University of Southern California Dissertations and Theses

Characterizing self-reported spatial and provider barriers to maternal health care utilization in Malawi

(USC Thesis Other)

Characterizing self-reported spatial and provider barriers to maternal health care utilization in Malawi

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content CHARACTERIZING SELF-REPORTED SPATIAL AND PROVIDER BARRIERS TO MATERNAL
HEALTH CARE UTILIZATION IN MALAWI
by
Lois Aareum Park
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POPULATION, HEALTH AND PLACE)
August 2022
Copyright 2022 Lois Aareum Park
Dedication
I dedicate this work to my family.
To my parents, Angela and David,
and to my sister and brother-in-law, Joanna and Eric,
for their love, unwavering support and encouragement.
Their faith in me gave me the confidence to move forward
through the most challenging times.
ii
Acknowledgements
I express a deep gratitude and appreciation to my co-chairs Yao-Yi Chiang and Meredith Franklin and
committee members Paul Marjoram and Emily Smith-Greenaway for their constant guidance and sup-
port. I would also like to thank Agbessi Amouzou and Sandy Eckel for serving on my proposal guidance
committee.
I would like to thank friends, colleagues, mentors, and supporters whose support has been vital to my
well-being and success: Kate, Emily, Avery, Li, Xiaozhe, Yan, Johanna, Yijun, Weiwei, Zekun, Population,
Health and Place colleagues, Knowledge Computing Lab colleagues, Amber, Sarah, Romesh, and Sacha,
Emma and Elie, Tricia, Emily, Talata, Safia, Aleks, Younger and Socci, Maribel, JB, JR, Scott, Namjoon,
Seokjin, Yoongi, Hoseok, Jimin, Taehyung, Jungkook, and canines Bob and Cody.
I would like to acknowledge Eugene Lang, who through the Lang Opportunity Scholarship at the Lang
Center for Civic and Social Responsibility at Swarthmore College, enabled me to step into the world of
public health, putting this aspect of my life’s trajectory into motion many years ago. I would also like to
acknowledge my teachers and friends from the Federal Way Public Academy, who nurtured and embraced
this curious mind.
I would like to acknowledge the International Cartographic Association (ICA) and the US National
Committee for the ICA which provided the funding and platform to present various components of this
work at the International Cartographic Conference (ICC) 2019 in Tokyo and ICC 2021 in Florence.
I would like to acknowledge the financial support of the PhD Program in Population, Health and Place.
I acknowledge all the unnamed supporters, teachers, and communities that supported me in the various
stages of learning in my life and all the individuals who spent many hours responding to questions to make
up the data that were used in this dissertation. Thank you.
iii
TableofContents
Dedication ii
Acknowledgements iii
ListofTables vii
ListofFigures viii
Abstract xi
Chapter1: Introduction 1
Chapter2: Howfaristoofar? Characterizingself-reporteddistanceasbarriertohealthcare
inruralMalawi(Paper1) 5
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Measures of geographic access to health care: distance and health service
environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.3 Malawi context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.1.1 2015-16 Malawi Demographic and Health Survey . . . . . . . . . . . . . 10
2.2.1.2 2013-14 Malawi Service Provision Assessment . . . . . . . . . . . . . . . 11
2.2.1.3 Malawi road network data . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.2 Measures of physical access to health care: Linking population and health facility
data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2.1 Euclidean distance link method . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2.2 Road network link method . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2.3 Euclidean buffer link (count) method . . . . . . . . . . . . . . . . . . . . 13
2.2.2.4 Double buffer catchment area (probability) method . . . . . . . . . . . . 14
2.2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3.1 Sensitivity analysis for Double Buffer Catchment Area (probability)
Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.3.2 Testing the four measures of physical access to health facilities . . . . . 16
2.2.3.3 Multilevel logistic modeling . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3.1 Measures of physical access to health facilities . . . . . . . . . . . . . . . . . . . . . 17
2.3.2 Multilevel logistic modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3.3 Sociodemographic differences in perceiving distance to be a barrier to health care . 20
2.3.3.1 Household wealth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3.3.2 Having a young child . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3.3.3 Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3.4 Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.3.5 Marital status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
iv
2.4.1 Measures of physical proximity to health care: distance and access environment . 23
2.4.2 How far is too far? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Chapter3: Self-reportedbarrierstohealthcareandtheutilizationofpregnancycareinrural
Malawi(Paper2) 26
3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.1 Maternal health in Malawi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 Indirect maternal deaths and antenatal care (ANC) . . . . . . . . . . . . . . . . . . 29
3.1.3 Enabling factors and barriers to antenatal care . . . . . . . . . . . . . . . . . . . . . 30
3.1.4 Self-reported barriers to care . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.2.1 Outcome variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.2.2 Explanatory variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.3 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3.2 Multilevel logistic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.2.1 Timely ANC initiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.2.2 ANC4+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3.2.3 District-level differences . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4.1 Age, education, and working status . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.2 Marital status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4.3 Unplanned pregnancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.4.4 Self-reported barriers to health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Chapter 4: Characterizing perceived provider barriers to health care and its relationship
withhealthproviderdensity(Paper3) 48
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.2.1.1 2015-16 Malawi Demographic and Health Survey . . . . . . . . . . . . . 51
4.2.1.2 2013-14 Malawi Service Provision Assessment . . . . . . . . . . . . . . . 52
4.2.1.3 WorldPop estimates of Malawi population . . . . . . . . . . . . . . . . . 54
4.2.2 Sociodemographic variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.3 Self-reported barriers to health care . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.4 Measures of skilled provider density . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.4.1 Closest facility provider density . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.4.2 Local provider density . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.5.1 Descriptive analysis and exploration of provider density measures . . . . 59
4.2.5.2 Multilevel logistic modeling . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.1 Measures of provider density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
v
4.3.2 Multilevel logistic models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Chapter5: Conclusions 75
5.1 Main contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.1.1 Substantive contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.1.2 Methodological contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Bibliography 79
vi
ListofTables
2.1 Background characteristics for women in rural DHS clusters (n=19,286), unweighted, by
whether distance is perceived to be a big barrier to obtaining health care . . . . . . . . . . 18
2.2 Summary of measures of spatial access, by whether distance is perceived to be a big
barrier to obtaining health care. p-values from Mann-Whitney U test to compare medians
in independent groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Summary of multilevel logistic models for whether distance is a big barrier in accessing
health care among women living in rural Malawi . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Summary of multilevel logistic models for whether distance is a big barrier in accessing
health care among women living in urban Malawi . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Background characteristics and explanatory variables among women in rural DHS clusters
who had a live delivery in the five years prior to the survey (n=11,082), unweighted, by
whether they initiated ANC in the first trimester and whether they attained ANC4+ . . . . 37
3.2 Summary of multilevel models for timely ANC initiation and ANC4+ among rural
Malawian women . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.1 Background characteristics perceiving high (2+) provider barrier to health care among
rural Malawian women (n=19,315) with different measures of health provider density. . . . 61
4.2 Summary of multilevel models for perceiving high (2+) provider barrier to health care
among rural Malawian women (n=19,315) with different measures of health provider density. 66
4.3 Summary of multilevel models for the concern that there will be no provider being a major
barrier to health care among rural Malawian women (n=19,315) with different measures
of health provider density. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Summary of multilevel models for ANC4+ among rural Malawian women (n=10,825) with
different measures of health provider density. . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5 Summary of multilevel models for ANC4+ among rural Malawian women (n=10,825) with
perceived provider barrier and local health provider density. . . . . . . . . . . . . . . . . . 70
vii
ListofFigures
1.1 Left: Map of Africa with Malawi shaded in purple; Right: Map of Malawi and its 28
administrative districts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Malawi trends for Maternal Mortality Rate (with 2015 target for Millennium Development
Goals (MDG) and 2030 target for Sustainable Development Goals (SDG)) and coverage
of key interventions for a healthy pregnancy: ANC1, ANC4+, Timely ANC initiation,
health facility delivery. Source: MDHS 2000, MDHS 2004, MDHS 2010, MDHS 2015-16 and
United Nations Maternal Mortality Estimation Inter-agency Group (MMEIG), September
2019 revision (updated 2021) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Map of Malawi (black outline) showing location of 977 health facilities surveyed in the
2013-14 Malawi Service Provision Assessment (red cross symbol) and 677 rural clusters
surveyed in the 2015-16 Malawi Demographic and Health Survey (blue circle symbol).
The inset map shows the African continent with Malawi highlighted in purple. . . . . . . . 9
2.2 Conceptual diagram of the four linking methods explored in this study: PanelA. Method
1: Shortest straight-line (Euclidean) distance to health facility; Panel B. Method 2:
Shortest routed distance to health facility; PanelC. Method 3: Count of facilities within
a 10km-radius buffer; PanelD. Method 4: Probability of not living in a 5-km catchment
area of a health facility. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Distribution of Measures 1-4 with medians (dashed lines) for those who report that
distance is (blue) and is not (red) a big barrier in getting health care. . . . . . . . . . . . . 16
3.1 Map of Malawi (black outline) and its 28 districts (yellow with red outline) showing
location of rural clusters surveyed in the 2015-16 Malawi Demographic and Health Survey 27
3.2 Malawi trends for Maternal Mortality Rate (with 2015 target for Millennium Development
Goals and 2030 target for Sustainable Development Goals) and coverage of key inter-
ventions for a healthy pregnancy: ANC1, ANC4+, Timely ANC initiation, health facility
delivery. Source: MDHS 2000, MDHS 2004, MDHS 2010, MDHS 2015-16 and United
Nations Maternal Mortality Estimation Inter-agency Group (MMEIG), September 2019
revision (updated 2021) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Histograms showing number of ANC visits by age group and timing of ANC initiation.
Dashed line shows median in each age-timing group. Average number of ANC visits
decreases as initiation of ANC is delayed across all age groups. . . . . . . . . . . . . . . . . 35
viii
3.4 Intercepts for districts for timely ANC initiation and ANC4+ (with ANC initiation) models.
Random effects are represented as odds ratios (blue represents positive effect and red
indicates negative effect) and their 95% confidence intervals. Lines between the left and
right panels link same district. Black circles/lines represent districts that had higher odds
of timely ANC initiation but reduced odds of ANC4+ and a drop in relative drop in ranking
of at least ten spots. Green circles/lines indicate districts that had reduced odds of timely
ANC initiation, but greater odds of ANC4+ accompanied by a relative improvement in
relative ranking of at least ten spots. All other districts are yellow. . . . . . . . . . . . . . . 40
3.5 Intercepts for districts for timely ANC initiation and ANC4+ (with ANC initiation)
models among rural women. Random intercepts are represented as odds ratios. The color
categories do not represent fixed OR ranges; the quantile method was used to classify a
similar number of districts to each color category. . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Map of Malawi (black outline) showing location of 977 health facilities surveyed in the
2013-14 Malawi Service Provision Assessment (red cross symbol) and 677 rural clusters
surveyed in the 2015-16 Malawi Demographic and Health Survey (blue circle symbol).
The inset map shows the African continent with Malawi highlighted in purple. . . . . . . . 53
4.2 Estimated total number of people per grid-cell for Malawi in 2014 by WorldPop. Each grid
is approximately a 100-meter square and has a value that is the total estimated number of
people covered by that area. Cell values range from 0 to 652 people, with a mean of about
1. Map of Malawi shown along with a zoomed-in inset to better demonstrate pixels. . . . . 55
4.3 Outline of steps for generating health facility-specific provider density and local (DHS
survey cluster) provider densities from three data sources: MSPA 2013-14, MDHS 2015-16,
and WorldPop population distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4 Top: Histogram of local skilled health provider density in total study population
(n=19,315); Bottom: Density plots of local skilled health provider density by the total
number of provider barriers (0-3) experienced by individuals. The three black vertical
lines in each density plot represent the Q1, Q2 (median), and Q3. The vertical dashed red
line in both top and bottom graphs represents the WHO-recommended density of 4.45
skilled providers per 1,000 population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.5 Top: Comparing the all health provider density of the closest facility (x-axis) vs local all
health provider density (y-axis). Bottom: Comparing the skilled health provider density
of the closest facility (x-axis) vs local skilled health provider density (y-axis). In both
top and bottom graphs, the red line represents best-fit line with shading representing
the confidence interval. Dashed black line is a reference line of slope 1. In both graphs,
the points above the dashed reference line represent survey clusters for which the local
provider density is higher than that of the closest health facility. . . . . . . . . . . . . . . . 63
ix
4.6 Summary of district-level estimates of skilled health provider density and the population
that is covered within the 5km buffer of a health facility. The top graph shows a positive
correlation with skilled provider density and the percentage of all health workers that are
skilled. The bottom graph demonstrates that an increase in skilled provider density is
associated with an increase in the proportion of the population that covered in a health
facility catchment area. Each circle represents a district and the size of the circles on the
bottom graph shows the size of the population that is not covered by a facility catchment
area to demonstrate relative size of the issue. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
x
Abstract
The overarching goal of this work was to characterize self-reported barriers to health care among rural
Malawian women with the aim of better understanding what these barriers represent in terms of objec-
tive measures. This work was motivated by the substantive challenge of understanding why only half
of pregnant women in Malawi obtained the WHO-recommended four or more antenatal care visits while
more than 90% of women got at least one visit in a way that went beyond linking outcomes to individ-
ual sociodemographic characteristics. This work also sought to demonstrate improved spatial methods of
linking population health surveys and facility surveys to contribute health system perspectives in analyses
of health outcomes.
The analyses presented here consistently demonstrate that perceived barriers are shown to reflect
objective measures of their correlates. Chapter 2 demonstrates that women who report distance to be a
major barrier to health care do, in fact, live farther from health facilities. Chapter 3 indicates that perceiving
barriers to care moderates the utilization of pregnancy health care. Finally, Chapter 4 shows that women
who report experiencing poor health system readiness live in areas of reduced skilled health provider
density. The main methodological contribution of this body of work is the innovative demonstration of a
spatial method of linking nationally-representative health and facility surveys.
The work presented here to characterize user perceptions of health care will ultimately allow for
decision-makers to leverage this rich perspective to allocate resources strategically and make policies to
improve the coverage, utilization, and impact of lifesaving health care.
xi
Chapter1
Introduction
Malawi, a small, landlocked country in south-central Africa, is has one of the highest rates of maternal
mortality in the world (Figure 1.1). While the maternal mortality ratio (MMR), the number of maternal
deaths per 100,000 live births, has been on the decline, the gains were not sufficient to reach the MMR
target in 2015 for the end of the Millennium Development Goal era ([84, 17]). With the most recent estimate
of MMR at 349 maternal deaths per 100,000 live births, an accelerated reduction is needed to meet the
Sustainable Development Goal (SDG) for MMR by the target year of 2030 ([84]). The SDG 3.1 Target for
MMR is a global average of 70 deaths per 100,000 live births, with no individual country above 140 deaths
per 100,000 live births ([28], Figure 1.2).
Figure 1.1: Left: Map of Africa with Malawi shaded in purple; Right: Map of Malawi and its 28 adminis-
trative districts
1
349
155
70
0
100
200
300
400
500
600
700
800
900
1000
2000 2005 2010 2015 2020
Maternal Mortality Ratio
(maternal deaths per 100,000 live births)
MMR
MDG 5 target
SDG 3.1 target
55
69
73
91 91
92
95 95
56
57
46
51
7
8
12
24
0
10
20
30
40
50
60
70
80
90
100
2000 2004 2010 2015
Coverage of service (%)
ANC1
Health facility delivery
ANC4+
Timely ANC initiation
Figure 1.2: Malawi trends for Maternal Mortality Rate (with 2015 target for Millennium Development Goals
(MDG) and 2030 target for Sustainable Development Goals (SDG)) and coverage of key interventions for
a healthy pregnancy: ANC1, ANC4+, Timely ANC initiation, health facility delivery. Source: MDHS 2000,
MDHS 2004, MDHS 2010, MDHS 2015-16 and United Nations Maternal Mortality Estimation Inter-agency
Group (MMEIG), September 2019 revision (updated 2021)
While the MMR is short of meeting global targets, Malawi intentionally and swiftly increased facility
deliveries through aggressive policies ([18]; Figure 1.2). Tracked alongside MMR is SDG Indicator 3.1.2, the
proportion of deliveries attended by skilled health personnel trained in delivering lifesaving interventions
during labor, delivery, and in the postpartum period ([82]). This measure includes doctors, nurses, and
midwives, and excludes traditional birth attendants. The indicator shown in Figure 1.2 is health facility
delivery, a proxy and somewhat conservative estimate for the indicator of having a skilled attendant at
birth (SAB), since all deliveries in a facility are assumed to be attended by a skilled provider and there
may be deliveries outside the health facility that are also attended by a skilled provider. Health facility
deliveries increased from 55% in 2000 to 91% in 2015 ([50], Figure 1.2). This meets the 90% global target
put forth by the World Health Organization (WHO) and UN Population Fund (UNFPA)-led global, multi-
partner initiative for Ending Preventable Maternal Mortality (EPMM) ([21]). EPMM also recommends a
2
90% global coverage of four or more antenatal care visits (ANC4+), with 90% of countries achieving at least
80% coverage ([21]).
In contrast to the near doubling of health facility delivery coverage from 2000 to 2015, the coverage
of ANC4+ declined from 56% to 51% in the same time period ([50], Figure 1.2). This persistent lack of
improvement in ANC4+ coverage stands out especially in the context of a consistently high rate (90% or
higher) of attending at least one ANC visit (ANC1) and an increasing trend of timely ANC initiation in the
first trimester, or first 12 weeks, of pregnancy. These contrasting trends suggest that there are persistent
barriers that prevent women from attending the recommended four or more visits during pregnancy even
while almost all women attend one visit and an increasing number of women initiate ANC early in preg-
nancy ([31, 38]). This conflicting portrait of maternal health care, with improvements and upward trends
on one hand and stagnation and decrease in the case of ANC4+ motivates the substantive aim for this body
of work: to characterize potential barriers to achieving ANC4+.
This dissertation centers around characterizing a series of questions on the DHS about perceived bar-
riers to care. The 2015-16 Malawi DHS specifically asks about seven factors that might be major barriers
including: (1) distance to the health facility; (2) getting permission to go to the doctor; (3) getting money
needed for advice or treatment; (4) not wanting to go alone; (5) concern that there may not be a female
provider; (6) concern that there may not be any health provider; (7) concern that there may be no drugs
available ([50]). These data on self-reported barriers in patient, provider, and system-level domains has
the potential to offer a unique perspective that can provide insight into the relationship between the pres-
ence of potential barriers, whether they are perceived as barriers, and whether these perceived barriers
influence health service utilization. While these questions have a lot of potential, they are underutilized in
health research because there is little understanding of their meaning and significance. The three subse-
quent chapters seek to characterize different facets of these self-reported barriers in order to demonstrate
their validity and value for use in future health research.
3
Chapter 2 explores whether perceiving distance to be a major barrier to health care is related to ob-
jective measures of distance to the nearest health facility and health care access. Chapter 3 then studies
whether these perceived barriers are associated with the utilization of antenatal care. Finally, Chapter 4
studies perceived provider barriers and their relationship with health provider density.
Another goal of this body of work is to demonstrate how innovative applications of spatial data pro-
cessing and analysis can link various data sources together. Instead of analyzing population health and
health facility data independently, their respective spatial attributes can be the key in linking these at-
tributes so that supply-side perspectives can complement our understanding of health outcomes.
The work presented here will ultimately allow for decision-makers to leverage these richer perspectives
to allocate resources strategically and make policies to improve the coverage, utilization, and impact of
lifesaving health care.
4
Chapter2
Howfaristoofar? Characterizingself-reporteddistanceasbarrierto
healthcareinruralMalawi(Paper1)
2.1 Introduction
2.1.1 Background
Acknowledging persistent disparities in proximity to health facilities, researchers have carefully docu-
mented how proximity to health facilities influences service utilization or health outcomes. Poor geo-
graphic access to health care is generally associated with poor. Greater distances increase travel time,
ultimately reduce health care utilization, and are associated with worse health outcomes and higher mor-
tality ([34, 7, 75, 63, 35]).
However, it cannot be assumed that the presence of a physical barrier like distance is the sole cause
of a reduction in health care utilization. A critical link between the two is whether a potential factor is
perceived by an individual to be a significant barrier to actually impact their behavior. The perception of
something like distance as a barrier to health care may be influenced by social, demographic, and cultural
factors, including individual lived experiences such as prior experience with the health system. Research
on distance to provider and health care utilization suggests there is a complex web of factors beyond
distance that influences patient care-seeking behaviors and distance tolerance in seeking health care ([57,
79]).
5
In this study, measures of geographic access to health facilities are compared to the perception that
distance is a major barrier to obtaining health care. The Demographic and Health Survey (DHS), a widely-
implemented population health survey in low-and middle-income countries (LMICs), asks reproductive-
aged women about seven factors and whether they are major barriers in seeking medical advice or treat-
ment for themselves when they are sick ([50]). These self-reported barriers include: (1) getting permission
to go to the doctor; (2) getting money needed for advice or treatment; (3) not wanting to go alone; (4)
distance to the health facility; (5) concern that there may not be a female provider; (6) concern that there
may not be any health provider; (7) concern that there may be no drugs available ([50]).
The self-assessed barriers measured by the DHS have the potential to lend an important perspective in
understanding individual health care-seeking behavior. Self-rated health has come to gain greater impor-
tance in medicine with increased evidence of being a valid measure of health status and even a predictor
of mortality ([36, 40]). Self-reported barriers to health care are reflective of how an individual internalizes
the potential impact of a barrier, considering not just the presence of the barrier, but other factors that
might exacerbate or reduce its impact on care-seeking behaviors. These perceived barriers may be better,
more inclusive indicators of individual context and health behaviors than more objective measures, such
as distance to the nearest health facility. However, these measures of perceived barriers need to be better
understood and characterized. Better characterization of these self-reported barriers allows us use them to
improve our understanding of health care utilization without relying on, or to complement external data
sources. Instead of the usual reliance on socioeconomic characteristics to motivate health care-seeking
behaviors, these barrier indicators may be able to offer a more nuanced and more actionable portrait of
the factors that motivate and hinder health care utilization.
Studying the relationship between objective geographic access and the perception that distance is a
barrier in the context of other sociodemographic factors allows us to understand inequities that might exist
6
in this dynamic. This study focuses on exploring these factors in rural populations, which face greater
challenges with distance to health care compared to urban populations ([35, 89, 6]).
The first goal of this study is to understand and quantify how the perception of distance as a major
barrier to health care is related to objective, physical measures of geographic access to health facilities.
This investigation allows us to quantify what is considered too far to pose as a barrier in seeking health
care. Understanding how far is far enough to discourage care-seeking behaviors can, in turn, inform future
studies of the geographic reach of health facilities and their services as perceived by users.
2.1.2 Measuresofgeographicaccesstohealthcare: distanceandhealthservice
environment
Another aim of this study is to advance the linking of household surveys and health facility assessments.
A common measure of geographic access to health care is the distance between a household and a service
delivery point ([67]). Given little training in geographic information system (GIS) software and tools,
a straight-line distance is simple to calculate for a set of household and health facility locations. With
road network data and a routing algorithm, a routing distance can be generated, which better reflects
actual distances people need to travel to reach their nearest health facility ([60]). However, some question
whether the cost of obtaining improved, sophisticated measures of distance is worth the gains in accuracy
([58]). Some argue that for most public health applications, straight-line distance is often good enough as
a proxy of health care access ([58]).
While distance to the nearest health facility is simple and relatively easy to derive, it is one-dimensional
and leaves much room for improvement as a general measure of health care access. In addition, a layer of
complexity is added by the displacement of the DHS survey cluster to protect respondent privacy ([11]).
The DHS makes publicly available the spatial coordinates for the center of a DHS survey cluster, which
is the sampling unit for the survey from which 30-33 households are selected for interviews ([50]). This
7
makes linking to just one, nearest facility vulnerable to error. Instead of a one-to-one linkage of one
survey cluster to one specific health facility, a more accurate strategy would be to approximate the service
environment ([73, 12]). As an improvement, the buffer linkage method, which is a count of health facilities
within a certain range of a survey cluster, has been proposed to better account for multiple facilities that
might be in close proximity ([89]). In addition to providing a more realistic idea of access to health services
by being able to account for multiple, close health facilities, this method allows for the selection of buffer
sizes that can reduce the error introduced by the displacement of survey cluster locations in the DHS ([12]).
Larger buffer sizes that cover more surface area are increasingly likely to include the true location of the
DHS cluster that has been obscured through the displacement procedure.
The limitation of the buffer linkage method is the assumption that all households within a survey clus-
ter buffer have equal access to a health facility anywhere within that buffer. The reality is that even within
the buffer, those close to the health facility will have greater access compared to households living far-
ther away. This paper introduces an innovative adaptation of an existing two-buffer method, the two-step
floating catchment area (2SFCA) method, to link DHS survey clusters to a service environment, improving
on the buffer linkage method ([47]). All of these metrics are described in greater detail in the methods
section.
2.1.3 Malawicontext
Malawi is a relatively small, landlocked country in southeast Africa (Figure 2.1). With a population of just
over 18 million and an average life expectancy at birth of 63 years for men and 68 years for women, it ranks
174 out of 189 countries on the UN Human Development Index ([85, 83]). Most of Malawi’s population is
young and rural. A little over 16% of the population lives in urban areas, which is one of the lowest rates
of urbanization in the world ([68]). The median age is 17, with almost 45% of the population under the age
of 15 and 15% under five years of age ([49]).
8
R u r a l D H S C l u s t e r s
H e a l t h F a c i l i t i e s
M a l a w i
Figure 2.1: Map of Malawi (black outline) showing location of 977 health facilities surveyed in the 2013-14
Malawi Service Provision Assessment (red cross symbol) and 677 rural clusters surveyed in the 2015-16
Malawi Demographic and Health Survey (blue circle symbol). The inset map shows the African continent
with Malawi highlighted in purple.
9
Malawi has several advantages as a study context for linking DHS, a population health survey, and the
