N.J. Madise. 2006. An investigation of district spatial
variations of childhood diarrhoea and fever morbidity in Malawi. Social Science &
Medicine 62 (5): 1138–52.
91
Kazembe, L.N., A.S. Muula, C.C. Appleton, and I. Kleinschmidt. 2007. Modelling the effect of
malaria endemicity on spatial variations in childhood fever, diarrhoea and pneumonia in
Malawi. International Journal of Health Geographics 6: 33. http://www.ij-
healthgeographics.com/content/6/1/33 (last accessed 29 August 2014).
Kulldorff, and Martin. 2014. SaTScan User Guide Version: SaTScan v9.3 released March 20
2014. http://www.satscan.org/techdoc.html (last accessed 29 August 2014).
Manyando, C., E.M. Njunju, J. Chileshe, S. Siziya, and C. Shiff. 2014. Rapid diagnostic tests for
malaria and health workers’ adherence to test results at health facilities in Zambia.
Malaria Journal 13: 166.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=4026818 (last accessed 29
August 2014).
Msellem, M.I., A. Mårtensson, G. Rotllant, A. Bhattarai, J. Strömberg, E. Kahigwa, M. Garcia,
M. Petzold, P. Olumese, A. Ali, and A. Björkman. 2009. Influence of rapid malaria
diagnostic tests on treatment and health outcome in fever patients, Zanzibar: a crossover
validation study. PLoS medicine 6 (4): e1000070.
Nhavoto, J.A., and A. Grönlund. 2014. Mobile technologies and geographic information systems
to improve health care systems: a literature review. JMIR mHealth and uHealth 2 (2):
e21. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=4114429 (last accessed
24 August 2014).
O’Dempsey, T.J., T.F. McArdle, B.E. Laurence, A.C. Lamont, J.E. Todd, and B.M. Greenwood.
1993. Overlap in the clinical features of pneumonia and malaria in African children.
Transactions of the Royal Society of Tropical Medicine and Hygiene 87 (6): 662–5.
O’Sullivan, D., and D.J. Unwin. 2010. Geographic information analysis 2nd ed. Hoboken, New
Jersey: John Wiley & Sons Inc.
Price, M. 2010. Mastering ArcGIS 4th ed. New York, New York, USA: McGraw-Hill.
Rafael, M., T. Taylor, A. Magill, Y. Lim, F. Girosi, and R. Allan. 2006. Reducing the burden of
childhood malaria in Africa: the role of improved. Nature 444 (Suppl 1): 39–48.
Rao, V., D. Schellenberg, and A. Ghani. 2013. Impact of improving appropriate treatment for
fever on malaria and non-malarial febrile illness management in under-5s: a decision-tree
modelling approach. PLoS ONE 8 (7): e69654.
http://dx.plos.org/10.1371/journal.pone.0069654.g005 (last accessed 1 September 2014).
Republic of Zambia Ministry of Health. 2011a. Annual Health Statistical Bulletin.
———. 2011b. National Health Strategic Plan 2011-2015.
———. 2012. National Malaria Control Programme Strategic Plan for FY 2011-2015.
92
Slater, H., and E. Michael. 2012. Predicting the current and future potential distributions of
lymphatic filariasis in Africa using maximum entropy ecological niche modelling. PLoS
ONE 7 (2): e32202. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3281123
(last accessed 22 August 2014).
Snow, R.W., K. Marsh, and D. le Sueur. 1996. The need for maps of transmission intensity to
guide malaria control in Africa. Parasitology Today 12 (12): 455–457.
Tanser, F.C., and D. le Sueur. 2002. The application of geographical information systems to
important public health problems in Africa. International Journal of Health Geographics
1: 4. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=149399 (last accessed 25
August 2014).
UNAIDS. Zambia Country Profile. http://www.unaids.org/en/regionscountries/countries/zambia/
(last accessed 5 August 2014).
