Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
THE MEDIATION ANALYSIS OF EDUCATION AND INCOME ON MAJOR CARDIOVASCULAR EVENTS
WIN KHAING
A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR
THE DEGREE OF DOCTOR OF PHILOSOPHY (CLINICAL EPIDEMIOLOGY)
FACULTY OF GRADUATE STUDIES MAHIDOL UNIVERSITY
2017
COPYRIGHT OF MAHIDOL UNIVERSITY
Thesis entitled
THE MEDIATION ANALYSIS OF EDUCATION AND INCOME ON MAJOR CARDIOVASCULAR EVENTS
............................................................. Mr. Win Khaing Candidate
............................................................. Assoc. Prof. Ammarin Thakkinstian, Ph.D. (Clinical Epidemiology & Community Medicine) Major advisor
............................................................ Assoc. Prof. Atiporn Ingsathit, M.D., Ph.D. (Clinical Epidemiology) Co-advisor
............................................................ Asst. Prof. Sakda Arj-Ong Vallibhakara, M.D., PhD. (Clinical Epidemiology) Co-advisor
....................................................... ........ ..................................................... Prof. Patcharee Lertrit, Assoc. Prof. Ammarin Thakkinstian, M.D., Ph.D. (Biochemistry) Ph.D. (Clinical Epidemiology & Dean Community Medicine) Faculty of Graduate Studies Program Director Mahidol University Doctor of Philosophy Program in Clinical Epidemiology Faculty of Medicine, Ramathibodi Hospital, Mahidol University
Thesis entitled
THE MEDIATION ANALYSIS OF EDUCATION AND INCOME ON MAJOR CARDIOVASCULAR EVENTS
was submitted to the Faculty of Graduate Studies, Mahidol University
for the degree of Doctor of Philosophy (Clinical Epidemiology) on
July 24, 2017 ……………………………………... Mr. Win Khaing Candidate
……………………………………...
Lect. Vijj Kasemsap, M.D., Ph.D. (Social & Administrative
Pharmacy) Chair ……………………………………... …………………………………… Assoc. Prof. Atiporn Ingsathit, Assoc. Prof. Ammarin Thakkinstian, M.D., Ph.D. (Clinical Epidemiology) Ph.D. (Clinical Epidemiology & Member Community Medicine) Member
……………………………………... ……………………………………... Assoc. Prof. Col. Nakarin Sansanayudh, Asst. Prof. Sakda Arj-Ong Vallibhakara, M.D., Ph.D. (Clinical Epidemiology) M.D., Ph.D. (Clinical Epidemiology) Member Member ……………………………………... ……………………………………... Prof. Patcharee Lertrit, Prof. Piyamitr Sritara, M.D., Ph.D. (Biochemistry) M.D., FRCPT, FACP, FRCP (T) Dean Dean Faculty of Graduate Studies Faculty of Medicine Ramathibodi Hospital Mahidol University Mahidol University
iii
ACKNOWLEDGEMENTS
I would like to express my sincerest gratitude to my major advisor, Assoc.
Prof. Dr. Ammarin Thakkinstain, for her valuable guidance, kind support, encouragement and very productive criticism in my PhD student life. I have learned many valuable lessons from her in terms of academics and professionalism in my short time here. Her high expectations for me have given me the confidence to pursue challenging topics, and, when I inevitably get stuck, she always is willing to make herself available to brainstorm and offer insight. She has a rare mix of brilliance, patience, kindness, genius, and generosity; I consider myself truly lucky to have her as a major advisor and a program director.
I am heartily thankful to my co-advisors, Asso. Prof. Dr. Atiporn Ingsathit, Asst. Prof. Dr. Sakda Arj-Ong Vallibhakara for providing me not only the excellence recommendations and intuitive ideas, but also giving me continuous emotional support and encouragement. I thank them for their patience and thoughtful insight. I also appreciate the dedication and continuing support on statistical analysis from Dr. Sasivimol Rattanasiri and Dr. Attawood Lertpimonchai in this project.
I wish to express my appreciation to Mr. Stephen John Pinder for his continuous support, encouragement, guidance, invaluable advice and diligence is also the sole reason why I have been able to complete this PhD course. I would like to show my gratitude to Ms. Paneevon Palakawong Na Ayutthaya, Ms. Sudasiri Sriwiang and all personnel of Section of Clinical Epidemiology and Biostatistics, Ramathibodi Hospital, Mahidol University for their grateful help and support especially about academic and administrative management of my project.
Thank you to my wonderful classmates Dr. Sudarat Eursiriwan, Dr. Teeranan Angkananard, Dr. Orawee Chinthakanan, Dr. Visasiri Tantrakul, Ms. Threechada Boonchan, and Ms. Sariya Udayachalerm for their support and for providing me with many good laughs throughout my PhD life.
Last but not least, I would like to express endless appreciation and deepest gratitude to my wife and my families for their infinite love, enduring support, persistent encouragement, and empathy throughout my study.
My PhD course was supported by Norwegian Scholarship Capacity Building for Institutes in Myanmar project which was co-funded by Norwegian Government in collaboration with Mahidol University.
Win Khaing
Fac. of Grad. Studies, Mahidol Univ. Thesis / iv
THE MEDIATION ANALYSIS OF EDUCATION AND INCOME ON MAJOR CARDIOVASCULAR
EVENTS
WIN KHAING 5736100 RACE/D
Ph.D. (CLINICAL EPIDEMIOLOGY)
THESIS ADVISORY COMMITTEE: AMMARIN THAKKINSTIAN, Ph.D., ATIPORN INGSATHIT, M.D.,
Ph.D., SAKDA ARJ-ONG VALLIBHAKARA, M.D., Ph.D.
ABSTRACT
Education and income are associated with major cardiovascular events (MCVE) but whether they are directly associated or they are mediators of each other is still unknown. This study is aimed to determine the direct and mediation effects of education through income on MCVE or vice versa.