Service Provision Assessment (SPA), a standardized health facility survey. The first is that the most recent
DHS (2015-16) and SPA (2013-14) for Malawi are relatively close to each other in time, which improves
the comparability ([50, 54]). Another advantage of setting this study in Malawi is that the SPA is a census
of health facilities, providing a complete database of facility locations and capacities, whereas the SPA is
usually a sample survey in other, larger countries.
Taken together, these two data sets have the potential to provide a rich narrative around health care
utilization and service environment. This study informs how best to link these two surveys so that their
full potential can be realized.
2.2 Methods
2.2.1 Datasources
The data used in this study come from three sources: the 2015-16 Malawi Demographic and Health Survey,
the 2013-14 Malawi Service Provision Assessment, and OpenStreetMap.
2.2.1.1 2015-16MalawiDemographicandHealthSurvey
The 2015-16 Malawi Demographic and Health Service (MDHS) is a nationally representative survey provid-
ing population-level estimates of key indicators of health including fertility and family planning, HIV/AIDS,
nutrition, infant, child, and maternal health and mortality, among other indicators of population demo-
graphic characteristics and health. The 2015-16 MDHS employed a two-stage sample survey design in-
formed by the most recent census prior to the survey in 2008 in that each cluster is an enumeration area
from the survey ([50]). In the first stage of sampling, 850 clusters (173 urban and 677 rural) was selected to
10
represent urban and rural areas of all 28 districts of Malawi. In the second stage of sampling, 26,361 house-
holds were selected from among the 850 survey clusters. These data, along with the centroid (geographic
coordinates of the center of the survey cluster) of each of the 850 clusters are publicly available.
In order to ensure the confidentiality of the respondents, the geographic data of Demographic and
Health Surveys undergo a process of geographic masking before public release, where locations are dis-
placed in accordance with well-documented procedures ([10]). The location of urban clusters is displaced
up to 2 kilometers (0-2km) and the location of rural clusters is displaced up to 5 kilometers (0-5km). A
random subset of 1% of the rural clusters are displaced up to 10 kilometers.
The dynamics of geographic barriers and access to healthcare likely differ between rural and urban
areas so this study focuses only on rural areas. The response of 19,286 respondents from 677 rural clusters
to the question about whether distance to a health facility is a big barrier in them accessing health care,
their sociodemographic characteristics, and the geographically displaced location of their survey cluster
are used in this study (Figure 2.1).
2.2.1.2 2013-14MalawiServiceProvisionAssessment
The 2013-14 MSPA is a census of all health facilities in Malawi collecting data on facility infrastructure and
readiness to deliver quality health services ([54]). The location of each of the 977 surveyed health facilities
is supplied with the survey data and not displaced prior to public release (Figure 2.1).
2.2.1.3 Malawiroadnetworkdata
Malawi road network data from OpenStreetMap, downloaded on July 7, 2021 via Geofabrik, were used to
generate the routing distances between DHS clusters and their closest health facility ([64]).
11
2.2.2 Measuresofphysicalaccesstohealthcare:Linkingpopulationandhealthfacility
data
Four methods of linking were studied (Figure 2.2). The linkage methods vary in terms of complexity and
ease of implementation to how conservative they are, given that the DHS survey cluster locations are
displaced.
2 facilities
within buffer
77%
probability of
NOT being in
health facility
catchment area
4.4 km
5.2 km
4.8 km
6.5 km
7.3 km
5.1 km
A. Method 1: Shortest straight-line distance B. Method 2: Shortest routed distance
C. Method 3: Euclidean buffer D. Method 4: Double buffer catchment area
Figure 2.2: Conceptual diagram of the four linking methods explored in this study: Panel A. Method 1:
Shortest straight-line (Euclidean) distance to health facility;PanelB. Method 2: Shortest routed distance
to health facility;PanelC. Method 3: Count of facilities within a 10km-radius buffer; PanelD. Method 4:
Probability of not living in a 5-km catchment area of a health facility.
12
2.2.2.1 Euclideandistancelinkmethod
Euclidean distance link is the straight-line distance from a DHS cluster to its closest health facility (Figure
2A). While this measure may not reflect a realistic distance traveled by users, it is simple to implement
with little technical skill and a commonly used method to link household and facility data.
2.2.2.2 Roadnetworklinkmethod
Road network linkage improves on the Euclidean distance link method by using road network data to
route a more realistic path from a DHS cluster to its closest health facility (Figure 2B). However, accurate
road network distances require an accurate and sufficiently detailed road network data and the means to
implement a routing algorithm that makes this more complex to implement than the Euclidean distance
link method. Using the Origin-Destination matrix tool in QGIS (3.16.10-Hannover), the routed distance
between every DHS cluster and every health facility was obtained and the shortest distance for every
DHS cluster was retained for this measure. A smaller value indicates greater proximity to a health facility
and implies greater access to health care. One DHS cluster in rural Zomba district, located on the border
with Mozambique, east of Lake Chilwa, was routed nearly 80 kilometers to the nearest health facility. This
cluster was excluded from the analysis as an outlier. All other rural DHS clusters were routed to its nearest
health facility within 25 kilometers.
2.2.2.3 Euclideanbufferlink(count)method
The Euclidean buffer link method draws a 10-kilometer radius buffer around each rural DHS cluster and
counts the number of health facilities within that buffer (Figure 2.2C). Since 10 kilometers is the maximum
displacement distance of the rural DHS clusters, this measure is viewed as a conservative estimate of the
potential health service environment to which a cluster may have access. By taking this approach, this
method tries to avoid misclassification or linking the household to a facility that it does not actually use
13
([89]). This is the method of linking household and health facilities recommended by ICF, which conducts
both the DHS and SPA datasets analyzed here ([89]). A smaller value here also indicates greater proximity
to a health facility and implies greater access to health care.
2.2.2.4 Doublebuffercatchmentarea(probability)method
We propose the double buffer probability area link method to improve the Euclidean buffer link method
above. One limitation of the Euclidean buffer link (count) method is that it is not able to account for any
health facilities that may be on the border of the 10-kilometer radius buffer, even if it may be the closest
facility to households, whose true location is unknown within the buffer. The other assumption made is
that the facilities within a buffer serves all populations within that buffer equally, when the reality is that
the facilities in the buffers are more accessible to those who are closer than those who live farther away.
Just as drawing a buffer around the DHS cluster allows for the consideration of the uncertainty introduced
by the displacement procedure, a buffer around the undisplaced health facilities could also account for the
reality that health facilities serve catchment areas and communities that live in relatively close proximity
(Figure 2.2D).
Five-kilometer radius buffers were drawn around each cluster and five-kilometer radius buffers were
drawn around each health facility. The World Health Organization (WHO) recommends that populations
live within a five-kilometer catchment area (approximately one-hour walking distance) of a health facility
to achieve Sustainable Development Goal (SDG) 3.8.1 for universal coverage of essential health services
([93]). Five kilometers is also a meaningful distance for clusters because it reflects the fact that 99 percent
of the rural clusters are displaced up to five kilometers. The DHS cluster and health facility catchment
area buffers were overlaid to identify the overlapping area. The overlapping area was divided by the total
area of the DHS buffer to derive a percent of the DHS cluster buffer covered by a health facility catchment
area. In cases where the buffers of two close facilities overlap with each other, such as in Figure 2.2D, the
14
overlapping area is not counted twice. This measure of health service environment at the cluster level
reflects the probability of a household being within an x kilometer catchment area of a health facility.
Finally, the inverse of this measure was used to align the direction of the measure and interpretation with
the Euclidean distance and routing distance methods where a smaller number (shorter distance to health
facility) is associated with greater access to health care.
While the five-kilometer buffer sizes have a well-founded rationale, a sensitivity analysis with various
buffer sizes was conducted to see if any other buffer size would result in a larger difference between the
two groups (perceiving distance to be/not be a barrier). However, the sensitivity analysis showed that the
measure was best (the difference between the averages is maximized) when both buffer sizes were five
kilometers (Table 2.2). The final measure is the probability that a household is not within a five-kilometer
catchment area of a health facility. A smaller number is associated with greater access. For example, 0%
indicates that all of the five-kilometer DHS cluster buffer overlaps with a five-kilometer catchment area of
a health facility and 100% indicates that none of the buffer overlaps with a five-kilometer catchment area
of a health facility.
2.2.3 Analysis
2.2.3.1 SensitivityanalysisforDoubleBufferCatchmentArea(probability)Method
The Double Buffer Catchment Area Method calculates the probability that a household lives within a
catchment area of a health facility. The first buffer in this method refers to the five or 10-kilometer buffer
around the DHS cluster to account for the displaced location. The second buffer refers to the catchment
area of the health facility, which is a one-, two-, five-, or 10-kilometer radius buffer. Given two buffer sizes
for the DHS cluster and four buffer sizes for the health facility catchment areas, this measure was calculated
for each combination to determine which measure had greater sensitivity (measured here as the greatest
difference between those who perceive distance to be a barrier and those who do not perceive distance
15
to be a barrier). The five-kilometer radius buffer for both DHS cluster and health facility proved to have
the greatest difference in measurement among the two groups and was used for this measure (Table 2.2,
Figure 2.3D).
No Y es
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
0
500
1000
1500
2000
2500
0
500
1000
1500
2000
2500
Euclidean Buffer Link (number of health facilities
in 10km buffer of DHS cluster)
Count
Is distance is
a big problem
in accessing
health care?
No
Y es
0.00
0.05
0.10
0.15
0.20
0 5 10 15
Shortest straight−line (Euclidean) distance
to nearest health facility (km)
Density
Is distance is
a big problem
in accessing
health care?
No
Y es
0.000
0.005
0.010
0.015
0.020
0.025
0 25 50 75 100
Probability (%) of NOT being in
5km catchment area of a health facility
Density
Is distance is
a big problem
in accessing
health care?
No
Y es
0.00
0.05
0.10
0.15
0 5 10 15 20
Shortest routed distance
to nearest health facility (km)
Density
Is distance is
a big problem
in accessing
health care?
No
Y es
A. Method 1: Shortest straight-line distance B. Method 2: Shortest routed distance
C. Method 3: Euclidean buffer
(Count of facilities in 10km radius of a DHS cluster)
D. Method 4: Double buffer catchment area (Probability
(%) of NOT being in 5km catchment area of a facility)
Worse access Better access Worse access Better access
Worse access Better access Better access Worse access
Figure 2.3: Distribution of Measures 1-4 with medians (dashed lines) for those who report that distance is
(blue) and is not (red) a big barrier in getting health care.
2.2.3.2 Testingthefourmeasuresofphysicalaccesstohealthfacilities
To assess these measures of geographic access, we compared the average of each of the four measures
among those who perceived distance to be a major barrier in accessing health care and those who did not
16
perceive distance to be a major barrier. We used the Mann-Whitney U test to determine the statistical
significance of the differences between the two groups.
2.2.3.3 Multilevellogisticmodeling
Given the nested structure of the data, we implemented multilevel logistic models to determine the rela-
tionship between each of the four objective measures of physical access and the single, self-reported sub-
jective measure, adjusting for age, educational attainment, household wealth, marital status, and whether
the individual has a young child under 5 years of age.
2.3 Results
2.3.1 Measuresofphysicalaccesstohealthfacilities
Those who perceive distance to be a big barrier in accessing health care live, on average, 4.5 kilometers
(Euclidean distance) or 5.9 kilometers (routed distance) from their nearest health facility (Figure 3). This
is significantly greater than the 3.1 kilometers (Euclidean distance) or 4.1 kilometers (routed distance)
estimated for their counterparts who say distance is not a major barrier (p< 0.001 for both Euclidean and
routed distances; Table 2.2). The differences between the Euclidean and routed distances to the nearest
health facility is not surprising, as it is expected for straight-line distances to be underestimates of routed
distances. Both groups have an average of 3 health facilities in a 10-kilometer radius area–confirming that
there is no relationship between the subjective measure and this objective one.
By comparison, those who said that distance poses a big problem in accessing health care were twice
as likely to not be living within a five-kilometer catchment area of a health facility ( p< 0.001; Table 2.2).
Whereas individuals for whom distance isnot a big barrier in getting health care live in areas where 80%
is within five-kilometers of a health facility, individuals who say distance is a big barrier in getting health
17
care live in places where about 60% of the area is within a catchment area of a health facility (p < 0.001;
Table 2.2).
Table 2.1: Background characteristics for women in rural DHS clusters (n=19,286), unweighted, by whether
distance is perceived to be a big barrier to obtaining health care
No,distancenotabarrier
(n=8,000)
%(n)
Yes,distanceisabarrier
(n=11,286)
%(n)
All
(n=19,286)
%(n)
Age
15-19 23.4% (1869) 20.2% (2277) 21.5% (4146)
20-34 50.7% (4052) 50.9% (5744) 50.8% (9796)
35-49 26.0% (2079) 28.9% (3265) 27.7% (5344)
Education
No education or primary school 76.2% (6096) 83.2% (9385) 80.3% (15481)
Secondary or higher 23.8% (1904) 16.8% (1901) 19.7% (3805)
Householdwealthquintile
(onlyamongruralHHs)
Poorest 15.1% (1208) 21.4% (2413) 18.8% (3621)
Poorer 17.2% (1372) 20.5% (2308) 19.1% (3680)
Middle 17.9% (1432) 20.1% (2264) 19.2% (3696)
Richer 19.3% (1542) 20.6% (2324) 20.0% (3866)
Richest 30.6% (2446) 17.5% (1977) 22.9% (4423)
Workingstatus
Not currently working 39.5% (3156) 34.9% (3944) 36.8% (7100)
Currently working 60.6% (4844) 65.1% (7342) 63.2% (12186)
Maritalstatus
Never in union 22.0% (1757) 17.8% (2010) 19.5% (3767)
Married or living with partner 65.6% (5246) 67.6% (7633) 66.8% (12879)
Widowed, divorced, or separated 12.5% (997) 14.6% (1643) 13.7% (2640)
Frequencyofmediaexposure
(radio/tv/newspaper/magazine)
None 45.0% (3602) 51.8% (5841) 49.0% (9443)
Less than once a week 17.7% (1413) 17.9% (2023) 17.8% (3436)
At least once a week 37.3% (2985) 30.3% (3422) 33.2% (6407)
Haschildunder5
No 44.2% (3534) 41.2% (4647) 42.4% (8181)
Yes 55.8% (4466) 58.8% (6639) 57.6% (11105)
2.3.2 Multilevellogisticmodeling
In general, women who are older, poorer, and those with a child under 5 years of age have a greater odds
of perceiving distance to be a big barrier in obtaining health care (Table 3). The multilevel models using
18
Table 2.2: Summary of measures of spatial access, by whether distance is perceived to be a big barrier to
obtaining health care. p-values from Mann-Whitney U test to compare medians in independent groups
No,distanceisnotabarrier
(n=8,000)
median(Q1-Q3)
Yes,distanceisabarrier
(n=11,286)
median(Q1-Q3)
p-value
m1(Euclideandistance) 3.1 (1.7-4.7) 4.5 (2.8-6.2) <0.001
m2(Routeddistance) 4.1 (2.4-6.4) 5.9 (3.7-8.2) <0.001
m3(Countoffacilitiesin10kmbufferaroundDHScluster) 3.0 (2.0-5.0) 3.0 (2.0-5.0) 0.0431
m4(ProbabilityofNOTbeinginXkmcatchmentareaofafacility)
m4a(10kmbufferaroundDHScluster)
1km health facility buffer 2.8 (1.7-4.4) 2.8 (1.7-4.1) 0.0779
2km health facility buffer 11 (6.6-16) 10 (6.5-15) 0.00655
5km health facility buffer 54 (39-73) 53 (38-69) <0.001
10km health facility buffer 100 (95-100) 100 (93-100) <0.001
m4b(5kmbufferaroundDHScluster)
500m health facility buffer 1.0 (0.85-1.9) 1.0 (0-1.0) <0.001
1km health facility buffer 4.0 (2.8-6.9) 3.5 (0-4.2) <0.001
2km health facility buffer 16 (9.8-24) 12 (2.0-18) <0.001
5km health facility buffer 78 (54-93) 60 (35-84) <0.001
10km health facility buffer 100 (100-100) 100 (100-100) <0.001
Euclidean distance and the DBCAP value as measures of physical access were better fitting (with relatively
lower AIC) compared to those with routed distance or count of near facilities.
Based on Model 1, which uses Euclidean distance, every kilometer increase in the straight-line distance
to the nearest health facility is associated with a 27 percent increase in the odds of perceiving distance to
be a big barrier (p< 0.001; Table 2.3). The increase in odds decreases slightly for routed distance (Model
2), which is expected given routed distances are underestimated by straight-line distances (18%,p< 0.001;
Table 2.3).
As foreshadowed in the descriptive analysis, the number of facilities in a 10-kilometer radius does not
influence whether or not distance is perceived to be a barrier to accessing health care (Model 3, Table 2.3).
In comparison, in Model 4 for the other measure of healthcare access, for every 10 percent increase the
chance of not being within a five-kilometer catchment area of a health facility, the odds of perceiving
distance to be a big barrier increase 23 percent (p < 0.001; Table 2.3). This impact is comparable to the
impact of living one (straight) kilometer farther from the nearest health facility.
Of these four objective measures of health care access, Model 1 and Model 4, using Euclidean distance
and the catchment area probability method, respectively, have lower AIC values compared to the other
19
Table 2.3: Summary of multilevel logistic models for whether distance is a big barrier in accessing health
care among women living in rural Malawi
Model1 Model2 Model3 Model4
Predictors OR CI p OR CI p OR CI p OR CI p
(Intercept) 0.54 0.42–0.69 <0.001 0.56 0.44–0.73 <0.001 1.49 1.19–1.87 0.001 0.67 0.52–0.86 0.002
Age
(ref: 15-19yearsold)
20-34 years old 1.23 1.09–1.38 <0.001 1.23 1.09–1.38 <0.001 1.23 1.09–1.38 <0.001 1.23 1.09–1.38 0.001
35-49 years old 1.35 1.19–1.54 <0.001 1.35 1.19–1.54 <0.001 1.35 1.19–1.54 <0.001 1.35 1.19–1.54 <0.001
Education
(ref: Noneorprimary)
Secondary or higher 1.00 0.91–1.10 0.945 1.00 0.91–1.10 0.970 1.00 0.91–1.10 0.952 1.01 0.92–1.10 0.906
Householdwealthquintile
(ref: Poorest)
Poorer 0.86 0.77–0.95 0.005 0.85 0.77–0.95 0.005 0.86 0.77–0.96 0.006 0.86 0.77–0.96 0.006
Middle 0.87 0.78–0.98 0.017 0.87 0.78–0.97 0.016 0.87 0.78–0.98 0.019 0.87 0.78–0.98 0.018
Richer 0.97 0.86–1.09 0.581 0.97 0.86–1.08 0.565 0.96 0.86–1.08 0.525 0.97 0.87–1.09 0.617
Richest 0.69 0.61–0.78 <0.001 0.69 0.61–0.77 <0.001 0.68 0.60–0.77 <0.001 0.69 0.61–0.78 <0.001
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.02 0.90–1.16 0.735 1.02 0.90–1.16 0.719 1.02 0.90–1.17 0.708 1.02 0.90–1.16 0.728
Widowed, divorced, or separated 1.16 1.00–1.36 0.055 1.16 1.00–1.36 0.054 1.16 0.99–1.35 0.063 1.17 1.00–1.36 0.053
Hasayoungchildunder5years 0.92 0.84–1.00 0.046 0.92 0.84–1.00 0.047 0.92 0.85–1.00 0.051 0.92 0.84–1.00 0.045
Measuresofaccesstohealthfacility
m1 (Straight-line distance) 1.27 1.22–1.32 <0.001
m2 (Routed distance) 1.18 1.15–1.22 <0.001
m3 (Count of facilities within
10km radius) 1.00 0.98–1.02 0.804
m4 (Chance of NOT being in 5km
catchment area of a facility) 1.23 1.19–1.27 <0.001
RandomEffects
σ 2
3.29 3.29 3.29 3.29
τ 00 1.05
cluster:district
1.09
cluster:district
1.37
cluster:district
1.05
cluster:district
0.16
district
0.15
district
0.18
district
0.21
district
ICC 0.27 0.27 0.32 0.28
AIC 22260.7 22283.2 22412.8 22269.6
models, suggesting better fit and greater relevance in explaining the perception that distance is a big barrier
in obtaining health care.
2.3.3 Sociodemographicdifferencesinperceivingdistancetobeabarriertohealthcare
2.3.3.1 Householdwealth
With the exception of those in the second-most wealthy quintile, greater household wealth has a protective
effect on perceiving distance to be a big barrier in accessing health care, adjusting for actual access to
health facilities and other sociodemographic characteristics. The odds of perceiving distance to be a barrier
decreases by 30% for those in the richest wealth quintile (p < 0.001; Table 2.3). This suggests greater
resources may enable women to better overcome the barrier that distance poses to obtaining health care.
20
The impact of household wealth is different for those living in urban areas (results available in the
appendix). In urban areas, there is a strong progressive decline in odds as household wealth increases
(Table 2.4). For the richest quintile, the odds of perceiving distance to be a big barrier decreases by 60%
(p< 0.001; Table 2.4). This suggests that household resources is more strongly associated with perceived
distance for urban households than rural households.
Table 2.4: Summary of multilevel logistic models for whether distance is a big barrier in accessing health
care among women living in urban Malawi
Model1 Model2 Model3 Model4
Predictors OR CI p OR CI p OR CI p OR CI p
(Intercept) 0.26 0.15–0.44 <0.001 0.27 0.16–0.47 <0.001 0.41 0.25–0.67 <0.001 0.38 0.24–0.61 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 0.75 0.60–0.94 0.013 0.75 0.60–0.94 0.013 0.75 0.60–0.94 0.014 0.75 0.60–0.94 0.014
35-49 years old 0.71 0.54–0.94 0.016 0.71 0.54–0.94 0.015 0.71 0.54–0.94 0.016 0.71 0.54–0.94 0.016
Education
(ref: Noneorprimary)
Secondary or higher 0.76 0.65–0.90 0.002 0.76 0.65–0.90 0.002 0.77 0.65–0.91 0.002 0.77 0.65–0.91 0.002
Householdwealthquintile
(ref: Poorest)
Poorer 0.88 0.70–1.10 0.267 0.88 0.70–1.10 0.262 0.87 0.69–1.09 0.234 0.87 0.69–1.09 0.224
Middle 0.77 0.61–0.98 0.033 0.77 0.61–0.98 0.032 0.76 0.60–0.97 0.028 0.76 0.60–0.96 0.023
Richer 0.66 0.52–0.85 0.001 0.66 0.52–0.85 0.001 0.65 0.51–0.84 0.001 0.65 0.51–0.83 0.001
Richest 0.39 0.29–0.52 <0.001 0.39 0.29–0.52 <0.001 0.38 0.28–0.51 <0.001 0.38 0.28–0.51 <0.001
Maritalstatus
(ref: Nevermarried)
Married or living with partner 0.88 0.69–1.12 0.295 0.88 0.69–1.12 0.297 0.88 0.69–1.12 0.287 0.88 0.69–1.12 0.282
Widowed, divorced, or separated 1.28 0.96–1.71 0.097 1.28 0.96–1.71 0.097 1.27 0.95–1.71 0.103 1.27 0.95–1.70 0.106
Hasayoungchildunder5years 0.86 0.71–1.02 0.087 0.85 0.71–1.02 0.086 0.86 0.72–1.02 0.087 0.86 0.72–1.02 0.088
Measuresofaccesstohealthfacility
m1 (Straight-line distance) 1.32 1.10–1.59 0.003
m2 (Routed distance) 1.20 1.03–1.40 0.019
m3 (Count of facilities within
10km radius) 0.99 0.96–1.01 0.192
m4 (Chance of NOT being in 5km
catchment area of a facility) 0.74 0.44–1.24 0.249
RandomEffects
σ 2
3.29 3.29 3.29 3.29
τ 00 0.55
cluster:district
0.57
cluster:district
0.56
cluster:district
0.57
cluster:district
1.02
district
1.02
district
1.17
district
1.06
district
ICC 0.32 0.32 0.34 0.33
AIC 4883.9 4887.0 4890.8 4891.1
2.3.3.2 Havingayoungchild
Another somewhat protective characteristic is having a young child (under 5 years of age). Adjusting for
all other variables, having a young child decreases the odds of perceiving distance to be a big barrier by
8% (p < 0.001; Table 2.3). Women with a young child may have more recent and/or a greater frequency
21
of interaction with the health system compared to those who do not have a young child due to pregnancy,
delivery, and raising a young child. The protective effect might be a reflection of a greater need or greater
sense of urgency for health services, and thus, a reduction in the perception of distance as a barrier. An-
other motivation of this protective effect might be that greater familiarity with engaging with the health
system reduces perceived barriers, including those related to physical access.
2.3.3.3 Age
On the other hand, greater age increases the odds of perceiving distance to be a barrier. Compared to
women in the 15-19 year age group, those 20-34 years had a 23% increase in odds and those in the oldest
age category, 25-49 years, had a 35% increase in the odds of perceiving distance to be a big barrier in ac-
cessing health care, all other characteristics held constant (Table 2.3). This is one of the characteristics that
demonstrates an opposite effect in urban areas, where greater age is increasingly protective ( p< 0.05; Ta-
ble 4). This suggests how aging processes may differ in urban and rural environments. For example, while
urban aging might be associated with the accumulation of material and social resources that makes dis-
tance increasingly less of a barrier, aging in rural areas might not have those same benefits. Instead, aging
in rural areas might be associated with the degradation of material resources, social networks and protec-
tions, making women more vulnerable and less able to access or utilize resources to overcome distance
barriers to health care.
2.3.3.4 Education
Another area where urban and rural areas are different is in how education effects perceiving distance to be
a barrier to health care. While those with secondary or higher education have no difference in perceiving
distance to be a big barrier to health care compared with those with no to primary education (Table 2.3).
In comparison, higher education is a protective factor for women in urban areas (p< 0.005; Table 2.4).
22
2.3.3.5 Maritalstatus
In both urban and rural areas, those who are married or are living with a partner have no difference in
perceiving distance to be a big barrier compared to those who were never married. However, though not
statistically significant, women in both urban and rural areas who are widowed, divorced, or separated
experience an increased odds of perceiving distance to be a big barrier in accessing health care (Table 2.4).
This increase in risk has borderline statistical significance in the case of women in rural areas and hints at
a potential relative social disadvantage of women who are widowed, divorced, or separated.
2.4 Discussion
2.4.1 Measuresofphysicalproximitytohealthcare: distanceandaccessenvironment
The most important contribution of this study is demonstrating that living farther from health facilities
and poorer access to health facilities is related to perceiving distance to be a big barrier in getting health
care. Showing how a subjective measure is related to objective measures of access and sociodemographic
context lends support to the importance and potential utility of these self-rated measures in understanding
the factors that drive or hinder healthcare utilization.
The four objective measures of physical proximity to health facilities studied here can be categorized
into two types: distance and access environment measures. Straight-line (Euclidean) and routed distances
are measures of distance, while the number of facilities within a 10-kilometer radius and the probability
of being within a five-kilometer health facility catchment area are both measures of the general access
environment. The access environment measures include an element of proximity and tries to consider
the uncertainty introduced by the displacement of the DHS cluster location. Since the outcome explicitly
centers around asking whether “distance” is a big barrier to health care, it might not be a surprise that both
straight-line and routed distances are associated with the outcome.
23
This study provides evidence for straight-line distance being an adequate measure proximity to health
services. Given the model does not meaningfully change when routed distance is provided instead, it
suggests that, in the case of rural Malawi, straight-line distances systematically underestimate actual paths
travelled and that it is an adequate substitute as a measure of geographic proximity.
The finding that the DBCAP value is associated with the outcome advances the linking of measures of
health service environment to large household surveys. Further study is needed, but the findings suggest
that this approach to describing the health care environment is as helpful a measure as distance to nearest
health facility in quantifying access to health care.
This kind of double buffer method is well-developed and widely used to measure spatial access to
healthcare and plan future services and healthcare infrastructure in data-rich settings ([47]). This study
demonstrates how it can also be adapted and applied to link large health surveys and health facility sur-
veys in the low- and middle-income country context. As with the widely-implemented two-step floating
catchment area (2SFCA) method, this double buffer method can be used in planning locations of future
health facilities to optimize the reduction of actual and perceived barriers to health care access.
2.4.2 Howfaristoofar?
Finally, this study sheds light on the distance that people consider far enough to be considered a big barrier
to health care. The mode of transport available to people will vastly change that perception and calculus,
but for rural Malawi, five kilometers can serve as a general threshold for what might be considered too far.
The average routed distance to the nearest health facility for those who said distance was not a big
barrier was 4.1km and 5.9km for those who considered distance to be a big barrier to accessing health
care. Also, the size of the buffer that resulted in the greatest distinctions between the two groups for the
probability of not living in a health facility catchment area was a 5km buffer around the DHS cluster and
a 5km catchment area for health facilities.
24
These results can be used to inform future studies on perceived barriers to health care. For example,
understanding how far people consider to be far enough to hinder seeking health care quantifies the oth-
erwise vague idea of distance tolerance, or how far people are generally willing to travel, among this rural
Malawian population. Understanding the extent of what people consider to be a service environment they
have access to can be applied to generate better measures of service environment attributes.
Finally, the count of facilities in a 10-kilometer buffer has no impact on the model and cannot be used to
distinguish between people who say distance is a big barrier and those who say it is not. The conservative
and inclusive intention of the 10-kilometer radius makes this measure too general for it to inform what a
more realistic threshold for “too far” might look like for people in rural Malawi.
2.5 Conclusion
This study demonstrates that perceiving distance to be a major barrier in seeking health care is associ-
ated with greater distance from the nearest health facility and lower physical access, even after adjusting
for sociodemographic characteristics. However, our findings also show that there is more to the access
problem than the physical distance alone. Improving health infrastructure will help to improve improve
access, but other contextual factors will be important for decision-makers to consider to make further
gains in reducing access barriers to health care in a way that also addresses sociodemographic inequities.
Further, this study suggests that measures of self-assessed barriers to health care can be utilized to better
understand health-care seeking behaviors. The close association between objective and subjective mea-
sures raise questions of the relative salience to health-related behaviors. This study also demonstrates the
potential of the DBCAP method, a new, general adaptation of the 2SFCA method, to assess spatial access to
health care accurately despite the uncertainty introduced by the displacement of DHS cluster locations to
protect respondent privacy. This method can be advanced to improve the measure of service environments
in studies linking health facility and population health data.
25
Chapter3
Self-reportedbarrierstohealthcareandtheutilizationofpregnancy
careinruralMalawi(Paper2)
3.1 Background
3.1.1 MaternalhealthinMalawi
Malawi, a small, landlocked country in south-central Africa, is has one of the highest rates of maternal