United Nations Children’s Fund. 2013. Malaria. UNICEF Zambia Fact Sheets.
http://www.unicef.org/zambia/5109_8454.html (last accessed 15 November 2013).
United States of America Central Intelligence Agency. 2014. CIA World Factbook Africa:
Zambia. The World Factbook. https://www.cia.gov/library/publications/the-world-
factbook/geos/za.html (last accessed 22 April 2014).
White, L.J., P.N. Newton, R.J. Maude, W. Pan-ngum, J.R. Fried, M. Mayxay, R.R. Maude, and
N.P.J. Day. 2012. Defining disease heterogeneity to guide the empirical treatment of
febrile illness in resource poor settings. PLoS ONE 7 (9): e44545.
WHO. 2013. World Malaria Report 2013.
http://www.who.int/malaria/publications/world_malaria_report_2013/en/ (last accessed
21 August 2014).
———. 2014. Zambia Country Profile. http://www.who.int/countries/zmb/en/ (last accessed 20
March 2014).
World Health Organization. 2013a. Children: Reducing Mortality Fact Sheet No. 178. Media
Centre http://www.who.int/mediacentre/factsheets/fs178/en/ (last accessed 19 November
2013).
———. 2013b. Diarrheal Disease Fact Sheet No. 330. Media Centre
http://www.who.int/mediacentre/factsheet/fs330/en/ (last accessed 19 November 2013).
Young, M., C. Wolfheim, D.R. Marsh, and D. Hammamy. 2012. World Health
Organization/United Nations Children’s Fund joint statement on integrated community
case management: an equity-focused strategy to improve access to essential treatment
services for children. The American Journal of Tropical Medicine and Hygiene 87 (5
Suppl): 6–10.
93
Zambia Tourism Board. Official Zambia Tourism Website. http://www.zambiatourism.com (last
accessed 25 January 2014).
Zeiler, M. 1999. Modeling our world 1rst ed. Redlands, California: ESRI Press
94
APPENDICES
95
Appendix A: Population Density Maps
96
Chongwe district population densities for Ward administrative areas with village and RHC locations
97
Chongwe population densities for SEA administrative boundary areas and RHC locations
98
Appendix B: Disease/Illness Count Proportional Symbols Maps
99
Cases of Coughs with Problems Breathing
100
Cases of Fevers, Vomiting, & Diarrhea
101
Cases of Malaria, Fever, and RDTs
102
Cases of Pneumonia with Fever
103
Cases of Problems Breathing, Chest Pain & Coughing
104
Cases of Fever and Diarrhea
105
Cases of Malaria with Fever
106
Cases of Malaria and RDT Usage
107
Cases of Fever & Vomit
108
Appendix C: Disease/Illness Incidence Maps
109
Incidence of Diarrhea & Fever
110
Incidence of Problems Breathing, Chest Pain, Cough and Fever
111
Incidence of Fevers with Fever History
112
Malaria Incidence
113
Incidence of Fever
114
Incidence of Fever, Cough, & Problem Breathing
115
Incidence of Severe Diarrhea (Diarrhea & Vomiting) not referred
116
Incidence of Malaria with Fever
117
Incidence of Cough & Problems Breathing (Pneumonia)
118
Appendix D: SaTScan v9.3 Cluster Analysis Maps
119
SaTScan v9.3 Fever Incidence Cluster Areas with 50% of Population at Risk for Fever in Chongwe
SEA boundary areas
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
120
SaTScan v9.3 Malaria Incidence Cluster Areas with 50% of Population at Risk for Malaria in Chongwe
SEA boundary areas.
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
121
SaTScan v9.3 Malaria with Fever Incidence Cluster Areas with 50% of Population at Risk for Malaria
with Fever in Chongwe SEA boundary areas.
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
122
SaTScan v9.3 Diarrhea Incidence Cluster Areas with 50% of Population at Risk for Diarrhea Illness in
Chongwe SEA boundary areas.