Data from the prospective cohort of Electricity Generating Authority of Thailand 1985 (called EGAT1) were used. All subjects from the second (in 1997) to fifth survey of the EGAT1 cohort (EGAT1/2- EGAT1/5) were included in this study. The subjects were excluded if they had outcomes of interest before or at the date of enrollment in 1997. Education and income were study factors of interest whereas MCVEs were the outcomes, of interest i.e., composites of myocardial infarction, ischemic stroke/transient ischemic attack, and cardiovascular death; these were identified and confirmed by a team of specialists. A total of 2,967 subjects enrolled in EGAT1/2 and 2,360, 1,958 and 1,609 of them participated with the survey EGAT 1/3, 1/4 and 1/5, respectively with a total 3,025 subjects who enrolled in either of these in EGAT1. Among them, 28 subjects had MCVEs before enrolment of EGAT1/2, leaving 2,997 subjects included in this study. The causal relationship pathway of education → income → MCVE was assessed using generalized structural equation. A bootstrapping with 1000-replications was used to estimate the potential mediation effects, i.e., low education through low income, low education through medium income, medium education through low income, and medium education through medium education on MCVEs when compared with high education and high income. Odds ratios of these corresponding effects were 8.95 (95%CI: 4.19, 19.56), 2.17 (95%CI: 1.50, 3.20), 3.47 (95%CI: 2.24, 5.61), and 1.46 (95%CI: 1.22, 1.80), respectively. The direct effects of low and medium education versus high education on MCVE were not significant with coefficient 0.06 (95%CI: -0.42, 0.59) and 0.25 (95%CI: -0.19, 0.69), respectively.
This study provided evidence education was not directly associated with MCVE but it indirectly affected on MCVEs through income.
KEY WORDS: CARDIOVASCULAR EVENTS / EDUCATION / GENERALIZED STRUCTURAL
EQUATION MODELLING / INCOME / MEDIATION ANALYSIS
170 pages
v
CONTENTS
Page
ACKNOWLEDGEMENTS iii
ABSTRACT iv LIST OF TABLES viii
LIST OF FIGURES x
LIST OF ABBREVIATIONS xii
CHAPTER I BACKGROUND AND RATIONALE 1
1.1 Background and rationale 1
1.2 Rationale 3
1.3 Research questions 4
1.4 Research objectives 4
CHAPTER II LITERATURE REVIEW 5
2.1 Epidemiologic transition of cardiovascular diseases 5
2.2 Impact of social determinants of health on cardiovascular
diseases 6
2.3 Socioeconomic status and cardiovascular diseases 7
2.4 Effects of education and income on cardiovascular outcomes:
Systematic review and meta-analysis 9
2.5 The association between education/income and cardiovascular
risk factors 18
2.6 Causal pathways between education/income and
cardiovascular diseases 19
2.7 Conceptual framework 20
CHAPTER III METHODOLOGY 64
3.1 Study design and setting 64
3.2 Study subjects 64
3.3 Study factors and measurements 65
3.4 Outcome of interest 65
vi
CONTENTS (cont.)
Page
3.5 Others risk factors and measurements 67
3.6 Data collection 70
3.7 Sample size estimation 71
3.8 Data management 72
3.9 Imputation 74
3.10 Data analysis 76
3.11 Ethics considerations 84
CHAPTER IV RESULTS 96
4.1 Characteristics of subjects 96
4.2 Imputation results 97
4.3 Education → Income → MCVE pathway 97
4.4 Income → Education → MCVE pathway 100
CHAPTER V DISCUSSION 119
5.1 Main findings 119
5.2 Income measurement 119
5.3 Education measurement 120
5.4 Causal relationship pathway between education and MCVE
through income 120
5.5 Education, income and sustainable development goals 122
5.6 Multiple imputation 123
5.7 The use of GSEM model 123
5.8 Strengths of this study 125
5.9 Limitations of this study 126
5.10 Clinical application 127
5.11 Suggestion for further studies 127
5.12 Conclusion 128
REFERENCES 130
vii
CONTENTS (cont.)
Page
APPENDICES 152
Appendix A Serarch terms and serarch strategy used 152
Appendix B Newcastle-Ottawa quality assessment scale (cohort
studies) 155
Appendix C Commands used for multiple imputation 157
Appendix D Commands used for medation analysis 165
Appendix E Diagnostics plot between missing and observed values 168
Appendix F Ethical approval 169
BIOGRAPHY 170
viii
LIST OF TABLES
Table Page
2.1 Characteristics of included studies 21
2.2 Risk of bias assessment of included studies 41
2.3 Estimations of pooled effects of education and income on
cardiovascular outcomes (co-variates adjusted studies only) 46
2.4 Pooled education and income effects on cardiovascular outcomes by
regions 47
2.5 Pooled education and income effect on coronary artery diseases
(subgroup analyses) 47
2.6 Pooled education and income effect on cardiovascular events
(subgroup analyses) 49
2.7 Pooled education and income effect on strokes (subgroup analyses) 50
2.8 Pooled education and income effect on cardiovascular deaths
(subgroup analyses) 51
3.1 Example of 3 subjects who had inconsistent, missing data and out-of-
range data for sex and height 86
3.2 Example of 3 subjects who had inconsistent data for education,
marital status, smoking status and alcohol consumption 87
3.3 Multiple imputation model per variable with their selected variables 89
4.1 Baseline characteristics of subjects included by EGAT periods 101
4.2 Report of number of missing data 103
4.3 Comparison of characteristics of subjects between original dataset and
imputed dataset 106
4.4 Mediation analysis of education and income (Univariate Analysis) 108
4.5 Mediation analysis of education and income (Multivariate Analysis) 109
4.6 Mediation analysis of MCVE (Univariate Analysis) 110
4.7 Mediation analysis of MCVE (Multivariate Analysis) 111
ix
LIST OF TABLES (cont.)