mortality in the world (Figure 3.1). While the maternal mortality ratio (MMR), the number of maternal
deaths per 100,000 live births, has been on the decline, the gains were not sufficient to reach the MMR
target in 2015 for the end of the Millennium Development Goal era ([84, 17]). With the most recent estimate
of MMR at 349 maternal deaths per 100,000 live births, an accelerated reduction is needed to meet the
Sustainable Development Goal (SDG) for MMR by the target year of 2030 ([84]). The SDG 3.1 Target for
MMR is a global average of 70 deaths per 100,000 live births, with no individual country above 140 deaths
per 100,000 live births ([28], Figure 3.2).
While the MMR is short of meeting global targets, Malawi intentionally and swiftly increased facility
deliveries through aggressive policies ([18]; Figure 3.2). Tracked alongside MMR is SDG Indicator 3.1.2, the
proportion of deliveries attended by skilled health personnel trained in delivering lifesaving interventions
during labor, delivery, and in the postpartum period ([82]). This measure includes doctors, nurses, and
midwives, and excludes traditional birth attendants. The indicator shown in Figure 3.2 is health facility
delivery, a proxy and somewhat conservative estimate for the indicator of having a skilled attendant at
birth (SAB), since all deliveries in a facility are assumed to be attended by a skilled provider and there
may be deliveries outside the health facility that are also attended by a skilled provider. Health facility
26
R u r a l D H S C l u s t e r s
D i s t r i c t s
M a l a w i
Figure 3.1: Map of Malawi (black outline) and its 28 districts (yellow with red outline) showing location of
rural clusters surveyed in the 2015-16 Malawi Demographic and Health Survey
deliveries increased from 55% in 2000 to 91% in 2015 ([50], Figure 3.2). This meets the 90% global target
put forth by the World Health Organization (WHO) and UN Population Fund (UNFPA)-led global, multi-
partner initiative for Ending Preventable Maternal Mortality (EPMM) ([21]). EPMM also recommends a
27
349
155
70
0
100
200
300
400
500
600
700
800
900
1000
2000 2005 2010 2015 2020
Maternal Mortality Ratio
(maternal deaths per 100,000 live births)
MMR
MDG 5 target
SDG 3.1 target
55
69
73
91 91
92
95 95
56
57
46
51
7
8
12
24
0
10
20
30
40
50
60
70
80
90
100
2000 2004 2010 2015
Coverage of service (%)
ANC1
Health facility delivery
ANC4+
Timely ANC initiation
Figure 3.2: Malawi trends for Maternal Mortality Rate (with 2015 target for Millennium Development
Goals and 2030 target for Sustainable Development Goals) and coverage of key interventions for a healthy
pregnancy: ANC1, ANC4+, Timely ANC initiation, health facility delivery. Source: MDHS 2000, MDHS
2004, MDHS 2010, MDHS 2015-16 and United Nations Maternal Mortality Estimation Inter-agency Group
(MMEIG), September 2019 revision (updated 2021)
90% global coverage of four or more antenatal care visits (ANC4+), with 90% of countries achieving at least
80% coverage ([21]).
In contrast to the near doubling of health facility delivery coverage from 2000 to 2015, the coverage
of ANC4+ declined from 56% to 51% in the same time period ([50], Figure 3.2). This persistent lack of
improvement in ANC4+ coverage stands out especially in the context of a consistently high rate (90% or
higher) of attending at least one ANC visit (ANC1) and an increasing trend of timely ANC initiation in
the first trimester, or first 12 weeks, of pregnancy. These contrasting trends suggest that there are persis-
tent barriers that prevent women from attending the recommended four or more visits during pregnancy
even while almost all women attend one visit and an increasing number of women initiate ANC early in
pregnancy ([31, 38]).
28
3.1.2 Indirectmaternaldeathsandantenatalcare(ANC)
Antenatal care is an important strategy to reduce indirect maternal deaths. In the 10th revision of the
International Classification of Diseases (ICD-10), the WHO categorizes maternal deaths as direct or indirect
deaths ([90]). Direct obstetric deaths are defined as “maternal deaths resulting from obstetric complications
of the pregnant state (pregnancy, labor, and the puerperium), from interventions, omissions, incorrect
treatment, or from a chain of events resulting from any of the above,” while indirect obstetric deaths are
“those resulting from previous existing disease or disease that developed during pregnancy and which was
not due to obstetric causes, but was aggravated by physiologic effects of pregnancy” ([90]).
The top causes of direct maternal mortality in sub-Saharan Africa are hemorrhage (25% of maternal
deaths), pregnancy-induced hypertension (16%), pregnancy-related sepsis (10%), and complications of un-
safe abortion (10%) ([71]). Having a skilled attendant at delivery who can provide emergency obstetric
care is an intervention with a strong evidence base for reducing these direct maternal deaths ([66, 92]).
Indirect causes, which are on the rise, but more difficult to capture, account for about 29% of maternal
deaths in sub-Saharan Africa, which is among the highest share in the regions of the world ([71, 30]).
The most common indirect causes of maternal death in this context are related to anemia, HIV/AIDS, and
malaria ([71]). Other causes of indirect maternal deaths include other sexually transmitted diseases, tu-
berculosis, poor nutrition (both undernutrition and obesity), and chronic diseases (e.g., diabetes, cardiac
disease) ([87]). The causes of direct maternal deaths have corresponding medical interventions that are
generally implemented in an urgent manner ([61, 30]). Given the scale of direct maternal deaths and pres-
ence of clear medical interventions to address their causes, there is an understandable focus on reducing
direct maternal deaths through emergency obstetric care during childbirth. In contrast, the prevention of
indirect maternal deaths requires a multifaceted and longer-term approach.
Antenatal care visits are the ideal vehicle for the screening, diagnosis, and treatment of the conditions
that result in indirect maternal deaths. The screening and treatment of HIV, malaria, anemia, and other
29
opportunistic infections and conditions are a part of the national antenatal care package, which is free
in Malawi ([48, 29]). By also screening for complicated and high-risk pregnancies, ANC also contributes
to preventing direct maternal deaths. WHO’s Focused ANC (FANC) policy, first published in 2001 and
adopted in Malawi in 2003, recommends a minimum of four ANC visits through which a package of high-
quality interventions are to be delivered to ensure the health and well-being of women during pregnancy,
childbirth, and the postnatal period ([48, 52]). Although WHO revised its ANC recommendation to a
minimum of eight visits in 2016, FANC is still the policy implemented in Malawi ([91, 29]).
Given the general health profile and demographic context of Malawi, the potential and future contri-
bution of improved ANC utilization on maternal morality reduction, especially through the reduction of
indirect maternal deaths, cannot be underestimated. In Malawi, malaria is endemic with a high risk of
transmission, HIV prevalence in the population is high, and nearly a third of reproductive-aged women,
15-49, are anemic ([81]). In addition, 41% of the female population in Malawi is under the age of 15, where
the average woman has 4.2 children ([86]). For the sake of the next generation of women who will be-
come of child-bearing age, it is imperative to better understand the factors that enable and discourage the
utilization of ANC.
3.1.3 Enablingfactorsandbarrierstoantenatalcare
Many sociodemographic characteristics are associated with attainment of adequate ANC, which is com-
prised of both timely initiation within the first trimester of pregnancy and a sufficient number of contacts
throughout the pregnancy ([56]). Studies in the low- and middle-income country context find that age ([38,
19, 59]), educational attainment([38, 95, 24, 5, 19, 59]), urban/rural place of residence ([95, 24]), household
wealth ([38, 24, 5, 19, 59]), marital status ([19]), working status ([95]), media exposure ([95]), parity ([38,
95, 19, 59]), whether the pregnancy was planned ([95]), and health insurance ([19]) are associated with
both timely ANC initiation and ANC4+.
30
In addition to sociodemographic characteristics, there are many exogenous factors not easily captured
through surveys that motivate and hinder careseeking behaviors. Several qualitative studies in Malawi and
similar contexts highlight the role of culture, community, interpersonal dynamics, and lived experiences in
health decisions. For example, women who did not bring their partners to the clinic reported being denied
access to ANC services ([61, 51, 80]). Barriers also included additional costs associated with free ANC,
including having to bring cloth wraps for the newborn or having to pay fees for authorization letters from
village heads for women with no partners ([61, 51, 26]). Another barrier was the distance to care and the
cost (money and time) associated with making the trip ([41, 39, 80]). Provider-related barriers, including
poor quality of care, lack of essential equipment, lack of providers, and unfriendly, rude treatment by
providers, are common themes of barriers articulated by women ([41, 15, 80, 26]). In addition to spouses,
the presence of other stakeholders in deciding when to obtain ANC include village elders and marriage
counselors from the husband’s side from whom women must receive advice before starting ANC ([69,
15]). Specifically in the context of Malawi, many cited the custom of hiding a pregnancy until it is visible
to all (usually in the fourth or fifth month) to avoid being bewitched ([69, 51, 15]). While HIV screening
is voluntary by policy, women also noted compulsory HIV testing and HIV stigma as barriers to ANC
utilization for themselves as well as a reason husbands refuse to accompany them to ANC ([15, 80]).
As a largely preventive service, ANC suffers the challenge of not being perceived as important or
valuable, relative to urgent care or emergency services that respond to acute and obvious need. As a
preventive service that requires routine, repeat visits, ANC also faces the burden of repeatedly having to
prove its value to users. In this context of preventive care, the influence of barriers on the decision to seek
ANC may disproportionately outweigh motivating factors. High rates of attending one ANC visit alongside
low rates of ANC4+ suggest the presence of barriers that discourage users from continued utilization of
ANC after the first visit (Figure 3.2).
31
3.1.4 Self-reportedbarrierstocare
The Demographic and Health Survey (DHS) offers a unique opportunity to study barriers to care. The
DHS collects information on self-reported barriers to care that span personal factors, provider factors, and
distance to health facilities. While these measures are subjective, they are interesting in that they reflect the
impact of potential barriers in light of individual context. In this way, self-reported barriers may be more
accurate in assessing the impact of a potential barrier. In fact, self-reported barriers in multiple domains
(i.e. patient, provider, and system-level factors) have been associated with delayed and forgone health care
([4]). Studying the relationship of these self-reported barriers and ANC utilization may provide a better
understanding of the impact of barriers and suggest ways forward on how to reduce certain barriers to
improve ANC4+.
3.2 Methods
3.2.1 Data
This study uses data from the 2015-16 Malawi Demographic and Health Survey (MDHS). It is a cross
sectional survey providing a nationally-representative portrait of sociodemographic characteristics and
population health, including indicators specific to reproductive, maternal, and child health ([50]). The
MDHS employs a two-stage sampling scheme, which selects survey clusters in the first stage and then
households within clusters in the second stage. This study is comprised of the 11,082 reproductive-aged
women (15-49 years) in 677 rural survey clusters across all 28 districts in Malawi who had a childbirth
event in the five years preceding the survey.
32
3.2.2 Measures
3.2.2.1 Outcomevariables
Adequate ANC is comprised of both timely initiation within the first trimester of pregnancy and sufficient
number of contacts throughout the pregnancy ([56]). Information on the timing of initiation and number
of ANC visits during the most recent pregnancy in the five years prior to the survey were collected in the
MDHS and recoded as dichotomous outcome measures for the analysis. The outcome timely ANC initiation
refers to whether or not ANC was started in the first 12 weeks of pregnancy. The outcome ANC4+ refers
to whether or not four or more ANC visits were obtained. Two individuals from the analysis who were
missing responses for both outcomes were excluded from the analysis. While most women had data for
both outcomes (n = 10,825), individuals who had data for either ANC initiation (n = 37) or the number
of ANC visits during pregnancy (n = 218) were retained for the analysis.
3.2.2.2 Explanatoryvariables
The MDHS asks women about series of seven factors that might be major barriers in seeking medical advice
or treatment for themselves when they are sick. These include: (1) distance to the health facility; (2) getting
permission to go to the doctor; (3) getting money needed for advice or treatment; (4) not wanting to go
alone; (5) concern that there may not be a female provider; (6) concern that there may not be any health
provider; (7) concern that there may be no drugs available ([50]). For this study, these seven factors have
been grouped thematically: distance, personal barriers (getting permission, getting money, not wanting to
go alone), and provider barriers (concern that there may not be a female provider, that there may not be any
provider, no drugs available). Grouping these factors allows for exploring the possibility of a cumulative
effect from client and provider sides that is not easily detected from considering each factor alone. The
personal and provider barrier groups, each with a total of three barriers, are treated as ordinal categorical
33
variables, with it being possible that a women experiences none of the barriers, one of the barriers, two
barriers, or all three barriers.
The study also accounts for the following sociodemographic variables: age at the time of most recent
child birth, educational attainment of the woman, household wealth quintile (among rural households),
parity, martial status, working status, whether the most recent pregnancy was planned, and the frequency
of media exposure (any of the following: radio, tv, newspaper, and magazine).
3.2.3 Analysis
First, descriptive statistics were generated for the explanatory variables by each of the two outcomes:
timely ANC initiation and ANC4+. Then, multilevel logistic models were implemented for the two out-
comes to appropriately account for the multi-stage sampling scheme of the MDHS where individuals were
nested in survey clusters, which were in turn nested in districts. Two models were constructed for the
ANC4+ outcome, one with and one without timely ANC initiation as an explanatory variable. The random
intercepts (for districts) were extracted from the models to demonstrate district-level differences. Analysis
was conducted in R (version 4.1.3) and the models were implemented with the glmer function in the lme4
package.
3.3 Results
3.3.1 Descriptivestatistics
Table 3.1 shows that only about a quarter of rural women initiate ANC in the first trimester. Most women
initiate ANC in the second trimester (Table 3.1; Figure 3.3). Almost 40% of women who achieve ANC4+
start ANC in the first trimester of pregnancy, compared to about 12% for women who do not achieve
ANC4+ (Table 3.1). Figure 3.2 shows that initiating ANC earlier is generally associated with a greater
number of total ANC contacts during pregnancy, across all age categories. Fewer than two percent of
34
rural women in this sample forgo ANC altogether, demonstrating that the vast majority of women receive
at least one ANC during pregnancy (Table 3.1). About 20% attend five or more ANC visits, though less
than two percent attend eight or more visits, which is the most current WHO recommendation for ANC
frequency ([91]).
15−19 years 20−34 years 35−49 years
ANC initiated FIRST trimester ANC initiated SECOND trimester ANC initiated THIRD trimester
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
0
500
1000
1500
2000
0
500
1000
1500
2000
0
500
1000
1500
2000
Number of ANC visits
Number of women
Figure 3.3: Histograms showing number of ANC visits by age group and timing of ANC initiation. Dashed
line shows median in each age-timing group. Average number of ANC visits decreases as initiation of ANC
is delayed across all age groups.
3.3.2 Multilevellogisticmodels
3.3.2.1 TimelyANCinitiation
Women in the richest wealth quintile have a 25% increased odds of timely ANC initiation compared to those
in the poorest wealth quintile (p < 0.01, Table 3.2). Compared to women who have never been married,
women who are married or living with a partner and those who are widowed, divorced, or separated
have a 58% and 68% increase in the odds of timely ANC initiation, respectively (p < 0.005, Table 3.2).
Compared to those with no media exposure, those with at least one instance of weekly media exposure
35
had a 17% increase in odds of timely ANC initiation (p< 0.005, Table 3.2). Women whose pregnancies were
not planned had a 20% decrease in odds of timely ANC initiation compared to those whose pregnancies
were planned (p < 0.001, Table 3.2). None of the self-reported barriers were associated with the timely
initiation of ANC. Though there was a progressively greater decrease in odds of timely ANC initiation
with an increasing number of provider barriers, it was not statistically significant (Table 3.2).
3.3.2.2 ANC4+
In the ANC4+ model without timely ANC initiation as an explanatory variable, marital status, current
working status, and having an unplanned pregnancy were significantly associated with the outcome (Ta-
ble 3.2). Greater age, education, and household wealth were also associated with an increased odds of
achieving ANC4+ (Table 3.2). However, after adding timely ANC initiation in the model, only age and un-
planned pregnancy remained statistically significant. Secondary or higher educational attainment showed
marginal significance (aOR=1.18, p = 0.054, Table 3.2).
Compared to those 15-19 years old, there was an 18% and 35% increase in odds of ANC4+ for those
20-34 years old and 35-49 years old, respectively (p < 0.05, Table 3.2). Adding in timely ANC initiation
only slightly muted the impact of an unplanned pregnancy (from aOR=0.80 to aOR=0.85), which decreased
the odds of ANC4+ by 15% (p < 0.001, Table 3.2). Like for timely ANC initiation, there is an increased
odds in ANC4+ among women who are married/living with a partner or widowed, divorced, or separated,
though the association is not statistically significant.
An increasing number of perceived provider barriers is associated with a lower odds of ANC4+ in both
models with and without timely ANC initiation. Even after accounting for timely ANC initiation, which
increases the odds of ANC4+ nearly five-fold (aOR=4.81, p < 0.001, Table 3.2), perceiving two provider
barriers reduces the odds of ANC4+ by 11% and perceiving three barriers reduces the odds of ANC4+ by
14% (p< 0.05, Table 3.2).
36
Table 3.1: Background characteristics and explanatory variables among women in rural DHS clusters who
had a live delivery in the five years prior to the survey (n=11,082), unweighted, by whether they initiated
ANC in the first trimester and whether they attained ANC4+
TimelyANCinitiation?
FourormoreANCvisits
(ANC4+)
No
(n=8,057)
Yes
(n=2,805)
No
(n=5,567)
Yes
(n=5,476)
Overall
(n=11,080)
%(n) %(n) %(n) %(n) %(n)
Age
15-19 9.1% (731) 7.3% (205) 9.9% (552) 7.4% (407) 8.7% (961)
20-34 70.2% (5657) 72.4% (2032) 70.4% (3921) 71.0% (3888) 70.7% (7836)
35-49 20.7% (1669) 20.2% (568) 19.7% (1094) 21.6% (1181) 20.6% (2283)
Education
No education or primary school 13.6% (1099) 12.1% (338) 14.2% (792) 12.4% (681) 13.3% (1478)
Primary school 69.3% (5584) 69.0% (1936) 69.6% (3877) 69.0% (3778) 69.3% (7677)
Secondary or higher 17.1% (1374) 18.9% (531) 16.1% (898) 18.6% (1017) 17.4% (1925)
Householdwealthquintile(rural)
Poorest 22.8% (1833) 20.1% (564) 23.2% (1289) 21.3% (1168) 22.2% (2463)
Poorer 21.1% (1700) 21.1% (592) 21.5% (1195) 20.7% (1132) 21.1% (2336)
Middle 20.1% (1618) 18.6% (522) 20.0% (1111) 19.5% (1069) 19.7% (2183)
Richer 19.0% (1533) 19.7% (552) 18.9% (1052) 19.2% (1054) 19.1% (2113)
Richest 17.0% (1373) 20.5% (575) 16.5% (920) 19.2% (1053) 17.9% (1985)
Parity
Median [Min, Max] 3 [1, 14] 3 [1, 13] 3 [1, 13] 3 [1, 14] 3 [1, 14]
Maritalstatus
Never in union 4.0% (326) 2.6% (72) 4.5% (252) 3.0% (164) 3.8% (417)
Married or living with partner 82.3% (6634) 83.6% (2346) 82.1% (4571) 82.9% (4542) 82.5% (9145)
Widowed, divorced, or separated 13.6% (1097) 13.8% (387) 13.4% (744) 14.1% (770) 13.7% (1518)
Workingstatus
Not currently working 32.5% (2621) 31.9% (896) 34.4% (1916) 31.0% (1697) 32.8% (3629)
Currently working 67.5% (5436) 68.1% (1909) 65.6% (3651) 69.0% (3779) 67.2% (7451)
Unplannedpregnancystatus
Pregnancy is planned 54.3% (4377) 59.9% (1681) 52.9% (2943) 58.0% (3178) 55.5% (6149)
Pregnancy is not planned 45.7% (3680) 40.1% (1124) 47.1% (2624) 42.0% (2298) 44.5% (4931)
Frequencyofmediaexposure
(radio/tv/newspaper/magazine)
None 51.0% (4110) 47.0% (1319) 51.6% (2874) 48.8% (2674) 50.2% (5567)
Less than once a week 18.0% (1449) 16.8% (472) 18.0% (1000) 17.4% (952) 17.7% (1958)
At least once a week 31.0% (2498) 36.1% (1014) 30.4% (1693) 33.8% (1850) 32.1% (3555)
Distanceisabigbarriertohealthcare
No 39.5% (3184) 42.0% (1178) 39.1% (2179) 41.1% (2253) 40.2% (4451)
Yes 60.5% (4873) 58.0% (1627) 60.9% (3388) 58.9% (3223) 59.8% (6629)
Personalbarrierstohealthcare
(gettingpermission,gettingmoney,notwantingtogoalone)
0 barriers 37.7% (3036) 41.0% (1151) 37.1% (2067) 39.6% (2166) 38.4% (4253)
1 barrier 33.9% (2735) 31.8% (892) 33.7% (1878) 32.9% (1804) 33.3% (3687)
2 barriers 19.4% (1564) 18.7% (524) 19.9% (1107) 18.8% (1028) 19.3% (2140)
All 3 are barriers 9.0% (722) 8.5% (238) 9.3% (515) 8.7% (478) 9.0% (1000)
Provider-relatedbarrierstohealthcare
(concernthattheremaynotbeafemaleprovider,anyprovider,nodrugsavailable)
0 barriers 28.4% (2290) 31.0% (870) 27.9% (1555) 30.0% (1645) 29.0% (3210)
1 barrier 18.3% (1476) 18.7% (524) 17.9% (996) 18.7% (1024) 18.3% (2029)
2 barriers 32.1% (2584) 30.8% (864) 31.9% (1776) 31.5% (1723) 31.7% (3509)
All 3 are barriers 21.2% (1707) 19.5% (547) 22.3% (1240) 19.8% (1084) 21.0% (2332)
TimingofANCinitiation
In first trimester 0% (0) 100% (2805) 11.5% (642) 39.1% (2141) 25.3% (2805)
In second trimester 91.6% (7379) 0% (0) 73.3% (4080) 60.0% (3284) 66.6% (7379)
In third trimester 8.4% (678) 0% (0) 11.3% (630) 0.9% (48) 6.1% (678)
Missing 0% (0) 0% (0) 3.9% (215) 0.1% (3) 2.0% (218)
NumberofANCvisits
0 0% (0) 0% (0) 3.6% (201) 0% (0) 1.8% (201)
1 2.6% (213) 0.6% (17) 4.1% (230) 0% (0) 2.1% (230)
2 13.0% (1048) 2.4% (67) 20.1% (1119) 0% (0) 10.1% (1119)
3 42.8% (3449) 19.9% (558) 72.2% (4017) 0% (0) 36.2% (4017)
4 27.0% (2175) 37.7% (1057) 0% (0) 59.1% (3234) 29.2% (3234)
5 to 7 13.5% (1089) 35.9% (1007) 0% (0) 38.3% (2097) 18.9% (2097)
8 or more 0.8% (68) 2.7% (77) 0% (0) 2.6% (145) 1.3% (145)
Missing 0.2% (15) 0.8% (22) 0% (0) 0% (0) 0.4% (37)
37
Table 3.2: Summary of multilevel models for timely ANC initiation and ANC4+ among rural Malawian
women
TimelyANCinitiation
ANC4+
(noinitiation)
ANC4+
(withinitiation)
Predictors aOR CI p aOR CI p aOR CI p
(Intercept) 0.19 0.13–0.27 <0.001 0.59 0.44–0.80 0.001 0.50 0.36–0.68 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 1.14 0.95–1.37 0.157 1.20 1.03–1.40 0.020 1.18 1.01–1.39 0.043
35-49 years old 1.13 0.89–1.45 0.316 1.37 1.11–1.68 0.003 1.35 1.08–1.68 0.008
Education
(ref: None)
Primary 1.02 0.88–1.18 0.805 1.08 0.96–1.23 0.214 1.08 0.94–1.23 0.275
Secondary or higher 1.03 0.85–1.24 0.799 1.21 1.02–1.42 0.024 1.18 1.00–1.41 0.054
Householdwealthquintile
(ref: Poorest)
Poorer 1.13 0.99–1.30 0.078 1.03 0.92–1.17 0.572 0.99 0.87–1.12 0.893
Middle 1.03 0.89–1.19 0.729 1.05 0.93–1.19 0.400 1.04 0.91–1.18 0.567
Richer 1.08 0.93–1.25 0.309 1.06 0.93–1.20 0.375 1.02 0.89–1.17 0.780
Richest 1.25 1.06–1.47 0.008 1.16 1.01–1.34 0.041 1.11 0.95–1.29 0.184
Parity 1.00 0.97–1.03 0.826 1.00 0.98–1.03 0.745 1.01 0.98–1.04 0.530
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.58 1.19–2.10 0.002 1.32 1.06–1.65 0.013 1.15 0.91–1.46 0.231
Widowed, divorced, or separated 1.68 1.24–2.28 0.001 1.42 1.12–1.81 0.004 1.23 0.95–1.59 0.108
Currentlyworking 1.02 0.92–1.13 0.652 1.11 1.02–1.21 0.020 1.08 0.99–1.19 0.089
Pregnancywasnotplanned 0.80 0.73–0.88 <0.001 0.80 0.74–0.87 <0.001 0.85 0.78–0.92 <0.001
Media(newspaper/magazine,radio,tv)exposure
(ref: noexposure)
Any media exposure less than once a week 1.03 0.90–1.17 0.687 0.99 0.89–1.11 0.923 0.98 0.87–1.10 0.709
Any media exposure at least once a week 1.17 1.05–1.31 0.004 1.07 0.98–1.18 0.143 1.02 0.92–1.13 0.703
Perceivedtobebigbarrierstohealthcare
Distance
(ref: nodistancebarrier)
Distance is a barrier 0.96 0.86–1.07 0.504 0.99 0.91–1.09 0.910 0.98 0.89–1.08 0.698
Personalbarriers(gettingpermission,gettingmoney,notwantingtogoalone)
(ref: nopersonalbarriers)
1 personal barrier (any one) 0.95 0.85–1.07 0.424 0.96 0.86–1.06 0.385 0.96 0.86–1.07 0.486
2 personal barriers (any two) 1.02 0.87–1.18 0.845 0.92 0.81–1.04 0.178 0.91 0.80–1.05 0.197
3 personal barriers (all three) 1.03 0.84–1.26 0.771 0.97 0.82–1.16 0.750 1.01 0.84–1.21 0.954
Provider-relatedbarriers(concernthattheremaynotbeafemaleprovider,anyprovider,nodrugsavailable)
(ref: noprovider-relatedbarriers)
1 provider-related barrier (any one) 0.99 0.86–1.14 0.925 0.98 0.87–1.11 0.799 0.98 0.86–1.11 0.726
2 provider-related barriers (any two) 0.93 0.81–1.05 0.236 0.88 0.79–0.98 0.021 0.89 0.79–1.00 0.048
3 provider-related barriers (all three) 0.90 0.77–1.05 0.177 0.83 0.72–0.95 0.006 0.86 0.74–0.99 0.036
TimelyinitiationofANCinfirsttrimester 4.81 4.34–5.33 <0.001
RandomEffects
σ 2
3.29 3.29 3.29
τ 00 0.21
cluster:district
0.09
cluster:district
0.09
cluster:district
0.11
district
0.08
district
0.08
district
ICC 0.09 0.05 0.05
AIC 12138.5 15069.5 13769.6
38
3.3.2.3 District-leveldifferences
Figure 3.4 shows the random intercepts for the 28 districts of Malawi, as odds ratios with their confidence
intervals from the models for timely ANC initiation (Figure 3.4, left panel) and ANC4+ (with timely ANC,
Figure 3.4, right panel). The odds ratios by district are mapped in Figure 3.5. The heterogeneity in the
intercepts suggest the presence of district-level differences. Living in Likoma, Chirazdzulu, Phalombe,
Dowa, and Ntichisi districts significantly increases the odds of timely initiation of ANC, whereas living in
Lilongwe, Mchinji, and Karonga significantly decreases the odds of initiating ANC in the first trimester
of pregnancy (p < 0.05 for all). The odds of achieving ANC4+ are significantly improved for residents
of Phalombe, Balaka, Likoma, and Nkhata Bay districts, whereas they are significantly reduced for those
living in Salima, Karonga, Rumphi, Chikwawa, and Mwanza districts(p< 0.05 for all).
While there are positive (Phalombe and Likoma) and negative (Karonga) outliers where the odds of
both outcomes is either improved or decreased, there are also districts that show discrepancies, where
the odds of one outcome are improved while the odds of the other outcome are reduced. The districts in
Figure 3.4 are sorted by the odds ratio with a line connecting each district to visually track the relative
position of each district in both models. Black lines connect districts in which the odds of timely ANC
initiation were improved, but the odds of ANC4+ were reduced, along with a more than 10-place loss in
relative position. Green lines connect districts experiencing the opposite, where the odds of timely ANC
initiation were reduced, but the odds of ANC4+ were improved, with a 10 or more place improvement in
relative position. In addition to the districts that are positive and negative outliers in the more traditional
sense, the further study of these “black” and “green” districts that are outliers because of their discrepant
impact on two related outcomes may also lead to greater insight about the set of factors that motivate the
initiation of ANC and the completing of four or more visits.
39
1.08
0.84
0.83
1.77
1.11
0.93
1.50
0.57
1.03
2.06
0.66
1.00
0.84
0.61
1.04
1.35
1.09
1.16
0.83
1.04
0.87
0.84
1.39
1.50
0.94
1.05
1.21
0.88
Karonga
Mchinji
Lilongwe
Nkhatabay
Chikwawa
Blantyre
Mangochi
Ntcheu
Nsanje
Zomba
Dedza
Rumphi
Machinga
Kasungu
Mulanje
Nkhotakota
Salima
Balaka
Mzimba
Chitipa
Neno
Thyolo
Mwanza
Ntchisi
Dowa
Phalombe
Chiradzulu
Likoma
0.1 0.2 0.5 1 2 5 10
Random effects of district (Intercept)
1.44
0.89
0.67
0.83
0.84
1.16
1.22
0.76
1.17
1.42
1.19
1.11
1.03
1.22
0.83
0.55
1.20
1.22
1.38
0.85
0.89
1.01
1.19
1.48
0.75
0.78
0.83
0.96
Mwanza
Chikwawa
Rumphi
Karonga
Salima
Mulanje
Thyolo
Chiradzulu
Chitipa
Nkhotakota
Blantyre
Nsanje
Zomba
Ntcheu
Mangochi
Machinga
Dedza
Kasungu
Lilongwe
Ntchisi
Mzimba
Dowa
Neno
Mchinji
Nkhatabay
Likoma
Balaka
Phalombe
0.1 0.2 0.5 1 2 5 10
Random effects of district (Intercept)
ANC timing ANC4+ (with timing)
Figure 3.4: Intercepts for districts for timely ANC initiation and ANC4+ (with ANC initiation) models.
Random effects are represented as odds ratios (blue represents positive effect and red indicates negative
effect) and their 95% confidence intervals. Lines between the left and right panels link same district. Black
circles/lines represent districts that had higher odds of timely ANC initiation but reduced odds of ANC4+
and a drop in relative drop in ranking of at least ten spots. Green circles/lines indicate districts that had
reduced odds of timely ANC initiation, but greater odds of ANC4+ accompanied by a relative improvement
in relative ranking of at least ten spots. All other districts are yellow.
40
T i m e l y A N C
D i s t r i c t i n t e r c e p t s ( O R )
0 . 5 7 3 - 0 . 8 3 6
0 . 8 3 6 - 0 . 9 3 4
0 . 9 3 4 - 1 . 0 3 9
1 . 0 3 9 - 1 . 2 1
1 . 2 1 - 1 . 7 6 8
A N C 4 +
D i s t r i c t i n t e r c e p t s ( O R )
0 . 5 4 9 - 0 . 8 2 5
0 . 8 2 5 - 0 . 8 8 9
0 . 8 8 9 - 1 . 1 0 8
1 . 1 0 8 - 1 . 2 0 4
1 . 2 0 4 - 1 . 4 7 9
Figure 3.5: Intercepts for districts for timely ANC initiation and ANC4+ (with ANC initiation) models
among rural women. Random intercepts are represented as odds ratios. The color categories do not repre-
sent fixed OR ranges; the quantile method was used to classify a similar number of districts to each color
category.
3.4 Discussion
The study findings suggest there is work to be done on both sides of supply and demand to improve preg-
nancy care. The findings suggest there may be different drivers and barriers to starting and maintaining
ANC visits. This is consistent with the finding among marginalized women in high-income countries that
while the decision to initiate ANC visits was most influenced by late pregnancy recognition and subse-
quent denial or acceptance, continued ANC visits depended more on a balance of perceived gains and
losses ([20]). While it is important to understand these two ANC outcomes individually, their relatedness
makes any differences we observe important to also understand. In their differences may lie a richer story
41
about what motivates women to seek adequate pregnancy care and hints about how to address the barriers
that hinder health-seeking behaviors.
3.4.1 Age,education,andworkingstatus
Older age, secondary and higher education, and current working status were associated with significantly
greater odds of ANC4+ when not accounting for timely ANC initiation. When accounting for timely ANC
initiation, only age remained statistically significant, with the impact of education and working status
reduced to marginally significant. In contrast, none of these factors were associated with timely ANC
initiation.
Greater age, higher education, and working status might indicate greater established independence,
autonomy, and empowerment enabling women to sustain multiple ANC visits to achieve ANC4+. The
fact that age has such a clear impact on ANC4+, but that parity, a proxy for maternal experience, lacks an
association with the outcome suggests there is a mechanism by which older age promotes ANC utilization
independent of the increase in maternal/general health system experience that comes with age. Overall,
these findings highlight the relative vulnerability of pregnancy among adolescents.
3.4.2 Maritalstatus
The single factor associated with the greatest increase in the odds of timely ANC initiation is marital status.
Compared to never married women, those married or living with a partner have an almost 60% increase in
the odds of timely ANC initiation. Those who were previously partnered (widowed, divorced, or separated)
have the same, if not greater, benefits as those who are currently partnered. When accounting for timely