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
123
SaTScan v9.3 Vomiting Incidence Cluster Areas with 50% of Population at Risk for Vomiting Illness in
Chongwe SEA boundary areas
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
124
SaTScan v9.3 Severe Diarrhea Incidence Cluster Areas with 50% of Population at Risk for Severe
Diarrhea Illness in Chongwe SEA boundary areas
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
125
SaTScan v9.3 Pneumonia Incidence Cluster Areas with 50% of Population at Risk for Pneumonia Illness
in Chongwe SEA boundary areas
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
126
SaTScan v9.3 Severe Pneumonia Incidence Cluster Areas with 50% of Population at Risk for Pneumonia
Illness in Chongwe SEA boundary areas
*Most-likely (Primary) Clusters are marked in darker shades of red and Secondary Clusters in lighter shades of Pink
Abstract (if available)
Abstract
The growing accessibility of mobile phones in developing countries has led to increased innovation and utilization of handheld technology in managing health outcomes. Mobile health (mHealth) technologies enabled significant gains in localized data collection methods and increased timeliness in disease surveillance and control programs. Mobile technology has become an important tool for point of care productivity and effective task shifting for Community Health Workers (CHW) in many developing countries. Concurrently, GIS technology has increasingly been utilized in public health research, planning, monitoring, and surveillance within many developing countries and low-resource settings. This has resulted in opportunities for better understanding of spatial variation of diseases and the correlations with environmental factors. ❧ To better understand community needs and burden of illnesses managed by CHWs, a geospatial analysis at the sub-district level was performed on CHW catchment area health data registries. Risk assessments and cluster analyses were conducted to identify high incidences of fever related illnesses for malaria, diarrhea, and pneumonia in community areas within the rural district area of Chongwe, Zambia. Seventy CHWs recorded 7,674 cases over a time-period of ten months, of which 3,130 cases were geocoded for geospatial analyses. One hundred forty-one village areas within 15 rural health center catchment areas were geocoded and mapped. Results were used to create thematic maps illustrating disease distribution and risks for malaria, pneumonia, and diarrheal illnesses for each sub-district village area manage by CHWs. The use of mobile technology integrated with GIS to manage community health data and the application of GIS to analyze community level data may provide further insight into local area disease distribution, variability, and community needs than systems focused solely on district level data analysis and lacking GIS integration.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand
PDF
Web GIS as a disease management workspace: enabling advocacy at multiple scales across multiple continents with the case of tungiasis
Asset Metadata
Creator
Metitiri, Mine
(author)
Core Title
Use of GIS for analysis of community health worker patient registries from Chongwe district, a rural low-resource setting, in Lusaka Province, Zambia
School
College of Letters, Arts and Sciences
Degree
Master of Science
Degree Program
Geographic Information Science and Technology
Publication Date
09/12/2014
Defense Date
09/04/2014
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
cluster analysis,community health,community health volunteers,community health workers,Development,DHIS2,eHealth,GIS,GIS integration,global health,Health policy,ICT,information communication technologies,low-resource settings,Lusaka,malaria,mHealth,mobile health,OAI-PMH Harvest,Public Health,rural disease surveillance,rural health centers,SaTScan,spatial epidemiology,spatial statistics,Zambia
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Longcore, Travis R. (
committee chair
), Ruddell, Darren M. (
committee member
), Warshawsky, Daniel N. (
committee member
)
Creator Email
metitiri@usc.edu,minemetitiri@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-475784
Unique identifier
UC11287201
Identifier
etd-MetitiriMi-2928.pdf (filename),usctheses-c3-475784 (legacy record id)
Legacy Identifier
etd-MetitiriMi-2928.pdf
Dmrecord
475784
Document Type
Thesis
Format
application/pdf (imt)
Rights
Metitiri, Mine
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
Tags
cluster analysis
community health
community health volunteers
community health workers
DHIS2
eHealth
GIS
GIS integration
global health
ICT
information communication technologies
low-resource settings
Lusaka
mHealth
mobile health
rural disease surveillance
rural health centers
SaTScan
spatial epidemiology
spatial statistics