Table Page
4.8 Mediation analysis of Education on MCVE that was mediated by
income (bias-corrected bootstrapped) 112
4.9 Causal effects of low education on MCVE through low income (bias-
corrected bootstrapped) 113
4.10 Causal effects of low education on MCVE through medium income
(bias-corrected bootstrapped) 114
4.11 Causal effects of medium education on MCVE through low income
(bias-corrected bootstrapped) 115
4.12 Causal effects of medium education on MCVE through medium
income (bias-corrected bootstrapped) 116
x
LIST OF FIGURES
Figure Page
2.1 Flow diagram for selection of studies 52
2.2 Pooling effects of education on cardiovascular outcomes 53
2.3 Funnel plots of relative risks of cardiovascular outcomes among
medium versus high education levels 54
2.4 Funnel plots of relative risks of cardiovascular outcomes among low
versus high education levels 55
2.5 Contour-enhanced plots of relative risks of cardiovascular outcomes
among medium versus high education levels 56
2.6 Contour-enhanced plot of relative risks of cardiovascular outcomes
among low versus high education levels 57
2.7 Pooling effects of income on cardiovascular outcomes 58
2.8 Funnel plots of relative risks of cardiovascular outcomes among
medium versus high income levels 59
2.9 Funnel plots of relative risks of cardiovascular outcomes among low
versus high income levels 60
2.10 Contour-enhanced plots of relative risks of cardiovascular outcomes
among medium versus high income levels 61
2.11 Contour-enhanced plots of relative risks of cardiovascular outcomes
among low versus high income level 62
2.12 Conceptual framework. Direct effect shown in solid line, and
mediated effect shown in dashed line 63
3.1 Time frame of first EGAT cohort and follow-up 90
3.2 Data management flow diagram 91
3.3 A causal effect pathway of education → income → MCVE 92
3.4 Generalized structure equation model of causal effect pathway of
education → income → MCVE 93
xi
LIST OF FIGURES (cont.)
Figure Page
3.5 A causal effect pathway of income → education → MCVE 94
3.6 Generalized structure equation model of causal effect pathway of
income → education → MCVE 95
4.1 Causal mediation pathway diagram of the relationship among
education, income and MCVE using generalized structural equation
modelling 117
4.2 Direct effects of education on MCVE and mediated effects of and
percentage of mediation from effect of education on MCVE through
low income 118
xii
LIST OF ABBREVIATIONS
ACS Acute coronary syndrome
AHA American Heart Association
AMI Acute myocardial infraction
ASA American Stroke Association
BMI Body mass index
BP Blood pressure
CAD Coronary artery disease
CBC Complete blood count
CHD Coronary heart disease
CI Confidence interval
CKD Chronic kidney disease
CT Computerized tomography
CVD Cardiovascular disease
CVE Cardiovascular event
CVRF Cardiovascular risk factors
CXR Chest X-ray
DBP Diastolic blood pressure
DE Direct effect
dL Deciliter
DM Diabetes mellitus
EGAT Electricity Generating Authority of Thailand
EKG Electrocardiography
exp Exponential
FMI Fraction of missing information
FPG Fasting plasma glucose
GSEM Generalized structural equation modelling
HDL High density lipoprotein
xiii
LIST OF ABBREVIATIONS (cont.)
HF Heart failure
HR Hazard ratio
HT Hypertension
IHD Ischemic heart disease
LDL Low density lipoprotein
LMICs Low and middle income countries
MAR Missing at random
MCVE Major cardiovascular event
ME Mediation effect
Mesh Medical subject headings
mg Milligram
MI Myocardial infraction
MICE Multiple imputation with chained equations
mmHg Millimeter mercury
MRI Magnetic resonance imaging
NCD Non-communicable disease
NSAIDs Non-steroidal anti-inflammatory drugs
OR Odds ratio
PRISMA Preferred Reporting Items for Systematic Reviews and Meta-
analysis
PROSPERO International prospective register of systematic reviews
RR Risk ratio
RVI Relative variance increases
SBP Systolic blood pressure
SD Standard deviation
SDG Sustainable development goal
SDH Social determinants of health
xiv
LIST OF ABBREVIATIONS (cont.)
SE Standard error
SEM Structural equation modelling
SES Socioeconomic status
SMK Smoking
TC Total cholesterol
TE Total effect
TG Triglyceride
TIA Transient ischemic attack
UHC Universal health coverage
US United State
var Variance
WHR Waist-to-hip ratio
WHO World health organization
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 1
CHAPTER I
BACKGROUND AND RATIONALE
1.1 Background and rationale
1.1.1 Non-communicable diseases and cardiovascular diseases burden
Of the global mortality in 2012, 38 out of 56 million (68%) were caused by
non-communicable diseases (NCD)1. A global epidemic of NCD strikes hardest to low-
and middle-income countries (LMICs) including Asian countries which accounts for
almost three quarters (28 million) of global NCD deaths1. The World Health
Organization (WHO) forecasted that NCD deaths are projected to rise to 52 million in
20301, which will account for more than two thirds of global mortality. Of these, 82%
would be cardiovascular diseases (CVD) were followed by cancers, respiratory diseases,
and diabetes. CVD is a major public health problem that accounts for about 30% of the
annual global mortality and 10% of the global disease burden1.
1.1.2 Cardiovascular disease risk factors
The Framingham Heart Study2, the WHO-MONICA Project3, and the
INTERHEART4 studies reported evidences for the major risk factors of CVD, which
could be classified as demographic (e.g., age, sex, race, family history, and etcetera),
behavioral (e.g., smoking, alcohol consumption, physical inactivity, dietary, and
etcetera) and metabolic (body mass index, blood glucose, cholesterol level, and etcetera)
domains5-7. Modification of these risk factors should lead to reduced cardiovascular
morbidity and mortality. Despite much effort invested in primary and secondary
prevention of CVD, it is still a problem in industrialized and high income countries, as
well as in LMICs1. Understanding of these risk factors is critical to the prevention of
cardiovascular morbidity and mortality. In addition, nontraditional markers (e.g., high-
sensitivity C-reactive protein8, lipoprotein(a)9, homocysteine10, small dense low-density
Win Khaing Background and Rationale / 2
lipoprotein-C particles11, fibrinogen12, and etcetera) were also identified with advanced
investigations.
1.1.3 Cardiovascular diseases and social determinants of health
Recently, the fifth epidemiological transition proposed that social
upheaval13 might break down the existing social and health structures leading to
increased CVD morbidity and mortality. The impacts of these consequences caused
much concern to all societies and economies, and were particularly devastating in poor
and vulnerable LMICs populations. Since then, many social determinants of health
(SDH, e.g., education, income, etcetera) have been increasingly considered and should
be included in a causal pathway with other traditional risk factors and markers of CVD.