ANC initiation in modeling ANC4+, marital status is not a significant predictor. Never married status as a
significant barrier to timely ANC initiation is consistent with reports that husband permission or presence
42
is an informal prerequisite for ANC care ([1, 51]). Single women may also incur additional costs (both
monetary and social) to get authorization letters from village heads to seek ANC without a partner ([51]).
This may suggest the need to carefully consider the impact, unanticipated consequences, and nuances
of policies that emphasize partner/stakeholder (e.g. mothers-in-law) involvement as a solution to im-
proving maternal health care utilization and outcomes. In other contexts, the well-intentioned initiative
of male partner involvement has been a barrier for those who are single and for those whose partners
are uncooperative ([65]). There may be a need for greater flexibility and nuance in recommendations
to encourage women to attend care with or without their partner and messaging around empowerment
and right-to-access independent of partnered status. Those in the never married category might also be
younger women, who already bear the burden of poor access to reproductive and maternal health services
due to age-related biases and social stigma ([72]).
In contrast to timely ANC initiation, marital status does not have the same impact on ANC4+. It is a
significant predictor in the model for ANC4+ without timely ANC initiation, but both the size of the effect
and its statistical significance are diminished when timely ANC initiation is added to the model, reinforcing
the idea with marital status has more impact on starting ANC than sustaining subsequent visits.
3.4.3 Unplannedpregnancy
Unplanned pregnancy is the only factor that significantly reduced the odds of both timely ANC initiation
and ANC4+. If the pregnancy is not expected there may be a delay in recognizing the pregnancy and the
subsequent period of denial or acceptance may take a longer time ([20, 23, 72]).
Since unexpected pregnancy remains a significant barrier to ANC4+ even after accounting for timely
ANC initiation, this suggests that the status of an unplanned pregnancy has an impact on ANC4+ that is
separate from the impact it has on the outcome as a factor that is also associated with timely ANC initiation.
Given the evidence that unintended pregnancies are more likely to occur among relatively vulnerable
43
demographic groups (e.g., unmarried women, those with four or more living children, adolescents and
women under 20), it will be doubly important to target interventions among these vulnerable groups to
improve ANC4+ ([32, 42]).
3.4.4 Self-reportedbarrierstohealth
The most important finding in this study is that even after accounting for timely initiation of ANC in
the first trimester of pregnancy, which increases the odds of ANC4+ nearly five fold, women who report
experiencing two or more provider barriers to care have a significantly reduced odds of achieving ANC4+.
This finding contributes the evidence that users who perceive there to be problems with health system
readiness have reduced utilization of health care. These provider barriers are not significantly associated
with timely ANC initiation, but like for ANC4+, a dose-dependent effect size is observed, with a greater
reduction in odds of timely ANC initiation as the number of provider barriers increases.
This study does not have the relevant data to determine whether the presence of provider barriers
indicates an insufficient package of ANC interventions, poor delivery, both, or neither. This is something
future studies can investigate through linking the DHS to health facility surveys, such as the Malawi
Service Provision Assessment ([54]). However, the three specific items that make up the provider barriers
score, including the concern that there may not be a female provider, concern that there may not be any
provider, and concern that there may be no drugs available, indicate issues with health system readiness
to deliver care.
Poor health facility quality, as measured by scores for facility infrastructure and service delivery quality
have been associated with reduced utilization of primary care services (antenatal care, postnatal care, and
vaccination) in Haiti ([27]). In the Malawi context, health facility quality has been associated with better
newborn care ([14]) as well as with a reduced newborn mortality rate ([45]). This supports the idea that
44
health facility readiness impacts both the ability to deliver quality care but also influences the user-side
decision to utilize services.
It is important to recognize how poor health facility readiness and infrastructure are related to poor
service quality and user experience. Malawi has endured a prolonged and severe shortage of health work-
ers, with only 1.48 health workers per 1,000 population, which falls far short of the WHO-recommended
minimum density of 4.45 doctors, nurses, and midwives per 1000 population ([29, 28]). Among other rea-
sons, studies indicate that staff shortages and lack of basic equipment, supplies and essential medicines
fail to create enabling environments for health providers, which in turn leads to unmotivated staff, lower
quality of care, and workforce attrition, which feeds into a negative feedback chain ([16, 9, 53]). Berman
and colleagues (2021) identify housing and facility-level improvements (e.g., reliable electricity, equipment,
drugs and supplies) as having the greatest impact on rural job choice among nurse midwives even over
salary, supportive management, and promotion potential, which are also important issues in health work-
force retention ([16]). The solution to rude provider attitudes and poor user experience of health services
must go beyond asking medical workers to stop being rude to patients as a Deputy Minister of Health was
known to do in response to a nurses strike ([62]).
Rude providers cannot be reduced to personal attitude problems that require introspection to correct.
If rude attitudes and lack of motivation is viewed as a product of an overworked, understaffed workforce
in an environment that lacks basic materials and is insufficiently supported, poor patient experience is the
result of a workforce that is neither empowered nor enabled to deliver their services effectively. The clear
and concrete solution is to improve health facility infrastructure, supplies, and increase the workforce.
These basic, essential needs feel overlooked in a time where improving the quality of maternal health
services seems to refer more to the technical skillsets of providers to provide specific interventions. The
qualitative studies on barriers to pregnancy care also indicate a need for greater professional competencies
45
on patient privacy, confidentiality, and understanding how provider-held cultural biases influences their
provision of care ([69, 72]).
The literature on the impact of distance to health services on the utilization of ANC in the low- and
middle-income country context is mixed, with some studies finding that increased distance is associated
with reduced ANC ([78, 76]) and others suggesting no association ([8, 43]). This study among rural women
in Malawi finds that perceiving distance as a major barrier to health care is not associated with either
timely ANC initiation or four or more visits. In contrast with the studies ([61, 51, 80, 26]) that suggest
personal barriers (financial, not empowered to make own health decisions, not wanting to go alone, etc.)
discourage pregnancy care utilization, this study finds no significant impact of self-reported personal bar-
riers and ANC4+. It is a limitation of this study that the DHS asks about these barriers as they pertain
to the attainment of general health care for self rather than health care during pregnancy. The barrier
measures derived from the DHS might underestimate barriers that are specific and pronounced to women
seeking health care during pregnancy. The lack of association might reflect a reduced specificity in the
measurement rather than a substantive issue.
3.5 Conclusion
This study demonstrates differences in the major barriers for timely ANC initiation and ANC4+. Timely
ANC initiation is especially poor for never married, single women. For ANC4+, women who perceive
poor facility readiness (availability of providers, drugs, and supplies) have a significantly reduced odds of
achieving ANC4+.
The main contribution of this study is evidence that women’s perception of health system readiness is
associated with the actual utilization of health services. The perceived barriers questions on the DHS can be
used in future studies to enhance understanding individual health behaviors. Since perceived barriers are
46
shown to pattern health care usage addressing these barriers is critical to ultimately improve the utilization
of pregnancy services to reduce maternal mortality in Malawi.
47
Chapter4
Characterizingperceivedproviderbarrierstohealthcareandits
relationshipwithhealthproviderdensity(Paper3)
4.1 Introduction
The increasing availability of geographically referenced health data has enabled the study of how spatial
factors impact health. Additionally, the availability of population health surveys and data on health facili-
ties has allowed for a richer perspective on how both supply and demand side factors play into health de-
cisions and health status. These studies that use both population health surveys and health system/facility
datasets require a linking procedure to appropriately assign health system attributes to specific popula-
tions.
The most common and simple method of linking population health surveys to health facility data is to
identify the nearest facility to a population measured by a straight-line distance. A more accurate way of
assessing proximity by distance is to account for the existing road network and to consider which facility
is nearest by routed distance. However, some characteristics of these datasets introduce uncertainty in
the geospatial attributes. In the case of Demographic and Health Surveys (DHS), which are widely and
regularly implemented in low- and middle-income countries, the location of the sampled survey cluster is
displaced before being made publicly available to protect the privacy of respondents. Also, health facility
data are often collected in sample surveys but only a subset of health facilities are included. In cases where
the health facility survey is a sample survey and not a census of health facilities, linking health facility
attributes to population health surveys is discouraged due to the heightened risk of misclassification ([73])
48
(e.g., erroneously matching people to health facilities where they do not get care). Misclassification hinders
the ability to accurately assess the impact of health system factors on health outcomes.
The Service Provision Assessment (SPA), implemented by the MEASURE DHS project that also imple-
ments the DHS, is a standard and representative source of data for health facilities in low- and middle-
income countries (LMICs) ([54]). SPAs are generally implemented as sample surveys, which leaves many
countries unable to adequately link health system data to population health surveys for a supply-side
perspective to individual and population health behaviors ([12]).
Instead, there is a series of questions on the DHS that ask women about factors perceived to be major
barriers in seeking medical advice or treatment. These questions have the potential to shed light on barriers
to care, including those related to the provider. The 2015-16 Malawi DHS specifically asks about seven
factors that might be major barriers including: (1) distance to the health facility; (2) getting permission
to go to the doctor; (3) getting money needed for advice or treatment; (4) not wanting to go alone; (5)
concern that there may not be a female provider; (6) concern that there may not be any health provider;
(7) concern that there may be no drugs available ([50]). These barriers were grouped into three domains:
demand-side factors (getting permission, getting money, not wanting to go alone), provider-side factors
(concern that there may not be a female provider, no provider, no drugs available), and distance, which
is a structural barrier that can be said to lie in both client and provider domains. Grouping these factors
allows for exploring the possibility of a cumulative effect from client and provider sides that is not easily
detected from considering each factor alone.
In lieu of, or as a complement to health facility surveys, these questions provide a unique perspective
on how potential barriers (such as distance) are internalized and perceived as barriers (or not) depending
on personal context. These questions can contribute a unique perspective on health outcomes and behav-
iors beyond what standard socioeconomic and demographic factors such as wealth and age can provide.
However, these specific factors are not well understood and thus, underutilized in analyses of health. To
49
this end, this study seeks to characterize self-reported provider-side barriers by linking the most recent
SPA and DHS surveys in Malawi. Malawi is a good setting for this work especially because the 2013-14
facility assessment was a census of facilities, providing a complete perspective of health system resources
at that time.
Provider density, or the number of health providers per 1,000 people, will be used as an objective mea-
sure in the characterization of provider barriers. Provider availability represents two of the three compo-
nents of the grouped provider-side barriers and has a global standard to benchmark and give context to the
measure: the WHO-recommended minimum density of skilled providers (doctors, nurses, and midwives)
is 4.45 per 1,000 people ([91, 82]). This study generates four variations of the provider density measure to
evaluate how well each measure compares to the self-reported provider barriers.
The first variation in the measure of provider density is the type of provider: skilled and all types.
To be consistent with the WHO definition of "skilled" provider, this category includes doctors, nurses,
and midwives ([91]). However, cadres outside this strict definition of skilled provider also provide health
services so an all provider density is also generated ([3]). The second variation is spatial. Every DHS
cluster will be linked to the provider density of the facility that is nearest according to a straight-line
(Euclidean) distance ([89]). A shortcoming of this simple linkage is that it does not account for other
facilities that might also be in relative close proximity. Also, since the facility provider density accounts
for a 5-kilometer radius catchment area around the facility, it does not necessarily account for the specific
population in the DHS cluster, unless the location of the cluster and health facility are the same.
One method of accounting for multiple facilities in proximity of a DHS cluster is to obtain an average
of health facility attributes of all the facilities that are within a certain range ([89, 12]). However, instead of
giving each facility an equal weight of influence on the service environment of nearby DHS clusters ([89,
12]), this study proposes an innovative measure that accounts for the influence of multiple facilities in a
50
more nuanced way: accounting for their proximity to the cluster and weighted by the population that is
covered in their respective catchment areas.
This study tests a more local (to the DHS survey cluster) measure of provider availability – one that
both accounts for the presence of multiple proximal facilities and the displacement of the DHS cluster
location by accounting for a 5-kilometer buffer around the survey cluster location. In this way, there are
four measures of provider density: a local skilled provider density, local all provider density, the skilled
provider density of the nearest facility, and the all provider density of the nearest facility.
The study aims to demonstrate two things: one substantive, one methodological. The first is to char-
acterize perceived provider barriers with sociodemographic factors and an objective measure of health
system readiness, provider density. The second is to explore what specific measures of provider density,
local or general, skilled provider or all provider, best contribute to this understanding of self-reported
provider barriers for use in future studies of health outcomes and behaviors.
4.2 Methods
4.2.1 Datasources
The data used in this study come from three sources: the 2015-16 Malawi Demographic and Health Survey,
the 2013-14 Malawi Service Provision Assessment, and the estimates of population distribution in Malawi
from WorldPop.
4.2.1.1 2015-16MalawiDemographicandHealthSurvey
The 2015-16 Malawi Demographic and Health Service (MDHS) is a nationally representative survey provid-
ing population-level estimates of key indicators of health, including fertility and family planning, HIV/AIDS,
nutrition, infant, child, and maternal health and mortality, among other indicators of population demo-
graphic characteristics and health. The 2015-16 MDHS employed a two-stage sample survey design with
51
850 primary sampling units or clusters (173 urban and 677 rural) drawn from a complete list of enumera-
tion areas (EAs) supplied by the 2008 Malawi Population and Housing Census ([50]). In the second stage
of sampling, 26,361 households were selected from among the 850 survey clusters. These data, along with
the centroid (geographic coordinates of the center of the survey cluster) of each of the 850 clusters are
publicly available (Figure 4.1).
In order to ensure the confidentiality of the respondents, the geographic data of Demographic and
Health Surveys undergo a process of geographic masking before public release, where locations are dis-
placed in accordance with well-documented procedures ([10]). The location of urban clusters is displaced
up to 2 kilometers (0-2km) and the location of rural clusters is displaced up to 5 kilometers (0-5km). A
random subset of 1% of the rural clusters are displaced up to 10 kilometers and all displacements are
constrained by district boundaries, meaning that even with the displacement, all cluster locations remain
inside the boundary of the original district ([10]).
This study focuses on rural areas because Malawi’s population is predominantly rural, and the dy-
namics of barriers and access to healthcare differ between rural and urban areas. In urban areas, both
population and facility densities are higher and there are greater transportation options and private sec-
tor providers. This study uses the data from 19,313 respondents across 677 rural clusters, their perceived
barriers to health care, their sociodemographic characteristics, and the location of the center of their cor-
responding survey cluster, which has been geographically displaced prior to public release.
4.2.1.2 2013-14MalawiServiceProvisionAssessment
The 2013-14 MSPA is a census of all health facilities in Malawi collecting data on facility infrastructure
and readiness to deliver quality essential health services ([54]). The location of each of the 977 surveyed
health facilities is supplied with the survey data and not displaced prior to public release (Figure 4.1).
52
R u r a l D H S C l u s t e r s
H e a l t h F a c i l i t i e s
M a l a w i
Figure 4.1: Map of Malawi (black outline) showing location of 977 health facilities surveyed in the 2013-14
Malawi Service Provision Assessment (red cross symbol) and 677 rural clusters surveyed in the 2015-16
Malawi Demographic and Health Survey (blue circle symbol). The inset map shows the African continent
with Malawi highlighted in purple.
53
4.2.1.3 WorldPopestimatesofMalawipopulation
In an effort to support global health and development efforts with spatial data, WorldPop, an interdisci-
plinary initiative develops, validates, and makes publicly available spatially referenced demographic and
health information ([77]). This study uses a gridded map that provides the number of people estimated
to be living in 100 by 100 square-meter cells (Figure 4.2). WorldPop develops annual estimates for each
country, but the 2014 estimate is used here in order to be generally aligned with the 2015-16 DHS and
2013-14 SPA surveys used in this study ([94]). The specific iteration of the population distribution map
used in this study is adjusted by WorldPop to match the corresponding official United Nations population
estimates ([68]). The data are available at a resolution of 3 arc seconds, which is approximately 100 meters
at the Equator.
4.2.2 Sociodemographicvariables
The study also accounts for the following sociodemographic variables: age, educational attainment of the
woman, household wealth quintile (among rural households), whether the woman has a child under 5
years of age, martial status, working status, and the frequency of media exposure (any of the following:
radio, tv, newspaper, and magazine).
4.2.3 Self-reportedbarrierstohealthcare
The MDHS asks women about series of seven factors that might be major barriers in seeking medical advice
or treatment for themselves when they are sick. These include: (1) distance to the health facility; (2) getting
permission to go to the doctor; (3) getting money needed for advice or treatment; (4) not wanting to go
alone; (5) concern that there may not be a female provider; (6) concern that there may not be any health
provider; (7) concern that there may be no drugs available ([50]). For this study, these seven factors have
been grouped thematically: distance, personal barriers (getting permission, getting money, not wanting
54
Total number of
people per grid
Low (0)
High (>650)
Figure 4.2: Estimated total number of people per grid-cell for Malawi in 2014 by WorldPop. Each grid is
approximately a 100-meter square and has a value that is the total estimated number of people covered by
that area. Cell values range from 0 to 652 people, with a mean of about 1. Map of Malawi shown along
with a zoomed-in inset to better demonstrate pixels.
to go alone), and provider barriers (concern that there may not be a female provider, that there may not
be any provider, no drugs available). The personal and provider barrier groups, each with a total of three
barriers, are treated as ordinal categorical variables, with it being possible that a women experiences none
of the barriers, one of the barriers, two barriers, or all three barriers.
55
4.2.4 Measuresofskilledproviderdensity
This study generated four measures of health provider density as an objective measure to compare with
levels of self-reported provider barriers. Separate measures were calculated for skilled providers (doctors,
nurses, and midwives) and for all providers, regardless of their type (pharmacists, laboratory technolo-
gists/scientists, assistants, dentists).
4.2.4.1 Closestfacilityproviderdensity
The number and type of providers associated with each facility is available in the MSPA. A provider density
for each facility was generated by drawing a 5km buffer around each facility and accounting for the total
population in that 5km catchment area of the facility.
The number of skilled and all providers was extracted from the MSPA dataset in R (version 4.1.3) and
joined with facility locations from the MSPA in QGIS (version 3.16.10 Hannover). The 5km catchment areas
around each facility were drawn, and the population in that area was summed in QGIS to derive skilled
and all provider density per 1,000 population for each facility. These steps of using facility attribute data
from the MSPA and population distribution data from WorldPop to generate facility provider densities is
represented in Step 1 of Figure 4.3.
4.2.4.2 Localproviderdensity
Steps 2 and beyond of Figure 4.3 represent the steps of how local, DHS cluster-specific provider densities
were derived. The goal in deriving this measure is in the interest of trying to obtain a more accurate
estimate of locally experienced health system readiness. A 5km buffer is drawn around the DHS cluster
locations to account for the displacement procedure prior to the public release of location data to preserve
the privacy of survey respondents. This displacement procedure randomly displaces 99% of all rural survey
cluster locations up to 5km from its original location and a random subset of 1% of rural clusters up to 10km.
56
MDHS 2015-16
- Survey cluster (EA)
locations
MSPA 2013-14
- Health facility locations
- # skilled providers per
facility
WorldPop
- Malawi population
distribution for
every 100m
2
in 2014
1. Generate provider
density for facility
catchment area
(5km radius)
2b. Intersect 2a. Overlay 2c. Clip
Result: Local provider density
3. Sum overlapping provider
densities in each survey cluster
+ +
2. Generate provider densities for
DHS survey clusters
4. Weight provider density by
underlying population to generate
local (survey cluster)
provider density
HIGH
LOW
HIGH
LOW
Figure 4.3: Outline of steps for generating health facility-specific provider density and local (DHS survey
cluster) provider densities from three data sources: MSPA 2013-14, MDHS 2015-16, and WorldPop popula-
tion distribution
57
The conceptual motivation is that a local estimate of provider density will a better estimate of access to
health care and provider availability than a proxy estimate represented by the provider density of the
closest health facility.
With the assumption that health facilities have a uniform reach of about 5km (estimated by WHO to
be an hour walking distance [93]) and that households could be anywhere in the 5km buffer of the pro-
vided cluster location, a local estimate of provider density was derived that would account for overlapping
catchment areas of multiple facilities as well as the population distribution of the 5km cluster buffer to
better account for more and less populated areas.
Given facility-specific provider densities for skilled and all providers, the Kernel Density Estimation
tool in QGIS was used to generate a gridded surface of 50 meter square pixels where every facility had a
5km radius reach with uniform (no distance decay applied) provider density values for that 5km catchment
area. The facility provider densities were converted from vector-type data to raster (gridded) format to
ease the subsequent steps of summing the densities of overlapping areas and then weighing by underlying
population to generate the final estimate of local provider density.
Once gridded 5km facility catchment were generated (blue circles in step 2a of Figure 4.3), 5km buffers
were drawn around each survey cluster from the DHS (beige circles in step 2a of Figure 4.3). The health
facility catchment areas overlapping with the DHS cluster buffers were clipped to just the areas where the
two layers intersected (steps 2b and 2c of Figure 4.3). Step 3 of Figure 4.3 shows the summing of provider
densities, represented by the shades of blue, of overlapping catchment areas to reflect the improved access
of those living in the vicinity of multiple facilities compared to those living within reach of just one or
fewer facilities. Finally, the WorldPop map of gridded population was overlaid and the raster calculator
and zonal statistics tools in QGIS were used to to generate a provider density for each DHS survey cluster
that was weighted by the underlying population. The final result was a skilled and all provider density per
1,000 population per DHS survey cluster.
58
4.2.5 Analysis
4.2.5.1 Descriptiveanalysisandexplorationofproviderdensitymeasures
Summary statistics were generated on the sociodemographic characteristics of the study population and
on the measures of perceived barriers and provider densities that were generated for this analysis. The
measures of skilled a provider density were summarized and explored to characterize their relationship
with each other and potentially with other variables, such as the number of perceived provider barriers.
4.2.5.2 Multilevellogisticmodeling
Accounting for the nested structure of the DHS data, we implemented multilevel logistic models to deter-
mine the relationship between the set of four measures of provider density and three different outcomes,
adjusting for various sociodemographic characteristics.
The first outcome was the odds of perceiving high provider barrier to health care. High provider barrier
was defined as experiencing two or more provider barriers. This designation is motivated by the results
of the previous chapter, which found that two or more provider barriers significantly reduced the odds of
antenatal care utilization.
The second outcome was the odds of perceiving the lack of any provider to be a major barrier to health
care. The concern that there will be no provider is one of the three components of the grouped provider
barrier index and the one that is most relevant to the measure of provider density.
The third outcome is the the attainment of four or more antenatal care visits (ANC4+), which is the
minimum number of ANC visits recommended by the WHO for a healthy and positive pregnancy and
delivery experience.
Exploring all three outcomes allows for the exploration of the relationship between provider density
and perceived provider barriers as well as actual utilization of health care.
59
4.3 Results
4.3.1 Measuresofproviderdensity
Overall, the study population of rural Malawian women live in areas with a mean of 1.52 skilled providers
per 1,000 people (Table 4.1). This is far short of the WHO-recommended 4.45 skilled provider per 1,000
population. The local provider density is consistently higher (for both skilled and all provider) than the
provider density for the closest health facility. Those perceiving 2+ provider barriers and those concerned
that there will be no provider have lower provider densities than their counterparts who do not experience
that concern, though these differences are not statistically significant (Table 4.1).
The top half of Figure 4.4 shows the distribution of local skilled provider density among the study
population. With a median of 0.63 skilled providers per 1,000 people, it is clear that the majority of the
population live in areas far below the WHO recommendation for skilled provider density. Still, there are
some who live in areas where the skilled provider density exceeds the WHO recommendation. The bottom
half of Figure 4.4 shows the distribution of local skilled provider density by the number of total provider
barriers perceived, from no barriers to a maximum of three barriers. The three vertical lines in each of the
density plots represent the first, second, and third quartiles (Q1, Q2, representing the overall median, and
Q3). While the differences are not pronounced, Figure 4.4 shows how the median local skilled provider
density decreases with an increase in the number of perceived provider barriers. The same figure shows
how the those living in areas with higher local skilled provider densities, above the WHO-recommended
density of 4.45, tend to be those experiencing no provider barriers (Figure 4.4).
For both all and skilled provider densities, there is a positive correlation between the local provider
density for each DHS survey cluster and the density of the closest facility (Figure 4.5). In general, the local
provider density is higher than that of the closest health facility (Table 4.1). This trend is shown by all
the points above the grey, dashed reference line (slope=1) in Figure 4.5. For example, the median local
60
Table 4.1: Background characteristics perceiving high (2+) provider barrier to health care among rural
Malawian women (n=19,315) with different measures of health provider density.
Doyouperceive2ormoreproviderfactors
tomajorbarrierstoobtaininghealthcare?
Isyourconcernthattherewillbenoprovider
amajorbarriertoobtaininghealthcare?
No
(n=9,321)
%(n)
Yes
(n=9,994)
%(n)
No
(n=9,453)
%(n)
Yes
(n=9,862)
%(n)
Overall
(n=19,315)
%(n)
Age
15-19 years 21.3% (1989) 21.6% (2162) 21.5% (2035) 21.5% (2116) 21.5% (4151)
20-34 years 50.8% (4739) 50.8% (5077) 50.8% (4800) 50.9% (5016) 50.8% (9816)
35-49 years 27.8% (2593) 27.6% (2755) 27.7% (2618) 27.7% (2730) 27.7% (5348)
Education
No education or primary school 78.1% (7281) 82.3% (8227) 78.2% (7388) 82.3% (8120) 80.3% (15508)
Secondary or higher 21.9% (2040) 17.7% (1767) 21.8% (2065) 17.7% (1742) 19.7% (3807)
Householdwealthquintile(rural)
Poorest 16.4% (1531) 21.0% (2099) 16.5% (1557) 21.0% (2073) 18.8% (3630)
Poorer 17.7% (1649) 20.4% (2038) 17.7% (1675) 20.4% (2012) 19.1% (3687)
Middle 19.2% (1785) 19.1% (1913) 19.2% (1818) 19.1% (1880) 19.1% (3698)
Richer 20.3% (1890) 19.8% (1977) 20.2% (1911) 19.8% (1956) 20.0% (3867)
Richest 26.5% (2466) 19.7% (1967) 26.4% (2492) 19.7% (1941) 23.0% (4433)
Haschildunder5years
No 43.5% (4058) 41.3% (4132) 43.5% (4111) 41.4% (4079) 42.4% (8190)
Yes 56.5% (5263) 58.7% (5862) 56.5% (5342) 58.6% (5783) 57.6% (11125)
Maritalstatus
Never in union 19.9% (1858) 19.2% (1914) 20.2% (1910) 18.9% (1862) 19.5% (3772)
Married or living with partner 66.1% (6162) 67.4% (6739) 65.9% (6232) 67.6% (6669) 66.8% (12901)
Widowed, divorced, or separated 14.0% (1301) 13.4% (1341) 13.9% (1311) 13.5% (1331) 13.7% (2642)
Workingstatus
Not currently working 40.0% (3732) 33.8% (3378) 40.1% (3793) 33.6% (3317) 36.8% (7110)
Currently working 60.0% (5589) 66.2% (6616) 59.9% (5660) 66.4% (6545) 63.2% (12205)
Frequencyofmediaexposure(radio/tv/newspaper/magazine)
None 47.8% (4451) 50.1% (5009) 47.8% (4517) 50.1% (4943) 49.0% (9460)
Less than once a week 18.2% (1693) 17.5% (1749) 18.2% (1721) 17.5% (1721) 17.8% (3442)
At least once a week 34.1% (3177) 32.4% (3236) 34.0% (3215) 32.4% (3198) 33.2% (6413)
Perceivedtobemajorbarriertohealthcare
Distanceisabigbarriertohealthcare
No 56.2% (5242) 27.7% (2766) 55.5% (5248) 28.0% (2760) 41.5% (8008)
Yes 43.8% (4079) 72.3% (7228) 44.5% (4205) 72.0% (7102) 58.5% (11307)
Personalbarrierstohealthcare(gettingpermission,gettingmoney,notwantingtogoalone)
0 barriers 57.1% (5319) 22.7% (2273) 56.3% (5324) 23.0% (2268) 39.3% (7592)
1 barrier 31.5% (2939) 33.2% (3320) 31.7% (2996) 33.1% (3263) 32.4% (6259)
2 barriers 9.8% (916) 28.5% (2846) 10.2% (965) 28.4% (2797) 19.5% (3762)
All 3 are barriers 1.6% (147) 15.6% (1555) 1.8% (168) 15.6% (1534) 8.8% (1702)
Provider-relatedbarrierstohealthcare(concernthattheremaynotbeafemaleprovider,anyprovider,nodrugsavailable)
0 barriers 62.5% (5827) 0% (0) 61.6% (5827) 0% (0) 30.2% (5827)
1 barrier 37.5% (3494) 0% (0) 34.8% (3289) 2.1% (205) 18.1% (3494)
2 barriers 0% (0) 59.1% (5903) 3.6% (337) 56.4% (5566) 30.6% (5903)
All 3 are barriers 0% (0) 40.9% (4091) 0% (0) 41.5% (4091) 21.2% (4091)
Measuresofproviderdensity
Skilledprovidersper1,000population(local)
Mean (SD) 1.82 (2.56) 1.24 (1.92) 1.80 (2.55) 1.25 (1.93) 1.52 (2.27)
Median [Min, Max] 0.732 [0, 13.5] 0.570 [0, 13.5] 0.731 [0, 13.5] 0.571 [0, 13.5] 0.630 [0, 13.5]
Allprovidersper1,000population(local)
Mean (SD) 4.79 (4.59) 3.77 (3.60) 4.76 (4.57) 3.79 (3.62) 4.26 (4.14)
Median [Min, Max] 3.37 [0, 26.1] 2.95 [0, 26.1] 3.31 [0, 26.1] 2.95 [0, 26.1] 3.09 [0, 26.1]
Skilledprovidersper1,000population(closesthealthfacility)
Mean (SD) 0.446 (0.756) 0.312 (0.511) 0.443 (0.753) 0.314 (0.511) 0.377 (0.644)
Median [Min, Max] 0.189 [0, 4.83] 0.179 [0, 4.83] 0.188 [0, 4.83] 0.180 [0, 4.83] 0.183 [0, 4.83]
Allprovidersper1,000population(closesthealthfacility)
Mean (SD) 1.39 (1.40) 1.19 (1.06) 1.38 (1.40) 1.19 (1.06) 1.29 (1.24)
Median [Min, Max] 0.998 [0.00500, 10.7] 0.997 [0.00500, 10.7] 0.998 [0.00500, 10.7] 0.997 [0.00500, 10.7] 0.998 [0.00500, 10.7]
skilled provider density is 0.63 compared to 0.18 for the closest health facility (Table 4.1). The median
local all provider density is 3.09 while the median all provider density of the closest health facility is 1.00
(Table 4.1). However, as shown by the points below the dashed reference line in Figure 4.5, this trend of a
61
0
500
1000
1500
0.0 2.5 5.0 7.5 10.0
Density of skilled health provider per 1,000 population
Total count
0 provider
barriers
1 provider
barriers
2 provider
barriers
3 provider
barriers
0.0 2.5 5.0 7.5 10.0
Density of skilled health provider per 1,000 population
Figure 4.4: Top: Histogram of local skilled health provider density in total study population (n=19,315);
Bottom: Density plots of local skilled health provider density by the total number of provider barriers
(0-3) experienced by individuals. The three black vertical lines in each density plot represent the Q1,
Q2 (median), and Q3. The vertical dashed red line in both top and bottom graphs represents the WHO-
recommended density of 4.45 skilled providers per 1,000 population.
62
higher local provider density compared to that of the closest health facility does not hold true for all DHS
survey clusters.
0
10
20
0 3 6 9
All health provider density per 1,000 population (closest facility)
All health provider density per 1,000 population (local)
0
5
10
0 1 2 3 4 5
Skilled health provider density per 1,000 population (closest facility)
Skilled health provider density per 1,000 population (local)
Figure 4.5: Top: Comparing the all health provider density of the closest facility (x-axis) vs local all health
provider density (y-axis). Bottom: Comparing the skilled health provider density of the closest facility (x-
axis) vs local skilled health provider density (y-axis). In both top and bottom graphs, the red line represents
best-fit line with shading representing the confidence interval. Dashed black line is a reference line of slope
1. In both graphs, the points above the dashed reference line represent survey clusters for which the local
provider density is higher than that of the closest health facility.
Figure 4.6 shows a summary of district-level estimates of skilled health provider density and its re-
lationship with other measures summarizing the population that is covered by a 5km buffer of a health
63
facility. The top graph of Figure 4.6 shows a positive correlation with skilled provider density and the
percentage of all health workers that are skilled. These two measures are not unrelated but show that
districts with low skilled provider density also have the lowest proportions of skilled health workers in
their general workforce. The bottom graph demonstrates that an increase in skilled provider density is
associated with an increase in the proportion of the population covered in a health facility catchment area.
In essence, a higher proportion of the district’s population is within reach of a health facility with an in-
creasing skilled provider density. Each circle represents a district, and the size of the circles on the bottom
graph shows the size of the population that is not covered by a facility catchment area. Proportions are
important for relative comparisons, but the size of the problem (as represented by the number of people
affected) is also important to consider when making decisions about resource allocations.
4.3.2 Multilevellogisticmodels
Table 4.2 shows how all four measures of health provider density are significantly associated with the
reduction in the odds of perceiving high provider barrier to health care. In both the case of local and closest
health facility provider density, skilled provider density is more impactful in reducing the experience of
high provider barrier compared to all provider density (Table 4.2). For the local measures: one skilled
provider increase per 1,000 people is associated with a 5% reduction (p = 0.001; Table 4.2) while an
increase in one provider of any skill level is associated with a 2% reduction (p = 0.011; Table 4.2) in the
odds of reporting high perceived provider barrier. For measures of the closest health facility: one skilled
provider increase is associated with a 12% reduction (p = 0.012; Table 4.2) and a one all provider increase
is associated with a 6% reduction (p = 0.032; Table 4.2) in the odds of reporting high perceived provider
barrier. Among the four model variations, the one using the local skilled provider density has the lowest
AIC value, implying a slightly better fit that the models using the other measures of provider density
(Table 4.2).
64
Chitipa
Karonga
Nkhata Bay
Rumphi
Mzimba
Kasungu
Nkhotakota
Ntchisi
Dowa
Salima
Lilongwe
Mchinji
Dedza Ntcheu
Mangochi Machinga
Zomba
Chiradzulu
Blantyre
Mwanza
Thyolo
Mulanje
Phalombe
Chikwawa
Nsanje
Balaka
30
40
50
60
0.4 0.6 0.8 1.0
Skilled health providers per 1,000 population
Percent of all health workers that are
'skilled' (doctors, nurses, midwives)
Region
North
Central
South
40
60
80
0.4 0.6 0.8 1.0
Skilled health providers per 1,000 population
Percent of district population covered by
5km catchment area of a facility (%)
Population outside
facility coverage
2e+05
4e+05
6e+05
Figure 4.6: Summary of district-level estimates of skilled health provider density and the population that is
covered within the 5km buffer of a health facility. The top graph shows a positive correlation with skilled
provider density and the percentage of all health workers that are skilled. The bottom graph demonstrates
that an increase in skilled provider density is associated with an increase in the proportion of the population
that covered in a health facility catchment area. Each circle represents a district and the size of the circles
on the bottom graph shows the size of the population that is not covered by a facility catchment area to
demonstrate relative size of the issue.
65
Experiencing other types of self-reported barriers increase the odds of perceiving high provider barrier.
Those reporting that distance is a barrier to health care have twice the odds of those who think distance
is not a barrier of perceiving high provider barrier (p < 0.001; Table 4.2). Compared to those who ex-
perience no personal barriers, those who experience any number of personal barriers (1-3 barriers) have
at least twice the odds of perceiving high provider barrier (p < 0.001; Table 4.2). In addition, having a
young child, currently working, and being in the category of greatest media exposure (at least one in-
stance of newspaper/magazine, radio, tv exposure per week) are all associated with an increase in the odds
of perceiving high provider barrier (Table 4.2).
Table 4.2: Summary of multilevel models for perceiving high (2+) provider barrier to health care among
rural Malawian women (n=19,315) with different measures of health provider density.
ModelA1 ModelA2 ModelA3 ModelA4
Predictors OR CI p OR CI p OR CI p OR CI p
(Intercept) 0.32 0.25–0.42 <0.001 0.33 0.25–0.43 <0.001 0.32 0.24–0.41 <0.001 0.32 0.25–0.43 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 0.93 0.83–1.05 0.232 0.93 0.82–1.05 0.225 0.93 0.82–1.05 0.224 0.93 0.82–1.05 0.218
35-49 years old 0.92 0.80–1.04 0.190 0.91 0.80–1.04 0.184 0.91 0.80–1.04 0.179 0.91 0.80–1.04 0.175
Education
(ref: Noneorprimary)
Secondary or higher 0.95 0.86–1.04 0.274 0.95 0.86–1.04 0.264 0.94 0.86–1.04 0.242 0.94 0.86–1.04 0.234
Householdwealthquintile
(ref: Poorest)
Poorer 1.04 0.93–1.16 0.504 1.04 0.93–1.16 0.501 1.04 0.93–1.16 0.493 1.04 0.93–1.16 0.495
Middle 0.93 0.83–1.05 0.239 0.93 0.83–1.05 0.234 0.93 0.83–1.05 0.236 0.93 0.83–1.05 0.233
Richer 1.06 0.94–1.19 0.327 1.06 0.94–1.19 0.341 1.06 0.94–1.19 0.354 1.05 0.94–1.18 0.369
Richest 1.04 0.91–1.18 0.571 1.03 0.91–1.17 0.608 1.03 0.91–1.17 0.646 1.03 0.91–1.16 0.678
Hasayoungchildunder5years 1.09 1.00–1.19 0.050 1.09 1.00–1.19 0.049 1.09 1.00–1.19 0.047 1.09 1.00–1.19 0.047
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.00 0.88–1.14 0.993 1.00 0.88–1.14 0.994 1.00 0.88–1.14 0.994 1.00 0.88–1.14 0.984
Widowed, divorced, or separated 0.90 0.77–1.05 0.172 0.90 0.77–1.05 0.172 0.89 0.76–1.05 0.167 0.89 0.76–1.05 0.166
Currentlyworking 1.13 1.04–1.22 0.003 1.13 1.04–1.22 0.003 1.13 1.04–1.22 0.002 1.13 1.04–1.22 0.002
Media(newspaper/magazine,radio,tv)exposure
(ref: noexposure)
Any media exposure less than once a week 1.02 0.93–1.13 0.636 1.02 0.93–1.13 0.628 1.02 0.93–1.13 0.633 1.02 0.93–1.13 0.638
Any media exposure at least once a week 1.17 1.07–1.27 <0.001 1.17 1.07–1.27 <0.001 1.17 1.07–1.27 <0.001 1.17 1.07–1.27 <0.001
Perceivedtobemajorbarrierstohealthcare
Distance
(ref: nodistancebarrier)
Distance is a barrier 2.01 1.86–2.18 <0.001 2.01 1.85–2.18 <0.001 2.02 1.86–2.19 <0.001 2.02 1.86–2.19 <0.001
Personalbarriers(gettingpermission,gettingmoney,notwantingtogoalone)
(ref: nopersonalbarriers)
1 personal barrier (any one) 2.04 1.88–2.22 <0.001 2.04 1.88–2.22 <0.001 2.04 1.88–2.22 <0.001 2.04 1.88–2.22 <0.001
2 personal barriers (any two) 4.85 4.36–5.40 <0.001 4.85 4.36–5.41 <0.001 4.86 4.36–5.41 <0.001 4.86 4.36–5.41 <0.001
3 personal barriers (all three) 15.60 12.87–18.90 <0.001 15.61 12.89–18.92 <0.001 15.62 12.89–18.92 <0.001 15.64 12.91–18.94 <0.001
Measuresofhealthproviderdensity(allper1,000population)
Skilled provider density (local) 0.95 0.93–0.98 0.001
All provider density (local) 0.98 0.96–1.00 0.011
Skilled provider density (closet health facility) 0.88 0.79–0.97 0.012
All provider density (closet health facility) 0.94 0.90–0.99 0.032
RandomEffects
σ 2
3.29 3.29 3.29 3.29
τ 00 0.36
cluster:district
0.37
cluster:district
0.37
cluster:district
0.37
cluster:district
0.37
district
0.38
district
0.37
district
0.38
district
ICC 0.18 0.18 0.18 0.19
AIC 21262.0 21268.9 21267.0 21268.7
66
For the second set of models for the outcome that the concern that there will be no provider is a major
barrier to health care, the results are very similar to the outcome of perceiving high provider barrier in
general (Table 4.3). Again, all four measures of provider density (local and closest health facility, skilled
and all provider) are statistically significantly associated with a reduction in the odds of the outcome (all
p< 0.05; Table 4.3), with the model using the local skilled provider density having the smallest AIC among
the four models. Again, experiencing other types of self-reported barriers increase the odds of the concern
that there will be no provider being a major barrier to health care (all p < 0.001; Table 4.3). Currently
working and being in the highest media exposure category similarly increase the odds of the outcome.
The only difference is that having a young child is no longer associated with the outcome (Table 4.3).
The third set of models had ANC4+ as the outcome. This set of models used the same covariates as the
models for ANC4+ in chapter 3 for comparability. The model for ANC4+ as implemented in chapter 3 is
replicated in Table 4.5 as Model C5. Unlike the self-reported barrier outcomes in the first and second set
of models, only the local provider densities were significantly associated with ANC4+. One local skilled
provider increase per 1,000 people increases the odds of attaining ANC4+ by 3% (p = 0.022; Table 4.4)
while an increase in any local provider increases the odds of ANC4+ by 2% (p = 0.003; Table 4.4). Though
not significant, one skilled provider increase at the closest health facility is associated with a greater impact
in ANC4+ attainment compared with an increase in any kind of provider (Table 4.4).
4.4 Discussion
While the general population may not know the type and number of the staff that employ their local
hospitals, they have lived and shared experiences that inform form their perceptions of health system
readiness to provide quality care. This study demonstrates that perceptions of provider-related barriers
are associated with objective measures of provider density and that there is a sensitivity to provider type,
with an indication of preference, or a positive bias, for skilled providers.
67
Table 4.3: Summary of multilevel models for the concern that there will be no provider being a major
barrier to health care among rural Malawian women (n=19,315) with different measures of health provider
density.
ModelB1 ModelB2 ModelB3 ModelB4
Predictors OR CI p OR CI p OR CI p OR CI p
(Intercept) 0.31 0.24–0.41 <0.001 0.32 0.24–0.42 <0.001 0.31 0.24–0.40 <0.001 0.32 0.24–0.42 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 0.93 0.82–1.04 0.208 0.93 0.82–1.04 0.201 0.93 0.82–1.04 0.201 0.92 0.82–1.04 0.196
35-49 years old 0.91 0.80–1.04 0.160 0.91 0.80–1.04 0.155 0.91 0.80–1.04 0.151 0.91 0.80–1.03 0.148
Education
(ref: Noneorprimary)
Secondary or higher 0.95 0.87–1.05 0.308 0.95 0.86–1.05 0.296 0.95 0.86–1.04 0.278 0.95 0.86–1.04 0.271
Householdwealthquintile
(ref: Poorest)
Poorer 1.03 0.92–1.15 0.568 1.03 0.92–1.15 0.565 1.03 0.93–1.16 0.556 1.03 0.92–1.16 0.559
Middle 0.92 0.82–1.03 0.155 0.92 0.82–1.03 0.152 0.92 0.82–1.03 0.153 0.92 0.82–1.03 0.152
Richer 1.06 0.95–1.19 0.318 1.06 0.94–1.19 0.332 1.06 0.94–1.19 0.341 1.06 0.94–1.18 0.356
Richest 1.03 0.91–1.17 0.661 1.02 0.90–1.16 0.700 1.02 0.90–1.16 0.732 1.02 0.90–1.15 0.767
Hasayoungchildunder5years 1.06 0.97–1.15 0.174 1.06 0.97–1.15 0.171 1.06 0.98–1.16 0.167 1.06 0.98–1.16 0.166
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.06 0.93–1.21 0.358 1.06 0.93–1.21 0.350 1.06 0.93–1.21 0.351 1.07 0.93–1.21 0.344
Widowed, divorced, or separated 0.96 0.82–1.13 0.641 0.96 0.82–1.13 0.641 0.96 0.82–1.13 0.631 0.96 0.82–1.13 0.630
Currentlyworking 1.14 1.06–1.23 0.001 1.14 1.06–1.23 0.001 1.14 1.06–1.23 0.001 1.14 1.06–1.23 0.001
Media(newspaper/magazine,radio,tv)exposure
(ref: noexposure)
Any media exposure less than once a week 1.02 0.92–1.12 0.720 1.02 0.93–1.12 0.712 1.02 0.92–1.12 0.717 1.02 0.92–1.12 0.723
Any media exposure at least once a week 1.16 1.07–1.26 <0.001 1.16 1.07–1.26 <0.001 1.16 1.07–1.26 <0.001 1.16 1.07–1.26 <0.001
Perceivedtobemajorbarrierstohealthcare
Distance
(ref: nodistancebarrier)
Distance is a barrier 1.93 1.78–2.10 <0.001 1.93 1.78–2.10 <0.001 1.94 1.79–2.10 <0.001 1.94 1.79–2.10 <0.001
Personalbarriers(gettingpermission,gettingmoney,notwantingtogoalone)
(ref: nopersonalbarriers)
1 personal barrier (any one) 1.99 1.83–2.16 <0.001 1.99 1.83–2.16 <0.001 1.98 1.83–2.16 <0.001 1.98 1.83–2.16 <0.001
2 personal barriers (any two) 4.57 4.11–5.09 <0.001 4.58 4.11–5.09 <0.001 4.58 4.11–5.09 <0.001 4.58 4.12–5.09 <0.001
3 personal barriers (all three) 13.55 11.28–16.27 <0.001 13.56 11.29–16.28 <0.001 13.56 11.30–16.29 <0.001 13.58 11.31–16.31 <0.001
Measuresofhealthproviderdensity(allper1,000population)
Skilled provider density (local) 0.96 0.93–0.98 0.002
All provider density (local) 0.98 0.97–1.00 0.025
Skilled provider density (closet health facility) 0.88 0.79–0.97 0.012
All provider density (closet health facility) 0.94 0.89–0.99 0.021
RandomEffects
σ 2
3.29 3.29 3.29 3.29
τ 00 0.35
cluster:district
0.36
cluster:district
0.36
cluster:district
0.36
cluster:district
0.36
district
0.37
district
0.37
district
0.37
district
ICC 0.18 0.18 0.18 0.18
AIC 21527.2 21531.8 21530.6 21531.5
For both the provider barrier index and the individual “no provider concern” barrier, both skilled and all
provider density as well as local and closest health facility measures of provider density were statistically
significant (Table 4.2, Table 4.3). In all cases, an increase in provider density was significantly associated
with a reduction in the odds of perceiving a provider barrier, which is the expected impact of improved
provider barrier. Also, an increase in skilled providers was always associated with a greater reduction in
the odds of perceiving provider barriers compared to its all provider counterpart. This suggests that people
may detect differences in provider types and have a preference for skilled providers, or at least associate
them positively with health care readiness. This is consistent with research suggesting that rather than
68
Table 4.4: Summary of multilevel models for ANC4+ among rural Malawian women (n=10,825) with dif-
ferent measures of health provider density.
ANC4+(ModelC1) ANC4+(ModelC2) ANC4+(ModelC3) ANC4+(ModelC4)
Predictors OR CI p OR CI p OR CI p OR CI p
(Intercept) 0.46 0.34–0.63 <0.001 0.44 0.32–0.60 <0.001 0.47 0.34–0.64 <0.001 0.47 0.34–0.64 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 1.17 1.00–1.38 0.051 1.17 1.00–1.38 0.051 1.18 1.00–1.38 0.049 1.18 1.00–1.38 0.047
35-49 years old 1.33 1.07–1.66 0.010 1.33 1.07–1.66 0.010 1.34 1.07–1.67 0.009 1.34 1.08–1.67 0.009
Education
(ref: Noeducation)
Primary school 1.08 0.94–1.23 0.278 1.07 0.94–1.23 0.288 1.08 0.95–1.23 0.257 1.08 0.95–1.23 0.253
Secondary or higher 1.18 0.99–1.40 0.063 1.17 0.99–1.39 0.070 1.18 1.00–1.41 0.054 1.19 1.00–1.41 0.052
Householdwealthquintile
(ref: Poorest)
Poorer 0.99 0.88–1.13 0.916 0.99 0.88–1.13 0.920 0.99 0.87–1.12 0.892 0.99 0.87–1.12 0.894
Middle 1.04 0.91–1.18 0.577 1.04 0.91–1.18 0.577 1.04 0.91–1.18 0.569 1.04 0.91–1.18 0.561
Richer 1.01 0.89–1.16 0.839 1.01 0.89–1.16 0.843 1.02 0.89–1.16 0.811 1.02 0.89–1.16 0.793
Richest 1.10 0.94–1.27 0.228 1.10 0.94–1.27 0.231 1.10 0.95–1.28 0.195 1.11 0.95–1.29 0.183
Parity 1.01 0.98–1.04 0.456 1.01 0.98–1.04 0.448 1.01 0.98–1.04 0.485 1.01 0.98–1.04 0.492
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.16 0.92–1.47 0.204 1.17 0.92–1.47 0.203 1.16 0.91–1.46 0.226 1.15 0.91–1.46 0.232
Widowed, divorced, or separated 1.24 0.96–1.61 0.097 1.24 0.96–1.60 0.098 1.24 0.96–1.60 0.104 1.24 0.96–1.60 0.105
Currentlyworking 1.08 0.99–1.19 0.096 1.08 0.99–1.19 0.097 1.08 0.98–1.18 0.105 1.08 0.98–1.18 0.106
Pregnancywasnotplanned 0.85 0.78–0.92 <0.001 0.85 0.78–0.92 <0.001 0.85 0.78–0.92 <0.001 0.84 0.78–0.92 <0.001
Media(newspaper/magazine,radio,tv)exposure
(ref: noexposure)
Any media exposure less than once a week 0.98 0.87–1.10 0.729 0.98 0.87–1.10 0.726 0.98 0.87–1.10 0.719 0.98 0.87–1.10 0.722
Any media exposure at least once a week 1.01 0.92–1.12 0.783 1.01 0.92–1.12 0.782 1.02 0.92–1.12 0.762 1.02 0.92–1.12 0.761
TimelyinitiationofANCinfirsttrimester 4.81 4.34–5.34 <0.001 4.82 4.34–5.34 <0.001 4.81 4.34–5.34 <0.001 4.81 4.34–5.34 <0.001
Perceivedtobemajorbarrierstohealthcare
Distance
(ref: nodistancebarrier)
Distance is a barrier 0.97 0.88–1.07 0.571 0.98 0.89–1.08 0.687 0.97 0.88–1.07 0.503 0.97 0.88–1.06 0.494
Personalbarriers(gettingpermission,gettingmoney,notwantingtogoalone)
(ref: nopersonalbarriers)
1 personal barrier (any one) 0.94 0.85–1.05 0.273 0.94 0.85–1.05 0.270 0.94 0.85–1.05 0.271 0.94 0.85–1.05 0.270
2 personal barriers (any two) 0.87 0.77–0.99 0.040 0.87 0.77–0.99 0.040 0.87 0.77–0.99 0.039 0.87 0.77–0.99 0.038
3 personal barriers (all three) 0.94 0.79–1.11 0.459 0.94 0.79–1.11 0.457 0.94 0.79–1.11 0.444 0.94 0.79–1.11 0.434
Measuresofhealthproviderdensity(allper1,000population)
Skilled provider density (local) 1.03 1.00 – 1.05 0.022
All provider density (local) 1.02 1.01–1.03 0.003
Skilled provider density (closet health facility) 1.06 0.97–1.15 0.190
All provider density (closet health facility) 1.02 0.98–1.07 0.305
RandomEffects
σ 2
3.29 3.29 3.29 3.29
τ 00 0.08
cluster:district
0.08
cluster:district
0.09
cluster:district
0.09
cluster:district
0.07
district
0.07
district
0.07
district
0.07
district
ICC 0.05 0.04 0.05 0.05
AIC 13766.8 13763.6 13770.4 13771.1
being passive consumers, patients in the rural African health care setting are “active patients” that seek
both the best-known provider and most appropriate facility for specific illnesses ([44]).
In modeling ANC4+, only the local estimates of provider density were significantly associated with the
measures of health care utilization. This suggests that while local or environmental provider density are
associated with the perception of provider-related barriers, actual behaviors and care-seeking decisions
are better predicted by local measures of health system readiness. This suggests that simple linking of
69
Table 4.5: Summary of multilevel models for ANC4+ among rural Malawian women (n=10,825) with per-
ceived provider barrier and local health provider density.
ANC4+(ModelC5) ANC4+(ModelC6)
Predictors OR CI p OR CI p
(Intercept) 0.50 0.36–0.68 <0.001 0.48 0.35–0.65 <0.001
Age
(ref: 15-19yearsold)
20-34 years old 1.18 1.01–1.39 0.043 1.18 1.00–1.38 0.047
35-49 years old 1.35 1.08–1.68 0.008 1.34 1.08–1.67 0.009
Education
(ref: Noeducation)
Primary school 1.08 0.94–1.23 0.275 1.07 0.94–1.22 0.296
Secondary or higher 1.18 1.00–1.41 0.054 1.17 0.99–1.40 0.067
Householdwealthquintile
(ref: Poorest)
Poorer 0.99 0.87–1.12 0.893 0.99 0.88–1.13 0.910
Middle 1.04 0.91–1.18 0.567 1.04 0.91–1.18 0.597
Richer 1.02 0.89–1.17 0.780 1.02 0.89–1.16 0.827
Richest 1.11 0.95–1.29 0.184 1.10 0.94–1.27 0.228
Parity 1.01 0.98–1.04 0.530 1.01 0.98–1.04 0.492
Maritalstatus
(ref: Nevermarried)
Married or living with partner 1.15 0.91–1.46 0.231 1.17 0.92–1.47 0.203
Widowed, divorced, or separated 1.23 0.95–1.59 0.108 1.24 0.96–1.60 0.100
Currentlyworking 1.08 0.99–1.19 0.089 1.09 0.99–1.19 0.083
Pregnancywasnotplanned 0.85 0.78–0.92 <0.001 0.85 0.78–0.92 <0.001
Media(newspaper/magazine,radio,tv)exposure
(ref: noexposure)
Any media exposure less than once a week 0.98 0.87–1.10 0.709 0.98 0.87–1.10 0.721
Any media exposure at least once a week 1.02 0.92–1.13 0.703 1.02 0.92–1.13 0.729
TimelyinitiationofANCinfirsttrimester 4.81 4.34–5.33 <0.001 4.81 4.34–5.33 <0.001
Perceivedtobemajorbarrierstohealthcare
Distance
(ref: nodistancebarrier)
Distance is a barrier 0.98 0.89–1.08 0.698 0.99 0.90–1.09 0.797
Personalbarriers(gettingpermission,gettingmoney,notwantingtogoalone)
(ref: nopersonalbarriers)
1 personal barrier (any one) 0.96 0.86–1.07 0.486 0.96 0.86–1.07 0.478
2 personal barriers (any two) 0.91 0.80–1.05 0.197 0.91 0.80–1.05 0.195
3 personal barriers (all three) 1.01 0.84–1.21 0.954 1.01 0.84–1.21 0.941
Provider-relatedbarrier(concernthattheremaynotbeafemaleprovider,anyprovider,nodrugs)
(ref: noproviderbarriers)
1 provider barrier (any one) 0.98 0.86–1.11 0.726 0.98 0.86–1.12 0.779
2 provider barriers (any two) 0.89 0.79–1.00 0.048 0.89 0.80–1.00 0.060
3 provider barriers (all three) 0.86 0.74–0.99 0.036 0.86 0.75–1.00 0.045
Measureofhealthproviderdensity(allper1,000population)
Skilled provider density (local) 1.03 1.00–1.05 0.028
RandomEffects
σ 2
3.29 3.29
τ 00 0.09
cluster:district
0.08
cluster:district
0.08
district
0.07
district
ICC 0.05 0.05
AIC 13769.6 13766.8
DHS data to the closest health facility for health system attributes might not be appropriate or specific
enough to contribute to understanding health behaviors. This also suggests that geospatial datasets and
70
spatial methods, such as those demonstrated in this analysis, can be used to contribute a more local, and
thus more relevant, local perspective.
Overall, this study suggests that local measures of skilled provider density are relevant to perceptions
of provider barrier and to health behaviors. Given the appropriate spatial datasets, it can be generated with
standard tools, as demonstrated here, in a geographic information system, such as QGIS, which is free and
open-source. While users must be cautious about making local conclusions, this study demonstrates a way
for users to generate local estimates that are relevant to aggregated analysis.
One interesting finding from both Model Set A (Table 4.2) Model Set B (Table 4.3) is that currently
working status and having the highest media exposure category are both associated with an increase
in the odds of perceiving provider barriers. Employment and media exposure are proxies of autonomy,
empowerment, and potential advantage/resourcefulness in overcoming barriers ([2, 22]). Instead, they are
each associated with a greater than 10% increase in the odds of perceiving provider barriers. To understand
these findings, we can refer to the body of work that finds that less educated individuals report greater
satisfaction with medical care than their counterparts with more education ([74, 70, 13, 33]). This body of
work suggests that greater knowledge may promote greater expectation and, in turn, a greater difference
in expectation and experience, leading to greater dissatisfaction.
Another interesting observation is that having a young child increases the odds of perceiving high
provider barrier according to the grouped provider barrier index (Table 4.2), but not the odds of perceiving
concern of no provider to be a barrier (Table 4.3). One of the reasons the seven barrier questions were
grouped into domains includes wanting to get a cumulative sense of the impact of general provider-related
barriers as perceived by people. Even though the three specific components of the grouped provider barrier
index ask specifically about concern about the availability of providers and drugs, the thought was that
these questions had the potential to capture, or at least be influenced by, additional, unspecified perceptions
71
of provider-related barriers, such as deficient infrastructure (water, electricity), long waits, and negative
attitudes/treatment.
The finding that having a young child increases the odds of perceiving high provider barrier might be
related the drug availability component of the compound index, which is not accounted for by any other
covariate in the model (Table 4.2). The SPA collects data on the availability of several different types of
drug stocks, including for HIV/AIDS, family planing, and those for the treatment of common childhood
diseases, such as Malaria, Tuberculosis, and Pneumonia. The availability of drug stocks vary by type ([54]).
For example, the 2013-14 SPA found that less than half of facilities that offer family planning methods
actually had all the methods they provide available on the day of the assessment([54]). In contrast, more
than 90% of facilities had first-line malaria treatment on-hand the day they were surveyed ([54]). While
the DHS specifies “care for self” as the context of the barrier questions, women’s answers may be informed
by recent or notable interactions with the health system, which in the case of women with young children
might be for a sick child rather than themselves.
While the Model C5, replicated in Table 4.5 from the chapter 3, found 2 or 3 provider barriers to
significantly decrease the odds of ANC4+, Model C6, which adds local skilled provider density, still shows
that 3 provider barriers (but not 2) is significantly associated with reduced odds of ANC4+. This suggests
that even after accounting for provider density, the grouped perceived barrier index captures a sense of
provider-related barriers that go beyond that which is measured by provider availability.
Taken together, this study supports the idea that these self-reported barriers include a rich personal
context and has the potential to enhance the understanding of health decisions in a unique way.
4.4.1 Limitations
There are several limitations of this work that can inform directions for future research. First, this study
made the assumption that each health facility had a catchment area of a 5-kilometer radius buffer. This is
72
reductionist in two ways, a circle does not accurately represent the shape of a catchment area around a
health facility and will be informed by the road network and modes of transportation that are available.
By applying a standard size catchment area this analysis ignores that different types of facilities will have
different reaches. Clinics and dispensaries will have a more local reach and higher level, larger facilities,
such as district hospitals providing more complex, specialized care will serve a larger catchment area.
Preference for certain facilities will also mean that people may be more willing to travel farther to obtain
health care ([73, 60]). Future studies can apply different buffer sizes to account for the broader reach of
larger facilities.
Availability is just one dimension of potential supply-side barriers to health care and health system
readiness. Future work can explore different measures that might make up provider-side barriers. Among
these, drug availability should be a priority factor to more fully characterize provider barriers ([37]). While
the volume of data collected by the SPA makes it a challenge to develop a meaningful composite measure
of health system readiness, such an indicator will be helpful in advancing the understanding perceived
barriers.
Finally, it must be acknowledged that this analysis was possible in a meaningful way because Malawi
had a SPA that was a census of health facilities and not a sample survey. While this analysis would be able
to be replicated meaningfully in only a few countries, the growing availability of routine data sources, such
as national health management information systems or health facility rosters, maintained and routinely
updated by the government, make approaches like this increasingly useful for a broader audience without
having to rely on survey data.
4.5 Conclusion
This study demonstrates that perceptions of provider-related barriers (lack of provider and drug availabil-
ity) said to hinder health care are associated with objective measures of provider density and that there
73
is a sensitivity to provider type, with an indication of preference for skilled providers (doctors, nurses,
and midwives). Local estimates of provider availability, generated from combining spatial data from the
DHS, SPA, and gridded population estimates is a better predictor of health service utilization than provider
availability at the nearest health facility. Even so, objective measures of provider density do not fully cap-
ture the personal context and lived experience that is expressed in a grouped index of perceived provider
barriers. The perceived barrier questions on the DHS should continued to be explored as a way to motivate
deeper understanding of the factors that facilitate and hinder care-seeking in countries that implement this
and similar surveys.
74
Chapter5
Conclusions
The overarching goal of this work was to characterize self-reported barriers to health care with the aim
of better understanding what these barriers represent in terms of objective measures. This work was
motivated by the substantive challenge of understanding why only half of pregnant women in Malawi
obtained the WHO-recommended four or more antenatal care visits while more than 90% of women got at
least one visit in a way that went beyond linking outcomes to individual sociodemographic characteristics.
This work also sought to demonstrate improved spatial methods of linking population health surveys and
facility surveys to contribute health system perspectives in analyses of health outcomes.
5.1 Maincontributions
5.1.1 Substantivecontributions
Perceived barriers are shown to reflect objective measures of their correlates. Chapter 2 demonstrates
that people who report distance to be a major barrier to health care do, in fact, live farther from health
facilities. Chapter 3 indicates that perceiving barriers to care are shown to moderate pregnancy health care