Many studies show that SDH indirectly influence CVD through behavioral
and metabolic cardiovascular risk factors (CVRF), psychosocial factors and
environmental living condition14, 15. Some landmark studies16-18 and numerous other
epidemiological studies19-22 show an inverse relationship between SDH and CVD
morbidity and mortality. For instance, low educated persons were more likely to have
CVRF (e.g., hypertension, diabetes, dyslipidemias, overweight, smoking and sedentary
lifestyle), and have less healthy dietary habits than high educated persons23-25. Evidence
also showed that lower education was directly associated with atherosclerosis, ischemic
heart disease, cerebrovascular diseases, CVD mortality and all-cause mortality26-28.
Similar to education, the inverse relationship of income on ischemic heart disease
(IHD), coronary events, pre-hospital coronary death and CVD mortality has also been
reported29-33. These effects of education and income were more consistent in developed
countries, but results were still inconclusive in LMICs34, 35
1.1.4 Effects of education/income on cardiovascular diseases
A number of narrative36-38 and systematic reviews39-41 were conducted to
assess the relationship of socioeconomic status (SES) with CVD including myocardial
infarct (MI), strokes, heart failure (HF), and death. Effects of education and income on
MI40 and CVD mortality41 were pooled mainly based on studies from developed
countries with a few studies from LMICs. In both studies, education and income were
roughly categorized as low versus high, resulting in an inability to assess SES gradients.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 3
We therefore conducted a systematic review and meta-analysis42 to pool effects of
education and income on various cardiovascular outcomes by including more studies
conducted in developing countries and also classifying education and income into low,
middle, and high groups. A total of 72 studies were included, most of them were from
high income countries (93.1%), which mainly were from the European region (54.2%),
and only 19.4% were from the Asian regions. Our findings indicated that low to middle
education and income carried higher risks of coronary artery diseases (CAD),
cardiovascular events (CVE), strokes and cardiovascular deaths when compared to high
education and income. Comparing medium and low versus high education groups,
pooled relative risks (RRs) were 1.21 (1.06, 1.40) and 1.36 (1.11, 1.66) for CAD, 1.27
(1.09, 1.48) and 1.50 (1.17, 1.92) for CVE, 1.17 (1.01, 1.35) and 1.23 (1.06, 1.43) for
strokes, and 1.21 (1.12, 1.30) and 1.39 (1.26, 1.54) for cardiovascular deaths. Pooled
RRs for medium and low versus high income groups were 1.27 (1.10, 1.47) and 1.49
(1.16, 1.91) for CAD, 1.05 (0.98, 1.13) and 1.17 (0.96, 1.44) for CVE, 1.24 (1.00, 1.53)
and 1.30 (0.99, 1.72) for strokes, and 1.34 (1.17, 1.54) and 1.76 (1.45, 2.14) for
cardiovascular deaths. Evidence from the Asian region were still lacking, especially, in
association of income with CVD outcomes.
1.2 Rationale Results of our systematic review42 indicated that education and income were
associated with CVD outcomes. Previous evidences showed that education itself was
also highly associated with income43, 44 or vice versa45, 46, i.e., low education led to low
income, and both may increase the risk of CVD. A question was raised whether
education is directly associated with CVD or its effect is passed through income, or vice
versa. However, our systematic and previous reviews could only answer direct effects
of education and income on CVD, but not for a causal relationship pathway. There was
still lack of empirical evidences for analyzing the causal pathways between
education/income and CVD outcomes, especially in Asian countries.
In order to answer these questions, a large-scale cohort which has sufficient
power to adjust for all the known CVRF and follow-up long enough to observe for CVD
outcomes was necessary, especially in Asian countries. Therefore, this study was
Win Khaing Background and Rationale / 4
conducted using data from the Electricity Generating Authority of Thailand (EGAT)
prospective cohort with pre-specified major cardiovascular events (MCVE) as primary
outcome by following research questions and objectives.
1.3 Research questions Does education directly associate with MCVEs or is its effect mediated
through income, or vice versa?
1.4 Research objectives The objectives of the study were:
1.4.1 To determine the direct and mediation effects of education through
income on MCVEs
1.4.2 To determine the direct and mediation effects of income through
education on MCVEs
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 5
CHAPTER II
LITERATURE REVIEW
2.1 Epidemiologic transition of cardiovascular diseases A global pattern of morbidity and mortality of CVD has been observed over
time. Historically, CVD are mainly concerned with the cause of infections like
rheumatic fever and syphilis on the heart and cardiomyopathies due to malnutrition, and
death from CVD accounts only less than 10%13, 47. During the second stage of
epidemiologic transition, with more advances in societies, major causes of CVD shifted
from a predominance of infection and nutritional causes to chronic degenerative causes
like diseases related to hypertension, such as hemorrhagic stroke and hypertensive heart
disease with deaths attributed to CVD increased up to 35%13, 47, 48. Because of life
expectancy improvement, during the third stage of transition, CVD related with poor
habits such as cigarette smoking, high-fat diets and sedentary lifestyles have become
more common. CVD, most frequently IHD and atherosclerotic thrombotic stroke
became prominent especially at ages below 50 years. Not surprisingly, CVD deaths
continued to rise from 35% to 65% of overall deaths13. With increased efforts to earlier
diagnose, treat promptly and understand more about preventive measures, CVD had
become able to delay to more advanced ages of CVD during the fourth stage. Therefore,
the relationships between CVD and CVRF such as old age, CVD in family history, high
blood pressure (BP), tobacco smoking, unhealthy diet, alcohol consumption, overweight
and obesity, diabetes, physical inactivity, and dyslipidemia have been extensively
explored by many researchers during that periods.
More recently, a fifth epidemiological transition was proposed and was
called “age of health regression and social upheaval”13. We are, in turn, facing
resurgence of conditions seen in the first two stages and also diseases of the third and
fourth stages still persist. Social upheaval or war breaks down the existing social and
health structures, leading to increased deaths due to both cardiovascular and non-
cardiovascular causes such as infectious diseases, violence, accidents. Accordingly,
Win Khaing Literature Review / 6
now, many researchers have suggested that SDH should not only be put together with
traditional risk factors acting directly on CVD, but also be examined as underlying
determinants of some CVD. Actually, these social risk factors might act along causal
chains, influencing the incidence and management of conventional risk factors. Many
researchers became increasing awareness that different socioeconomic factors could
affect health at different times in the life cycle, effective at different levels and through
different loop of causal pathways.