utilization. Finally, chapter 4 shows that people who report experiencing poor health system readiness live
in areas of reduced provider density.
5.1.2 Methodologicalcontributions
The main methodological contribution of this body of work is the innovative demonstration of a spatial
method of linking nationally-representative health and facility surveys. This body of work, specifically
75
in Chapters 2 and 4 show that a the basic assumption of 5-kilometer catchment area of a facility and a 5-
kilometer buffer of a rural survey cluster accounts for the uncertainty introduced by displacement enough
to result in locally relevant measures health facility access and provider density.
The two-buffer approaches demonstrated here take advantage of contemporary spatial datasets (i.e.,
gridded population) to improve on existing methods that do not account for the influence of facilities that
are close, but not explicitly within survey cluster buffers or the overlapping influence of multiple facilities
in proximity of households. By additionally accounting for heterogeneity in the population distribution
in a survey cluster buffer, the aggregate local measures of provider density generated from this approach
more accurately reflect health system readiness experienced by residents of a survey cluster.
5.2 Limitations
A limitation of this work is that only individual women’s perspectives are considered on a topic, health care
utilization during pregnancy, which we know to be impacted by many influential stakeholders including
husbands, partners, and family elders (e.g., mothers-in-law). One hope was that these husband/family
opinions, especially if they were at odds with the woman’s own perspective, would present in some way
in the composite measure of personal barriers (i.e., getting permission to go, getting money to go, or not
wanting to go alone are major barriers to getting health care). However, there may be other ways to better
and more explicitly account for the perspective of other, external stakeholders. While the DHS might not
have the data to explore the perceptions of men and other stakeholders in the pathway to pregnancy health
care utilization, it is important to continue to pursue the narrative of influential stakeholders to gain a more
complete understanding of the factors that facilitate and hinder women’s utilization of health care.
A significant limitation of this work is that the data requirements make this kind of analysis difficult
to replicate in other settings ([73, 12]). As a relatively small country, Malawi had the benefit of the Service
76
Provision Assessment being a census of all facilities, which is normally prohibitively costly and energy-
intensive to implement. Malawi was able to survey all the facilities in the country, amounting to less than
1,000 total, but in neighboring Tanzania, for example, their most recent SPA surveyed 1,200 among the
more 7,000 total facilities in the country ([54, 55]). Facility sample surveys introduce an uncertainty due to
incomplete information that this analysis did not have to consider. Incomplete understanding of the health
system would introduce uncertainty and limit the validity of the methods as described in this dissertation.
However, continued improvements in routine data systems that regularly collect data on health system
attributes will allow this analysis to be replicated in settings without having to reply on SPA-like surveys
([46, 88]).
On a related note, while difficulty to quantify in this work, it is important to acknowledge the errors
that are introduced by the temporal difference in the data collection periods for the 2013-14 SPA and 2015-
16 DHS. While this difference in time period is not negligible, there are two reasons it is still acceptable
to link these two data sources in analyses. The first is that the current guidance from the DHS program,
which implements both the DHS and SPA surveys suggests that as a rule of thumb, surveys conducted
within a year of one another are fairly well-matched ([12]). The second reason is that the DHS questions
on pregnancy care refer to the most recent pregnancy in the five years preceding the survey. This time
frame includes the data collection period of the 2013-14 SPA, which can thus be said to provide a snapshot
of health system readiness that is temporally aligned to the DHS data.
Another limitation is that many times, measures that result from spatial processes are not straightfor-
ward or intuitive to a general audience. Effort must be made to improve interpretation and communication
to better demonstrate the need and value of integrating spatial methods in health research.
77
5.3 Futurework
A next step for this work is to include a temporal perspective. An exploration of trends could strengthen
the value of the self-reported barrier variables and promote its potential to contribute an important, under-
explored perspective to health research. Specifically, the 2010 Malawi DHS asks the same perceived barrier
questions and could be studied with the responses from the 2015-16 DHS to understand if the relationship
between self-reported barriers in the various domains (i.e., personal barriers, provider barriers, distance)
and pregnancy care utilization outcomes have changed.
The replication of this work in different countries with other population health and facility surveys,
like the DHS and SPA, would also contribute an understanding of how the dynamics vary by country con-
text. Results are expected to be different in different contexts. Relative to neighboring countries, Malawi
is more rural, more poor, and more equitable in poverty (i.e. smaller absolute difference in wealth be-
tween those in the highest wealth quintile and those in the lowest wealth quintile) ([25]). Other low- and
middle-income countries with different sociodemographic profiles (e.g., larger wealth inequities, more ur-
ban population, etc.) and different health care structures and insurance schemes will likely show different
dynamics between perceived barriers and health care utilization.
The work presented here to characterize user perceptions of health care will ultimately allow for
decision-makers to leverage this richer perspective to allocate resources strategically and make policies to
improve the coverage, utilization, and impact of lifesaving health care.
78
Bibliography
[1] Pauliina Aarnio, Effie Chipeta, and Teija Kulmala. “Men’s Perceptions of Delivery Care in Rural
Malawi: Exploring Community Level Barriers to Improving Maternal Health”. In: Health Care for
Women International 34.6 (2013), pp. 419–439.issn: 0739-9332.doi: 10.1080/07399332.2012.755982.
[2] Solomon Kibret Abreha and Yacob Abrehe Zereyesus. “Women’s Empowerment and Infant and
Child Health Status in Sub-Saharan Africa: A Systematic Review”. In: Maternal and Child Health
Journal 25.1 (2021), pp. 95–106.issn: 15736628.doi: 10.1007/s10995-020-03025-y.
[3] Adetoro Adegoke, Bettina Utz, Sia E. Msuya, and Nynke van den Broek. “Skilled Birth attendants:
Who is who? a descriptive study of definitions and roles from nine Sub Saharan African
countries”. In: PLoS ONE 7.7 (2012).issn: 19326203.doi: 10.1371/journal.pone.0040220.
[4] Elizabeth M. Allen, Kathleen T. Call, Timothy J. Beebe, Donna D. Mcalpine, and Pamela Jo Johnson.
“Barriers to Care and Health Care Utilization Among the Publicly Insured”. In: Medical Care 55.3
(2017), pp. 207–214.issn: 0025-7079.doi: 10.1097/mlr.0000000000000644.
[5] Mastewal Arefaynie, Bereket Kefale, Melaku Yalew, Bezawit Adane, Reta Dewau, and
Yitayish Damtie. “Number of antenatal care utilization and associated factors among pregnant
women in Ethiopia: zero-inflated Poisson regression of 2019 intermediate Ethiopian Demography
Health Survey”. In: Reproductive Health 19.1 (2022).issn: 1742-4755.doi:
10.1186/s12978-022-01347-4.
[6] George Ashiagbor, Richard Ofori-Asenso, Eric K. Forkuo, and Seth Agyei-Frimpong. “Measures of
geographic accessibility to health care in the Ashanti Region of Ghana”. In: Scientific African 9
(2020), e00453.issn: 2468-2276.doi: 10.1016/j.sciaf.2020.e00453.
[7] T.T. Awoyemi, O.A. Obayelu, and H.I. Opaluwa. “Effect of Distance on Utilization of Health Care
Services in Rural Kogi State, Nigeria”. In: Journal of Human Ecology 35.1 (2011), pp. 1–9.doi:
10.1080/09709274.2011.11906385.
[8] Sumera Aziz Ali, Savera Aziz Ali, Anam Feroz, Sarah Saleem, Zafar Fatmai, and
Muhammad Masood Kadir. “Factors affecting the utilization of antenatal care among married
women of reproductive age in the rural Thatta, Pakistan: Findings from a community-based
case-control study”. In: BMC Pregnancy and Childbirth 20.1 (2020), pp. 1–12.issn: 14712393.doi:
10.1186/s12884-020-03009-4.
[9] Leslie Berman, Levison Nkhoma, Margaret Prust, Courtney Mckay, Mihereteab Teshome,
Dumisani Banda, Dalitso Kabambe, and Andrews Gunda. “Analysis of policy interventions to
attract and retain nurse midwives in rural areas of Malawi: A discrete choice experiment”. In:PLOS
ONE 16.6 (2021), e0253518.issn: 1932-6203.doi: 10.1371/journal.pone.0253518.
79
[10] Clara R. Burgert, John Colston, Thea Roy, and Blake Zachary. DHS Spatial Analysis Report 7:
Geographic displacement procedure and georeferenced data release policy for the Demographic and
Health Surveys. Tech. rep. Calverton, Maryland, USA: ICF International, 2013.
[11] CR Burgert, J Colston, T Roy, and B Zachary. “Geographic displacement procedure and
georeferenced data release policy for the Demographic and Health Surveys”. In: DHS Spatial
Analysis Reports No. 7 (2013).url: http://dhsprogram.com/pubs/pdf/SAR7/SAR7.pdf.
[12] CR Burgert and D Prosnitz. “Linking DHS Household and SPA Facility Surveys: Data
Considerations and Geospatial Methods”. In: DHS Spatial Analysis Reports No. 10 (2014).url:
https://dhsprogram.com/pubs/pdf/SAR10/SAR10.pdf.
[13] Matthew J. Carlson, James A. Shaul, Susan V. Eisen, and Paul D. Cleary. “The Influence of Patient
Characteristics on Ratings of Managed Behavioral Health Care”. In: Journal of Behavioral Health
Services and Research 29.4 (2002), pp. 481–489.issn: 10943412.doi: 10.1007/BF02287354.
[14] Liliana Carvajal–Aguirre, Vrinda Mehra, Agbessi Amouzou, Shane M Khan, Lara Vaz,
Tanya Guenther, Maggie Kalino, and Nabila Zaka. “Does health facility service environment
matter for the receipt of essential newborn care? Linking health facility and household survey data
in Malawi”. In: Journal of Global Health 7.2 (2017).issn: 2047-2978.doi: 10.7189/jogh.07.020508.
[15] Chancy S. Chimatiro, Precious Hajison, Effie Chipeta, and Adamson S. Muula. “Understanding
barriers preventing pregnant women from starting antenatal clinic in the first trimester of
pregnancy in Ntcheu District-Malawi”. In: Reproductive Health 15.1 (2018).issn: 1742-4755.doi:
10.1186/s12978-018-0605-5.
[16] Wanangwa Chimwaza, Effie Chipeta, Andrew Ngwira, Francis Kamwendo, Frank Taulo,
Susan Bradley, and Eilish Mcauliffe. “What makes staff consider leaving the health service in
Malawi?” In: Human Resources for Health 12.1 (2014), p. 17.issn: 1478-4491.doi:
10.1186/1478-4491-12-17.
[17] Tim Colbourn, Sonia Lewycka, Bejoy Nambiar, Iqbal Anwar, Ann Phoya, and Chisale Mhango.
“Maternal mortality in Malawi, 1977-2012”. In: BMJ Open 3.12 (2013).issn: 20446055.doi:
10.1136/bmjopen-2013-004150.
[18] Lotte Danielsen. “Enforcing ‘Progress’: A Story of an MDG 5 Indicator and Maternal Health in
Malawi”. In: Development and Change 48.3 (2017), pp. 429–451.issn: 14677660.doi:
10.1111/dech.12307.
[19] Hanifa M. Denny, Agung D. Laksono, Ratu Matahari, and Bina Kurniawan. “The Determinants of
Four or More Antenatal Care Visits Among Working Women in Indonesia”. In: Asia Pacific Journal
of Public Health 34.1 (2022), pp. 51–56.issn: 1010-5395.doi: 10.1177/10105395211051237.
[20] S. Downe, K. Finlayson, D. Walsh, and T. Lavender. “’Weighing up and balancing out’: A
meta-synthesis of barriers to antenatal care for marginalised women in high-income countries”. In:
BJOG: An International Journal of Obstetrics and Gynaecology 116.4 (2009), pp. 518–529.issn:
14700328.doi: 10.1111/j.1471-0528.2008.02067.x.
80
[21] Ending Preventable Maternal Mortality (EPMM). Ending Preventable Maternal Mortality (EPMM): A
renewed focus for improving maternal and newborn health and wellbeing. Tech. rep. World Health
Organization (WHO) and UNFPA, 2021.url:
https://www.who.int/publications/i/item/9789240040519.
[22] Fernanda Ewerling, John W. Lynch, Cesar G. Victora, Anouka van Eerdewijk, Marcelo Tyszler, and
Aluisio J.D. Barros. “The SWPER index for women’s empowerment in Africa: development and
validation of an index based on survey data”. In: The Lancet Global Health 5.9 (2017), e916–e923.
issn: 2214109X.doi: 10.1016/S2214-109X(17)30292-9.
[23] Amon Exavery, Almamy Malick Kanté, Ahmed Hingora, Godfrey Mbaruku, Senga Pemba, and
James F Phillips. “How mistimed and unwanted pregnancies affect timing of antenatal care
initiation in three districts in Tanzania”. In: BMC Pregnancy and Childbirth 13.1 (2013), p. 35.issn:
1471-2393.doi: 10.1186/1471-2393-13-35.
[24] Adeniyi Francis Fagbamigbe, Olugbenga Olaseinde, and Omon Stellamaris Fagbamigbe. “Timing of
first antenatal care contact, its associated factors and state-level analysis in Nigeria: a
cross-sectional assessment of compliance with the WHO guidelines”. In: BMJ Open 11.9 (2021),
e047835.issn: 2044-6055.doi: 10.1136/bmjopen-2020-047835.
[25] Günther Fink, Cesar G. Victora, Kenneth Harttgen, Sebastian Vollmer, Luís Paulo Vidaletti, and
Aluisio J. D. Barros. “Measuring Socioeconomic Inequalities With Predicted Absolute Incomes
Rather Than Wealth Quintiles: A Comparative Assessment Using Child Stunting Data From
National Surveys”. In: American Journal of Public Health 107.4 (2017), pp. 550–555.issn: 0090-0036.
doi: 10.2105/ajph.2017.303657.
[26] Priscilla Funsani, Hong Jiang, Xiaoguang Yang, Atupele Zimba, Thokozani Bvumbwe, and
Xu Qian. “Why pregnant women delay to initiate and utilize free antenatal care service: a
qualitative study in theSouthern District of Mzimba, Malawi”. In: Global Health Journal 5.2 (2021).
Special Issue: Reproductive Health for Global Women, pp. 74–78.issn: 2414-6447.doi:
https://doi.org/10.1016/j.glohj.2021.04.001.
[27] Anna D. Gage, Hannah H. Leslie, Asaf Bitton, J. Gregory Jerome, Jean Paul Joseph,
Roody Thermidor, and Margaret E. Kruk. “Does quality influence utilization of primary health
care? Evidence from Haiti”. In: Globalization and Health 14.1 (2018).issn: 1744-8603.doi:
10.1186/s12992-018-0379-0.
[28] General Assembly, United Nations. Resolution adopted by the General Assembly on 25 September
2015: Transforming our world: the 2030 Agenda for Sustainable Development. Tech. rep. New York,
USA: United Nations, 2015.url: https://sdgs.un.org/2030agenda.
[29] Government of the Republic of Malawi. Health Sector Strategic Plan II (2017-2022). Tech. rep.
Lilongwe, Malawi: GoM, 2017, p. 122.url:
https://www.usaid.gov/sites/default/files/documents/1864/MH%5C%20Strategy_web_red.pdf.
81
[30] A Metin Gülmezoglu, Theresa A. Lawrie, Natasha Hezelgrave, Olufemi T. Oladapo,
João Paulo Souza, Marijke Gielen, Joy E. Lawn, Rajiv Bahl, Fernando Althabe, Daniela Colaci, and
G Justus Hofmeyr. Chapter 7: Interventions to Reduce Maternal and Newborn Morbidity and
Mortality. Ed. by Robert E. Black, Ramanan Laxminarayan, Marleen Temmerman, and Neff Walker.
Washington, DC, USA: The International Bank for Reconstruction and Development / The World
Bank, 2016.
[31] Shivam Gupta, Goro Yamada, Rose Mpembeni, Gasto Frumence, Jennifer A. Callaghan-Koru,
Raz Stevenson, Neal Brandes, and Abdullah H. Baqui. “Factors Associated with Four or More
Antenatal Care Visits and Its Decline among Pregnant Women in Tanzania between 1999 and
2010”. In: PLOS ONE 9.7 (2014), e101893.issn: 1932-6203.doi: 10.1371/journal.pone.0101893.
[32] Jennifer Hall, Geraldine Barrett, Nicholas Mbwana, Andrew Copas, Address Malata, and
Judith Stephenson. “Understanding pregnancy planning in a low-income country setting:
validation of the London measure of unplanned pregnancy in Malawi”. In: BMC Pregnancy and
Childbirth 13.1 (2013), p. 200.issn: 1471-2393.doi: 10.1186/1471-2393-13-200.
[33] Judith A. Hall and Michael C. Dornan. “Patient sociodemographic characteristics as predictors of
satisfaction with medical care: A meta-analysis”. In: Social Science Medicine 30.7 (1990),
pp. 811–818.issn: 0277-9536.doi: https://doi.org/10.1016/0277-9536(90)90205-7.
[34] F Hierink, EA Okiro, A Flahault, and N Ray. “The winding road to health: A systematic scoping
review on the effect of geographical accessibility to health care on infectious diseases in low- and
middle-income countries”. In: PLOS ONE 16.1 (2021), e0244921.issn: 1932-6203.doi:
10.1371/journal.pone.0244921.
[35] C A Hjortsberg. “Cost of access to health services in Zambia”. In: Health Policy and Planning 17.1
(2002), pp. 71–77.issn: 1460-2237.doi: 10.1093/heapol/17.1.71.
[36] EL Idler and Y Benyamini. “Self-Rated Health and Mortality: A Review of Twenty-Seven
Community Studies”. In: Journal of Health and Social Behavior 38.1 (1997), pp. 21–37.doi:
10.2307/2955359.
[37] Elizabeth F. Jackson, Ayesha Siddiqui, Hialy Gutierrez, Almamy Malick Kanté, Judy Austin, and
James F. Phillips. “Estimation of indices of health service readiness with a principal component
analysis of the Tanzania Service Provision Assessment Survey”. In: BMC Health Services Research
15.1 (2015), pp. 1–8.issn: 14726963.doi: 10.1186/s12913-015-1203-7.
[38] Safia S Jiwani, Agbessi Amouzou-Aguirre, Liliana Carvajal, Doris Chou, Youssouf Keita,
Allisyn C Moran, Jennifer Requejo, Sanni Yaya, Lara Me Vaz, Ties Boerma, and et al. “Timing and
number of antenatal care contacts in low and middle-income countries: Analysis in the
Countdown to 2030 priority countries”. In: Journal of Global Health 10.1 (2020).issn: 2047-2978.
doi: 10.7189/jogh.10.010502.
[39] Te Jongh, I Gurol-Urganci, E Allen, N Jiayue Zhu, and R Atun. “Barriers and enablers to integrating
maternal and child health services to antenatal care in low and middle income countries”. In:
BJOG: An International Journal of Obstetrics Gynaecology 123.4 (2016), pp. 549–557.issn:
1470-0328.doi: 10.1111/1471-0528.13898.
82
[40] M Jylhä. “What is self-rated health and why does it predict mortality? Towards a unified
conceptual model”. In: Social Science Medicine 69.3 (2009), pp. 307–316.issn: 0277-9536.doi:
https://doi.org/10.1016/j.socscimed.2009.05.013.
[41] C Kambala, T Morse, S Masangwi, and P Mitunda. “Barriers to maternal health service use in
Chikhwawa, Southern Malawi”. In: Malawi Medical Journal 23.1 (2011).issn: 1995-7262.doi:
10.4314/mmj.v23i1.67673.
[42] Nanzen Caroline Kaphagawani and Ezekiel Kalipeni. “Sociocultural factors contributing to teenage
pregnancy in Zomba district, Malawi”. In: Global Public Health 12.6 (2017), pp. 694–710.doi:
10.1080/17441692.2016.1229354.
[43] Nicholas N.A. Kyei, Oona M.R. Campbell, and Sabine Gabrysch. “The Influence of Distance and
Level of Service Provision on Antenatal Care Use in Rural Zambia”. In: PLoS ONE 7.10 (2012).issn:
19326203.doi: 10.1371/journal.pone.0046475.
[44] Kenneth L . Leonard. “‘ACTIVE PATIENTS’ IN RURAL AFRICAN HEALTH CARE:
IMPLICATIONS FOR WELFARE, POLICY AND PRIVATIZATION”. In: (2003).
[45] Hannah H. Leslie, Günther Fink, Humphreys Nsona, and Margaret E. Kruk. “Obstetric Facility
Quality and Newborn Mortality in Malawi: A Cross-Sectional Study”. In: PLOS Medicine 13.10
(2016), e1002151.issn: 1549-1676.doi: 10.1371/journal.pmed.1002151.
[46] Michelle Li, Isabel Brodsky, and Eric Geers. Tech. rep. MEASURE Evaluation, Carolina Population
Center, University of North Carolina at Chapel Hill.
[47] W Luo. “Using a GIS-based floating catchment method to assess areas with shortage of
physicians”. In: Health and Place 10.1 (2004), pp. 1–11.issn: 1353-8292.doi:
https://doi.org/10.1016/S1353-8292(02)00067-9.
[48] Malawi Ministry of Health. Malawi Road Map for the Reduction of Maternal and Neonatal Mortality
and Morbidity. Tech. rep. Lilongwe, Malawi: Government of the Republic of Malawi, 2012, p. 53.
url: https://www.healthynewbornnetwork.org/hnn-content/uploads/Malawi-Roadmap-for-Reducing-
MN-mortality-2012.pdf.
[49] Malawi National Statistical Office. Malawi Population and Housing Census Report - 2018. Tech. rep.
Zomba, Malawi: Malawi National Statistical Office, 2019.
[50] Malawi National Statistical Office and ICF Macro. Malawi Demographic and Health Survey
2015-2016. Tech. rep. Zomba, Malawi: Malawi National Statistical Office and ICF Macro, 2017.
[51] Kondwani Chidzammbuyo Mamba, Adamson S Muula, and William Stones. “Facility-imposed
barriers to early utilization of focused antenatal care services in Mangochi District, Malawi – a
mixed methods assessment”. In: BMC Pregnancy and Childbirth 17.1 (2017), p. 444.issn: 1471-2393.
doi: 10.1186/s12884-017-1631-y.
83
[52] Martina Mchenga, Ronelle Burger, and Dieter Von Fintel. “Examining the impact of WHO’s
Focused Antenatal Care policy on early access, underutilisation and quality of antenatal care
services in Malawi: a retrospective study”. In: BMC Health Services Research 19.1 (2019).issn:
1472-6963.doi: 10.1186/s12913-019-4130-1.
[53] Abi Merriel, Michael Larkin, Julia Hussein, Charles Makwenda, Address Malata, and
Arri Coomarasamy. “Working lives of maternity healthcare workers in Malawi: an ethnography to
identify ways to improve care”. In: AJOG Global Reports 2.1 (2022), p. 100032.issn: 2666-5778.doi:
https://doi.org/10.1016/j.xagr.2021.100032.
[54] Ministry of Health (MoH) [Malawi] and ICF International. Service Provision Assessment (SPA)
2013-14. Tech. rep. Lilongwe, Malawi, and Rockville, Maryland, USA: MoH and ICF International,
2014.
[55] Ministry of Health and Social Welfare (MoHSW) [Tanzania Mainland], Ministry of Health (MoH)
[Zanzibar], National Bureau of Statistics (NBS), Office of the Chief Government Statistician
(OCGS), and ICF International. Tanzania Service Provision Assessment Survey (TSPA) 2014-15.
Tech. rep. Dar es Salaam, Tanzania, and Rockville, Maryland, USA: MoHSW, MoH, NBS, OCGS,
and ICF International, 2015.
[56] Ann-Beth Moller, Max Petzold, Doris Chou, and Lale Say. “Early antenatal care visit: a systematic
analysis of regional and global levels and trends of coverage from 1990 to 2013”. In: The Lancet
Global Health 5.10 (2017), e977–e983.issn: 2214-109X.doi: 10.1016/s2214-109x(17)30325-x.
[57] GF Nemet and AJ Bailey. “Distance and health care utilization among the rural elderly”. In: Social
Science Medicine 50.9 (2000), pp. 1197–1208.issn: 0277-9536.doi:
https://doi.org/10.1016/S0277-9536(99)00365-2.
[58] Robin C Nesbitt, Sabine Gabrysch, Alexandra Laub, Seyi Soremekun, Alexander Manu,
Betty R Kirkwood, Seeba Amenga-Etego, Kenneth Wiru, Bernhard Höfle, Chris Grundy, and et al.
“Methods to measure potential spatial access to delivery care in low- and middle-income countries:
a case study in rural Ghana”. In: International Journal of Health Geographics 13.1 (2014), p. 25.issn:
1476-072X.doi: 10.1186/1476-072x-13-25.
[59] Wingston Felix Ng’Ambi, Joseph H. Collins, Tim Colbourn, Tara Mangal, Andrew Phillips,
Fannie Kachale, Joseph Mfutso-Bengo, Paul Revill, and Timothy B. Hallett. “Socio-demographic
factors associated with early antenatal care visits among pregnant women in Malawi: 2004–2016”.
In: PLOS ONE 17.2 (2022), e0263650.issn: 1932-6203.doi: 10.1371/journal.pone.0263650.
[60] Abdisalan M Noor, Abdinasir A Amin, Peter W Gething, Peter M Atkinson, Simon I Hay, and
Robert W Snow. “Modelling distances travelled to government health services in Kenya”. In:
Tropical Medicine International Health 11.2 (2006), pp. 188–196.issn: 1360-2276.doi:
10.1111/j.1365-3156.2005.01555.x.
[61] Nawal M. Nour. “An introduction to maternal mortality”. In: Reviews in obstetrics gynecology 1.2
(2008), pp. 71–81.
84
[62] Nyasa Times Staff Writer. “Medical staff asked not to be rude, as Balaka nurses strike continues”.
In: Nyasa Times (2012).url: https://www.nyasatimes.com/medical-staff-asked-not-to-be-rude-as-
balaka-nurses-strike-continues/.
[63] Catherine E. Oldenburg, Ali Sié, Mamadou Ouattara, Mamadou Bountogo, Valentin Boudo,
Idrissa Kouanda, Elodie Lebas, Jessica M. Brogdon, Ying Lin, Fanice Nyatigo, and et al. “Distance to
primary care facilities and healthcare utilization for preschool children in rural northwestern
Burkina Faso: results from a surveillance cohort”. In: BMC Health Services Research 21.1 (2021).
issn: 1472-6963.doi: 10.1186/s12913-021-06226-5.
[64] OpenStreetMap via Geofabrik. Malawi road network. http://download.geofabrik.de/africa.html.
Accessed: 2021-07-07.
[65] Jessica Påfs, Aimable Musafili, Pauline Binder-Finnema, Marie Klingberg-Allvin, Stephen Rulisa,
and Birgitta Essén. “‘They would never receive you without a husband’: Paradoxical barriers to
antenatal care scale-up in Rwanda”. In: Midwifery 31.12 (2015), pp. 1149–1156.issn: 0266-6138.
doi: https://doi.org/10.1016/j.midw.2015.09.010.
[66] A. Paxton, D. Maine, L. Freedman, D. Fry, and S. Lobis. “The evidence for emergency obstetric
care”. In: International Journal of Gynecology Obstetrics 88.2 (2005), pp. 181–193.issn: 0020-7292.
doi: 10.1016/j.ijgo.2004.11.026.
[67] David H. Peters, Anu Garg, Gerry Bloom, Damian G. Walker, William R. Brieger, and
M. Hafizur Rahman. “Poverty and Access to Health Care in Developing Countries”. In: Annals of
the New York Academy of Sciences 1136.1 (2008), pp. 161–171.issn: 0077-8923.doi:
10.1196/annals.1425.011.
[68] Population Division of the Department of Economic and Social Affairs of the United Nations. 2018
Revision of World Urbanization Prospects. https://un.org/wup. Accessed: 2022-02-22.
[69] Joni Roberts, Helen Hopp Marshak, Diadrey Anne Sealy, Lucinda Manda-Taylor, Ron Mataya, and
Peter Gleason. “The Role of Cultural Beliefs in Accessing Antenatal care in Malawi: A Qualitative
Study”. In: Public Health Nursing 34.1 (2016), pp. 42–49.issn: 15251446.doi: 10.1111/phn.12242.
[70] Patrick J. Roohan, Scott J. Franko, Joseph P. Anarella, Laura K. Dellehunt, and Foster C. Gesten.
“Do commercial managed care members rate their health plans differently than medicaid managed
care members?” In: Health Services Research 38.4 (2003), pp. 1121–1134.issn: 00179124.doi:
10.1111/1475-6773.00166.
[71] Lale Say, Doris Chou, Alison Gemmill, Özge Tunçalp, Ann Beth Moller, Jane Daniels,
A. Metin Gülmezoglu, Marleen Temmerman, and Leontine Alkema. “Global causes of maternal
death: A WHO systematic analysis”. In: The Lancet Global Health 2.6 (2014), pp. 323–333.issn:
2214109X.doi: 10.1016/S2214-109X(14)70227-X.
[72] Andrew Self, Samuel Chipokosa, Amos Misomali, Tricia Aung, Steven A. Harvey,
Mercy Chimchere, James Chilembwe, Lois Park, Chrissie Chalimba, Edson Monjeza,
Fannie Kachale, Jameson Ndawala, and Melissa A. Marx. “Youth accessing reproductive health
services in Malawi: Drivers, barriers, and suggestions from the perspectives of youth and parents”.
In: Reproductive Health 15.1 (2018), pp. 1–10.issn: 17424755.doi: 10.1186/s12978-018-0549-9.
85
[73] Martha Priedeman Skiles, Clara R Burgert, Siân L Curtis, and John Spencer. “Geographically
linking population and facility surveys: methodological considerations”. In: Population Health
Metrics 11.1 (2013), p. 14.issn: 1478-7954.doi: 10.1186/1478-7954-11-14.
[74] Shoshanna Sofaer and Kirsten Firminger. “PATIENT PERCEPTIONS OF THE QUALITY OF
HEALTH SERVICES”. In: Annual Review of Public Health 26.1 (2005). PMID: 15760300, pp. 513–559.
doi: 10.1146/annurev.publhealth.25.050503.153958. eprint:
https://doi.org/10.1146/annurev.publhealth.25.050503.153958.
[75] Mariam Tanou and Yusuke Kamiya. “Assessing the impact of geographical access to health
facilities on maternal healthcare utilization: evidence from the Burkina Faso demographic and
health survey 2010”. In: BMC Public Health 19.1 (2019).issn: 1471-2458.doi:
10.1186/s12889-019-7150-1.
[76] Mariam Tanou, Takaaki Kishida, and Yusuke Kamiya. “The effects of geographical accessibility to
health facilities on antenatal care and delivery services utilization in Benin: a cross-sectional
study”. In: Reproductive Health 18.1 (2021), pp. 1–11.issn: 17424755.doi:
10.1186/s12978-021-01249-x.
[77] Andrew J. Tatem, Abdisalan M. Noor, Craig von Hagen, Antonio Di Gregorio, and Simon I. Hay.
“High Resolution Population Maps for Low Income Nations: Combining Land Cover and Census in
East Africa”. In: PLOS ONE 2.12 (Dec. 2007), pp. 1–8.doi: 10.1371/journal.pone.0001298.
[78] Teketo Kassaw Tegegne, Catherine Chojenta, Theodros Getachew, Roger Smith, and
Deborah Loxton. “Antenatal care use in Ethiopia: A spatial and multilevel analysis”. In: BMC
Pregnancy and Childbirth 19.1 (2019), p. V.issn: 14712393.doi: 10.1186/s12884-019-2550-x.
[79] Sandra C. Thompson, Shelley Cheetham, and Siddhartha Baxi. “The enablers, barriers and
preferences of accessing radiation therapy facilities in the rural developed world – a systematic
review”. In: BMC Cancer 17.1 (2017).issn: 1471-2407.doi: 10.1186/s12885-017-3790-7.
[80] Cecilie Skaarup Uldbjerg, Stine Schramm, Felix Ocaka Kaducu, Emilio Ovuga, and
Morten Sodemann. “Perceived barriers to utilization of antenatal care services in northern Uganda:
A qualitative study”. In: Sexual Reproductive Healthcare 23 (2020), p. 100464.issn: 1877-5756.doi:
10.1016/j.srhc.2019.100464.
[81] UNICEF and WHO. Tracking Progress towards Universal Coverage for Reproductive, Newborn and
Child Health: The 2017 Report. Tech. rep. Washington, DC, USA: United Nations Children’s Fund
(UNICEF) and the World Health Organization (WHO), 2017, p. 259.url:
https://www.countdown2030.org/pdf/Countdown-2030-complete-with-profiles.pdf.
[82] United Nations. Global SDG Indicator Platform. Tech. rep. United Nations, 2022.url:
https://sdg.tracking-progress.org/indicator/3-1-2-proportion-of-births-attended-by-skilled-
health-personnel/.
[83] United Nations Development Programme (UNDP). Human Development Index 2020.
https://hdr.undp.org/en/content/latest-human-development-index-ranking. Accessed: 2022-02-22.
86
[84] United Nations Maternal Mortality Estimation Inter-agency Group (MMEIG). Trends in Maternal
Mortality 2000 to 2017. Tech. rep. New York, USA: WHO, UNICEF, UNFPA, World Bank Group and
the United Nations Population Division, 2019.url:
https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/unpd_
unicef_level_and_trends_in_maternal_mortality_2000-2017-eng.pdf.
[85] United Nations Population Fund (UNFPA). Demographic Dividend: Malawi.
https://www.unfpa.org/data/demographic-dividend/MW. Accessed: 2022-02-22.
[86] United Nations, Department of Economic and Social Affairs, Population Division. World Population
Prospects 2019, Volume II: Demographic Profiles (ST/ESA/SER.A/427) . Tech. rep. New York, USA:
United Nations, 2019.url: https://population.un.org/wpp/.
[87] USAID. Ending Preventable Maternal Mortality: USAID Maternal Health Vision for Action Evidence
for Strategic Approaches. Tech. rep. Washington, DC, USA: US Agency for International
Development (USAID), 2009, p. 53.url:
https://www.usaid.gov/sites/default/files/documents/1864/MH%5C%20Strategy_web_red.pdf.
[88] Bradley H. Wagenaar, Kenneth Sherr, Quinhas Fernandes, and Alexander C. Wagenaar. “Using
routine health information systems for well-designed health evaluations in low- and
middle-income countries”. In: Health Policy and Planning 31.1 (2016), pp. 129–135.issn: 14602237.
doi: 10.1093/heapol/czv029.
[89] WJ Wang, R Winter, L Mallick, L Florey, C Burgert-Brucker, and E Carter. The relationship between
the health service environment and service utilization: linking population data to health facilities data
in Haiti and Malawi. Tech. rep. ICF International, 2015.url:
http://dhsprogram.com/pubs/pdf/AS51/AS51.pdf.
[90] WHO. The WHO application of ICD-10 to deaths during pregnancy, childbirth and puerperium: ICD
MM. Tech. rep. Geneva, Switzerland: World Health Organization, 2012, p. 67.url:
https://apps.who.int/iris/bitstream/handle/10665/70929/9789241548458_eng.pdf.
[91] WHO. WHO recommendations on antenatal care for a positive pregnancy experience. Tech. rep.
Geneva, Switzerland: World Health Organization, 2016, p. 196.url:
%7Bhttps://www.who.int/publications/i/item/9789241549912%7D.
[92] WHO, UNFPA, UNICEF, AMDD. Monitoring Emergency Obstetric Care: A Handbook. Tech. rep.
Geneva, Switzerland: World Health Organization, 2009, p. 152.url:
http://apps.who.int/iris/bitstream/handle/10665/44121/9789241547734_eng.pdf?sequence=1.
[93] World Health Organization. “Primary Health Care on the Road to Universal Health Coverage:2019
Monitoring Report”. In: (2019), p. 162.url:
https://www.who.int/healthinfo/universal_health_coverage/report/uhc_report_2019.pdf.
[94] WorldPop (www.worldpop.org - School of Geography and Environmental Science, University of
Southampton; Department of Geography and Geosciences, University of Louisville; Departement
de Geographie, Universite de Namur) and Center for International Earth Science Information
Network (CIESIN), Columbia University. Malawi Population 2014.
https://www.worldpop.org/geodata/summary?id=27395. Accessed: 2022-02-22. 2018.
87
[95] Donghua Zhou, Zhonghe Zhou, Cheng Yang, Lu Ji, Bishwajit Ghose, and Shangfeng Tang.
“Sociodemographic characteristics associated with the utilization of maternal health services in
Cambodia”. In: BMC Health Services Research 20.1 (2020).issn: 1472-6963.doi:
10.1186/s12913-020-05652-1.
88