2.2 Impact of social determinants of health on cardiovascular diseases WHO defined the term SDH49 as “the conditions in which individuals are
born, grow, live, work and age, which are shaped by the distribution of money, power
and resources at global, national and local levels”. WHO comprehended that the SDH
was the most important issue of unfair and avoidable “health inequities” between groups
of people within countries and between countries. Social and economic conditions can
effect on their lives and determine their risk of illness and their decisions to prevent
them becoming ill or treat illness when it happens.
In order to push towards progressive achievement of universal health
coverage (UHC), health inequities need to be reduced, and both SDH and UHC need to
be take action in an integrated and systematic manner.
SDH (e.g., the level of education, income, race, ethnicity, culture and
language, health care system, working conditions, employment and job stability,
residential environment and social support or social network) directly or indirectly
influenced CVD by impacting behavioral and metabolic cardiovascular risk factors,
psychosocial factors and environmental living conditions14, 15. The Whitehall study16
and Whitehall II study18 were well known studies that showed evidence of an inverse
association between SDH and CVD morbidity and mortality. The Evans County
Study19, the US National Longitudinal Mortality Study20, the Charleston Heart Study21
and the Alameda Country Study22 also showed the similar trends. Work-related stress
and depression were found to be associated with hypertension and arthrosclerosis.
Negative social relationship was found to be linked with increased BP. The poor have
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 7
limited choice of healthy lifestyle and health care access which may explain the link
between SES and CVD14, 15.
2.3 Socioeconomic status and cardiovascular diseases SES has been widely accepted as the most powerful SDH. Three common
measures of SES, i.e., education, income and occupation have been extensively explored
with regard to their relationship to cardiovascular health. In general, lower SES is
associated with a higher prevalence of CVD risk factors and a greater incidence of
mortality resulting from CVD37, 50, 51. Researchers showed low education and income
are associated with higher mortality from coronary heart disease52-54. Socioeconomic
gradient in stroke is also greatly influenced by traditional stroke risk factors like
diabetes, hypertension and alcohol abuse55, 56.
Education is the most widely accepted measurement of SES because it is
relatively easy to obtain, more likely to respond their education level, and has less recall
bias as people tends to remember their education level accurately. It is also usually fixed
in late childhood or early adulthood, precedes health outcomes, compared to income
which is far more likely to change over the life cycle.
Since, education shapes future earning potential and occupational
opportunities, higher education provides individuals with higher income, and provides
better knowledge and life skills to get more access to information and resources to
promote health. Therefore, better-educated people can utilize better healthcare resources
and healthier foods, and can also enable more leisure time for exercise. Higher level of
education and income also tend to stand in higher social class, status and social
network57 from which many positive benefits can be gained like beneficial behavioral
norms, positive materials and emotional support.
Measurement of education is generally favored to use years of schooling,
but not reflect difference in school prestige or resources, which may affect to differences
in future earnings. Measurement of income at individual, family, and community levels
remains a great challenge. In comparison, education is typically established in early
adulthood and remains stable throughout the life, whilst income is dynamic and might
Win Khaing Literature Review / 8
change extensively from early adulthood to middle-age and then into retirement and late
old age.
2.3.1 Education and cardiovascular diseases
Lower levels of educational attainment are associated with a higher
incidence of CVD, higher prevalence of cardiovascular risks and greater cardiovascular
mortality26, 37. Tromsø Heart Study23 in a 12,368 Norwegians cohort found that higher
educated persons were less likely to smoke, likely to be overweight, but more physically
active and had a healthier diet. Educational differences in ischemic heart disease,
cerebrovascular diseases and CVD mortality in the US and 11 Western European
countries was studied by Mackenbach et al26 who found that lower education individuals
have higher mortality with inequality in smoking and excessive alcohol consumption in
all countries. A study27, which included 308 asymptomatic women from the Healthy
Women Study has shown lower education was associated with greater early stage
atherosclerosis. Stanford Five City Project which included 2,380 subjects also showed
a consistent trend between lower educational attainment and high exposure to
cardiovascular risk factors. Cirera et al24 studied cardiovascular risk factors and
educational attainment in 3091 Spanish adults, and showed that people without
schooling have two to three times higher prevalence of overweight in women and
hypertension in both genders when compared to people at the university level. Recently,
the ATTICA study25 showed that people with low education were 52% higher risk of
developing CVD compared to those with high education. Low educated people were
also higher prevalence of hypertension, diabetes and dyslipidemias and more likely to
be smokers and sedentary, with less healthy dietary habits. The GREECS longitudinal
study28 also shown that all-cause mortality was 2 times higher in low education group
as compared to medium and high education groups (40%, vs. 22% and 19%,
respectively, p<0.001). Reversed association in developing countries has also been
observed elsewhere. A study conducted by Fernald & Alder34 in Mexico found that
educational attainment showed an inverse association with systolic BP in low-income
rural women.
2.3.2 Income and cardiovascular diseases
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 9
In parallel with education, many studies documented the association of
income and cardiovascular outcome. Andersen et al29 studied income and risk of IHD
in 22,782 people in Nordic countries. Inverse effect of income on IHD was seen with
hazard ratio for highest versus lowest deciles of income of 0.53 (95%CI: 0.44, 0.65).
The FINMONICA study30 reported low-income men have 2 times higher risk of pre-
hospital coronary death compared to high-income men and 1-year mortality rate was
also significantly higher in low-income patients in those who survived after MI. The
FINAMI study31 also showed that lower income people are 5.21 times and 11.13 times
more likely to develop coronary events than higher income people among 35 to 64 year-
old men and women, respectively. Similar findings were also reported by Alter et al.32
and Rao et al.33, with opposing, reverse findings which were reported from a study35 in
China. This Chinese study found that people with higher family average income were
1.94 times more likely to develop strokes compared to those with lower average family
income after adjustment for demographic and traditional risk factors. A review by
Harper et al also agreed that there is no evidence of consistent associations between
income inequality and prevalence of CVD risk factors and outcomes.
2.4 Effects of education and income on cardiovascular outcomes:
Systematic review and meta-analysis
2.4.1 Methods
The review protocol has been registered with the international prospective
register of systematic review (PROSPERO number CRD42016046615).
2.4.1.1 Search strategy
Relevant studies were identified from MEDLINE and Scopus
databases since inception to 30th July 2016. Titles and abstracts were screened, and full
articles were read if decision of selection could not be made. Reference lists were also
checked for studies that were not identified by our searching. The following search
terms were used for MEDLINE: "Cardiovascular Disease"[Mesh], "cardiovascular
event", "Myocardial Infraction"[Mesh], "Heart Failure"[Mesh], "Ventricular Function,
Left"[Mesh], "Coronary Disease"[Mesh], "Coronary Restenosis"[Mesh], "restenosis",
Win Khaing Literature Review / 10
"re-stenosis", "coronary flow", "coronary blood flow", "ejection fraction", "stroke",
"cardiovascular death", "cardiovascular mortality", education[Mesh], "education
status"[Mesh], "education level" and income[Mesh]. Search strategies for both
databases are described in Appendix A.
2.4.1.2 Selection of studies
Retrospective or prospective cohorts published in English were
selected if they met the following criteria: assessed associations between
education/income and cardiovascular outcomes in either a general or specific types of
adult population; measured education (either education years/groups) or income in
terms of money or in category; had at least one of outcome of interest (i.e., coronary
artery diseases (CAD), cardiovascular events (CVE), strokes and cardiovascular
deaths); had contingency data between education/income and cardiovascular outcomes,
or a beta-coefficient. Studies were excluded from the review if data for education and
income were combined; income was assessed based on ownership of car/house/health
insurance/zip-code. In cases of missing data, we made 3 attempts to contact authors to
request additional data.
2.4.1.3 Study factors
Education and income were our study factors; which were
assessed and reported differently across studies. To standardize data for pooling across
studies, they were re-categorized into 3 groups as low, medium, and high for education
years ≤ 9 (i.e., illiteracy/ no education/ basic/ primary education), 10 – 12 (i.e.,
secondary/ high school/ medium/ technical/ apprenticed/ trade/ vocation), and > 12
years (i.e., university/ college/ associates/ master/ professional/ PhD), respectively. In
addition, income expressed in other currencies was converted to US currency/year using
the reported exchange rates or the exchange rate at the time of study publication by using
online currency converter58. The salary income was re-categorized as ≤20,000, 20,001
to 40,000, and >40,000 US$ for low, medium, and high, respectively. If the original
studies reported income as quartile, these were re-categorized: 1st = low, 2nd = medium,
and 3rd + 4th quartiles = high income. If studies reported income as quintile, these were
re-categorized: 1st + 2nd = low, 3rd = medium, and 4th + 5th quintiles = high, respectively.
2.4.1.4 Outcomes
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 11
The outcomes of interest were CVD including CAD (e.g., acute
myocardial infarct (AMI), IHD, coronary heart disease (CHD)), CVE (e.g., HF, hospital
admission due to cardiac causes, revascularization and composite CVD, e.g., IHD or
strokes), strokes (ischemic or hemorrhagic strokes), and cardiovascular deaths. These
were defined according to the original studies.
2.4.1.5 Data extraction
Two reviewers (WK and SV) independently extracted general
information (the first author’s last name, the publication year) and characteristics of
studies/patients (i.e., study country, mean age, gender, mean body mass index (BMI),
diabetes mellitus, physical activity, smoking, alcohol drinking, hypertension,
dyslipidemia and chronic renal failure). In addition, education and income and type of
outcomes were also extracted. Furthermore, cross-tabulated data between
education/income groups and individual outcome were extracted for pooling. Summary
statistics (e.g., odds ratio, risk ratio, or hazard ratio) along with its 95% confidence
interval (CI) were extracted instead, if frequency data were not reported. Authors were
contacted if insufficient data were insufficient. Data entry, cleaning and cross check
validations were performed separately for each study. Entries were compared for
accuracy and any disagreements were solved by consensus.
2.4.1.6 Risk of bias assessment
The quality of studies were independently assessed by two
reviewers (WK and SV) using the Newcastle and Ottawa risk of bias criteria (see
Appendix B). The following three domains were evaluated, i.e., selection of study
groups (4 subdomains), comparability of groups (2 subdomains) and ascertainment of
exposure and outcome (3 subdomains). Each was graded as 0 to 1 with a total grade
ranging from 0 to 9. A total grade of seven or more was regarded as higher quality or
lower risk of bias.
2.4.1.7 Statistical analysis
Relative risks (RR) of having each outcome between low versus
high (RR1) and medium versus high (RR2) education/income groups were recalculated
from frequency data for studies whose frequency data were available. These were then
appended with reported summary statistics where frequency data were not available. A
multivariate random-effect meta-analysis59 was applied for pooling two RRs
Win Khaing Literature Review / 12
simultaneously. Variance-covariance between RR1 and RR2 was assumed to be zero for
those studies reporting summary RRs without frequency data.
Heterogeneity and degree of heterogeneity were assessed by
Cochrane’s Q test and I-squared statistic, respectively. Heterogeneity was considered to
be present if the p value of Q test was <0.1 or I-squared ≥25%.
Subgroup analyses were conducted to examine potential sources
of heterogeneity by fitting each of the co-variables (i.e., country, country income level60,
number of co-variables adjustment, age group, BMI, percentage of males, diabetes,
obesity, hypertension, high physical activity, smoking, alcohol drinking, dyslipidemia
and chronic kidney disease) in a multivariate meta-regression model.
Finally, exploration of potential publication bias was visualized
using a funnel plot and Egger's test. If any of these indicated asymmetry, a contour
enhanced funnel plot was constructed to distinguish whether the cause of the asymmetry
was due to publication bias or heterogeneity.
All analyses were performed using Stata61 version 14.2. P-
values <0.05 were considered as statistically significant, except for the test of
heterogeneity where p <0.10 was used.