Abstract (if available)

Abstract The overarching goal of this work was to characterize self-reported barriers to health care among rural Malawian women with the aim of better understanding what these barriers represent in terms of objective measures. This work was motivated by the substantive challenge of understanding why only half of pregnant women in Malawi obtained the WHO-recommended four or more antenatal care visits while more than 90\% of women got at least one visit in a way that went beyond linking outcomes to individual sociodemographic characteristics. This work also sought to demonstrate improved spatial methods of linking population health surveys and facility surveys to contribute health system perspectives in analyses of health outcomes.

The analyses presented here consistently demonstrate that perceived barriers are shown to reflect objective measures of their correlates. Chapter 2 demonstrates that women who report distance to be a major barrier to health care do, in fact, live farther from health facilities. Chapter 3 indicates that perceiving barriers to care moderates the utilization of pregnancy health care. Finally, Chapter 4 shows that women who report experiencing poor health system readiness live in areas of reduced skilled health provider density. The main methodological contribution of this body of work is the innovative demonstration of a spatial method of linking nationally-representative health and facility surveys.

The work presented here to characterize user perceptions of health care will ultimately allow for decision-makers to leverage this rich perspective to allocate resources strategically and make policies to improve the coverage, utilization, and impact of lifesaving health care.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Homeless encampments and access to Water, Sanitation, and Hygiene (WaSH) services in Los Angeles, CA