2.4.2 Results
We identified 354 and 1335 studies from MEDLINE and Scopus databases
with 11 additional studies from reference lists. Of these 1700 studies, 115 were
duplicates, leaving 1585 studies to be screened. After screening titles and abstracts, 1399
studies did not address our primary question, leaving 72 studies for inclusion. Reasons
for exclusion of the studies are presented in Figure 2.1 following the Preferred Reporting
Items for Systematic Review and Meta-analysis (PRISMA) guideline.
2.4.2.1 General Characteristics of included studies
Characteristics of the 72 included cohorts published between
1982 and 2016 are shown in Table 2.1. Among them, 14, 39 and 19 studies were
conducted in Asia, Europe, and the United States, respectively. Most studies were from
high-income countries (93.1%); mean age and mean BMI ranged from 38.5 to 78 years
and 23.02 to 30.33 kg/m2, respectively. Percentage of males and proportion with
diabetes, smoking and hypertension varied from 35.9% to 78%, 1.3% to 42%, 7.28% to
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 13
72.64%, and 6.25% to 72.5% respectively. Thirty-three studies assessed association
between education and cardiovascular outcomes, 10 studies assessed effect of income,
and 29 studies assessed effects of both education and income, with a sample size ranging
from 128 to 4,157,202.
2.4.2.2 Risk of bias assessment
Results of the “risk of bias” assessment of the included studies
are shown in Table 2.2. Total scores ranged from 5 to 9 with a median of 7. Among the
included studies, 45 out of 72 (62%) had low risk of bias and 27 out of 72 (38%) had
high risk of bias.
2.4.2.3 Education and cardiovascular outcomes
A total of 62 studies assessed association between education and
cardiovascular deaths (N = 35 and 31 for low and medium vs high), CAD (N = 21 and
18 for low and middle vs high), CVE (N = 13 and 15 for low and middle vs high) and
strokes (N = 15 and 13 for low and middle vs high).
Among them, there were very few studies (4 in cardiovascular
death and CAD, 3 in CVE, and 2 in strokes) where risks were estimated from unadjusted
or raw frequency data. To be consistent, only co-variates adjusted studies were pooled
to see the effects of education. Results are presented in Table 2.3. Effects of education
on these outcomes were heterogeneous across studies with the I2 ranging from 83% to
99%, see Table 2.3. Multivariate meta-analysis was applied indicating significant
educational effects on all outcomes, see Table 2.3 and Figure 2.2. The strongest
education effect was on CVE, where low and medium education increased CVE by 50%
(RR 1.50, 95% CI: 1.17, 1.92) and 27 % (RR 1.27, 95% CI: 1.09, 1.48) compared to
high education. A similar trend occurred for cardiovascular deaths, in which the risks
for these education levels were 39% (RR 1.39, 95% CI: 1.26, 1.54) and 21% (RR 1.21,
95% CI: 1.12, 1.30). In addition, patients with low education showed 36% (RR 1.36,
95% CI: 1.11, 1.66) higher risk, and patients with medium education showed 21% (RR
1.21, 95% CI: 1.06, 1.40) higher risks for CAD. Furthermore, low and medium
education levels were associated with 23% (RR 1.23, 95% CI: 1.06, 1.43) and 17% (RR
1.17, 95% CI: 1.01, 1.35) higher risks, respectively, for developing strokes when
compared to high education level.
Win Khaing Literature Review / 14
Sources of heterogeneity were next explored by meta-regression
or subgroup analyses, see Tables 2.4 – 2.8. Geographical regions were grouped as Asia,
Europe, and US, but only a small number of studies in the Asian setting were available
for most outcomes. Effects of both low/middle educations still remained on all 4
cardiovascular outcomes for pooling within Europe and US, but not for Asia, likely due
to small numbers of studies, see Table 2.4.
We performed subgroup analyses by co-variables including
number of adjusted variables, age (≤60 vs >60 years), BMI (<25 kg/m2 vs ≥ 25 kg/ m2),
percentages of male, diabetes, and smoking (see Tables 2.5 – 2.8), and none of these
was identified as a source of heterogeneity. However, education levels were associated
with all four CVD outcomes in the subgroup younger than 60 years (see Tables 2.5 –
2.8). The risk of cardiovascular deaths and CAD outcomes was higher in the studies
comprising a higher percentage of male subjects. Likewise, the risk of CVD outcomes
(except CAD) was higher in the studies with a higher proportion of diabetic subjects.
The association between BMI and CVE was detected in the BMI subgroup ≥ 25 kg/m2
(see Tables 2.5 – 2.8).
There was no evidence of publication bias using Egger’s test
except for low versus high education level on CVD outcomes (Egger’s test: β=2.33,
p=0.008) which corresponded with funnel plots showing asymmetry (see Figures 2.3
and 2.4). A contour enhanced funnel plot showed that some studies fell in both non-
significant and significant areas, so asymmetry was more likely due to heterogeneity
(see Figures 2.5 and 2.6). No individual study significantly changed the overall
estimates based on the results of the sensitivity analysis.
2.4.2.4 Income and cardiovascular outcomes
A total of 39 studies assessed association between income and
cardiovascular deaths (N = 22 and 13 for low and middle vs high), respectively, CAD
(N = 13 and 14 for low and middle vs high), CVE (both N = 8 for low and middle vs
high) and strokes (both N = 7 for low and middle vs high). Amongst, risk estimations
from unadjusted or raw frequency data (1 in CVD, 4 in CAD, 2 in CVE and 1 in strokes)
were excluded in order to pool the effects of income from co-variates adjusted studies.
Results are shown in Table 2.3. Effects of income on these outcomes were pooled using
multivariate meta-analysis with random-effect models, see Table 2.3 and Figure 2.2.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 15
Substantial heterogeneity across studies was found with I2 ranging from 95% to 99%,
see Table 2.3. The largest income effect was also on cardiovascular deaths, with 76%
(RR 1.76, 95% CI: 1.45, 2.14) and 34% (RR 1.34, 95% CI: 1.17, 1.54) higher risk of
cardiovascular death for low and medium versus high income, respectively. Comparable
effects were seen on CAD, with 49% (RR 1.49, 95% CI: 1.16, 1.91) and 27% (RR 1.27,
95% CI: 1.10, 1.47) higher risks, respectively. Furthermore, patients with low income
showed 17% (RR 1.17, 95% CI: 0.96, 1.44) higher risk, and patients with middle income
showed 5% (RR 1.05, 95% CI: 0.98, 1.13) higher risks for CVE. Additionally, low and
medium incomes showed about 30% (RR 1.30, 95% CI: 0.99, 1.72) and 24% higher
risks (RR 1.24, 95% CI: 1.00, 1.53) of developing strokes when compared to high
income.