PDF

Integrating spatial visualization to improve public health understanding and communication

PDF

Incomplete data & insufficient methods: transgender population health research in the US

PDF

Barriers in care access for the marginalized individual: a provider’s role

PDF

Modeling patient access to point-of-care diagnostic resources in a healthcare small-world network in rural Isaan, Thailand

PDF

Covariance-based distance-weighted regression for incomplete and misaligned spatial data

PDF

Lead exposure and health outcomes in older adults

PDF

The Women and Girls Center for Maternal Health intergenerational approaches to address the Black maternal health crisis

PDF

Understanding the role of population and place in the dynamics of seasonal influenza outbreaks in the United States

PDF

Stigma-free pregnancy: a recruitment and retention strategy for healthcare systems to engage pregnant women with substance use disorder in collaborative care

PDF

Barriers to surgery for in low- and middle-income countries: a cross-sectional study of cleft lip and cleft palate patients in Vietnam

PDF

Health care utilization and spending of the U.S. aging population

PDF

Implementation of peer providers in integrated health care settings

PDF

Investigating daily effects of activity space-based built environment exposures on physical activity behaviors in Hispanic women during pregnancy and early postpartum

PDF

Spatiotemporal studies of out-of-hospital cardiac arrests and bystander cardiopulmonary resuscitation in Los Angeles

PDF

Uncertainty quantification in extreme gradient boosting with application to environmental epidemiology

PDF

Defining neighborhood for health research in Arizona

PDF

Implicit provider bias in cardiovascular disease care of Black women: the lives of Black women with heart disease matter

PDF

Personal PM2.5 exposure during pregnancy in an environmental health disparities population

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

Asset Metadata

Creator Park, Lois Aareum (author)

Core Title Characterizing self-reported spatial and provider barriers to maternal health care utilization in Malawi

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Health and Place,Population

Degree Conferral Date 2022-08

Publication Date 06/29/2022

Defense Date 06/10/2022

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

Tag antenatal care,barriers to health,GIS,Malawi,maternal health,OAI-PMH Harvest,pregnancy health,spatial analysis

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Chiang, Yao-Yi (committee chair), Franklin, Meredith (committee chair), Marjoram, Paul (committee member), Smith-Greenaway, Emily (committee member)

Creator Email lois.a.park@gmail.com,loispark@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111352082

Unique identifier UC111352082

Legacy Identifier etd-ParkLoisAa-10796

Document Type Dissertation

Format application/pdf (imt)

Rights Park, Lois Aareum

Type texts

Source 20220706-usctheses-batch-950 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

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

Repository Email cisadmin@lib.usc.edu