Sources of heterogeneity were next explored by meta-regression
or subgroup analyses (see Tables 2.4 – 2.8). By geographical regions, European studies
showed effects of income similar to the overall effect, see Table 2.4.
In subgroup analyses performed by age group and percentages
of males, low income was associated with higher risks for cardiovascular deaths, CAD
and CVE, in the studies consisting of subjects aged 60 years and younger (see Tables
2.5 – 2.8).
No publication bias was identified by Egger’s test except in
medium versus high income level with CAD outcome (Egger’s test: β=2.98, p=0.009),
but funnel plots showed asymmetry (see Figures 2.8 and 2.9). The contour enhanced
funnel plots suggested that asymmetry was more likely due to heterogeneity (see Figures
2.10 and 2.11). Overall estimates were similar to the sensitivity analyses.
2.4.3 Discussion
We performed a systematic review and meta-analysis to pool effects of
education and income on CVD. Our findings indicated that low to middle education and
income carried higher risks of CAD, CVE, strokes and cardiovascular death when
compared to high education and income. The pooled RRs for low and middle versus
high education were 1.36 and 1.21 for CAD, 1.50 and 1.27 for CVE, 1.23 and 1.17 for
strokes, and 1.39 and 1.21 for cardiovascular death. The pooled RRs for low and middle
Win Khaing Literature Review / 16
versus high income for these corresponding outcomes were 1.49 and 1.27, 1.17 and 1.05,
1.30 and 1.24, and 1.76 and 1.34, respectively.
Direct or indirect mechanisms between education and income on CVD have
been described in which behavioral risk factors53, lifestyle or living environment
condition62, health literacy63 and psychological factors64, 65 play important roles. Low
education and low income persons had a higher prevalence of risk behaviors (smoking,
obesity, physical inactivity, unhealthy diet, and etcetera), and were more likely to live
in poor polluted environment, have poor health literacy (ability to read/understand
comprehend medical information, lack of awareness of impact of lifestyle behavior,
poor adherence/incorrect medication, ignorance of medical checkup), and have higher
prevalence of depression with poorer coping in response to cumulative stress.
Consequently, mortality was high, potentially due to delayed access to medical care,
poor understanding in disease progress management, and lack of post-disease cardiac
rehabilitation66-71.
Moreover, education and income have mutual causal influences on CVD
morbidity and mortality and one should not rely on a single, potentially biased
parameters72. Combined effects of education and income have been studied
previously57, and persons with low income and education had the highest risk of incident
CHD, when compared to those with high education/low income, low education/high
income, and high education/high income. However, some researchers have suggested
that education and income should not be combined and should not be interchangeable46,
because they may affect CVD outcomes through different, potentially independent,
causal pathways. For example, Ahmed et al73 found that low income was a significant
independent predictor of HF regardless of education level in community-dwelling older
adults age ≥65 years population. To prove this hypothesis, individual patient data
containing education and income variables are required, and mediation analysis should
be applied.
Many studies46, 74 have compared the difference between the highest and the
lowest strata of socioeconomic measure. This approach does not make maximal use of
the data and one loses the ability to see a “dose-response curve”18, 75, 76. In this study, to
increase comparability across the studies and to study the full gradient of exposure, the
medium-level education and income categories were maintained. This approach
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 17
confirmed the social gradient effect of education and income. Although there was high
heterogeneity in the results, statistical significance was seen, except in effects of income
on CVE and strokes outcomes. This may be due to the possibility of different definitions
and classifications of education and income categories between individual studies, and
between different geographical regions, economies, educational systems and cultures.
Differences in study periods over time could lead to variability in the scales used to
classify the exposure.
2.4.3.1 Strength and limitation
Our meta-analysis has some strength. We believe, it is the first
meta-analysis assessing levels of education and income effects on major CVD
outcomes. To increase comparability across the studies and to study the social gradient
effects, three strata of education and income were categorized and considered to yield
more details than previous meta-analyses40, 41. Effects of education/income were
simultaneously pooled using multivariate meta-analyses. In addition, we included only
cohort studies that could provide more reliable effects of education and income on CVD
outcomes. This review was also conducted in accordance with PRISMA guidelines77.
However, our study has also some limitations. Pooled estimates
were highly heterogeneous, which may be due to differences in characteristics of the
study populations, differences in definitions and classifications of education and income
in both developed and developing countries, and differences in timing of measurement
of education and income categories across the studies. Although many efforts were
made to explore the heterogeneity, we could not identify the sources. We also did not
have access to primary data and many of the studies did not adjust and report for
confounding variables, and thus the estimated risk might be confounded.
2.4.3.2 Clinical Implications and further research
Braveman et al46 explained educational influence on general and
health-related knowledge, health literacy, and problem-solving skills, which can change
one’s health outcome. The results of our meta-analysis provided some evidence of the
effects of education and income on CVD outcomes. However, whether education or
income is directly associated with CVD outcomes72, or education is indirectly associated
with CVD outcomes through income as mediator78, or both education and income are
Win Khaing Literature Review / 18
indirectly associated with CVD outcomes through other risk factors such as BMI79,
diabetes, smoking as mediators has not been clearly answered in studies.
Further research should focus on the causal pathway between
education and income on CVD outcomes with more advanced statistical analysis, such
as a mediation/moderation analysis80.
2.4.4 Conclusion
In conclusion, low/medium education and income increase the risks of
CAD, CVE, stroke and cardiovascular death. Further studies should be conducted to
assess causal pathway of education/income on cardiovascular outcomes to confirm our
findings, especially in Asian countries.
2.5 The association between education/income and cardiovascular risk
factors Winkley and colleague50 studied the relationship between education/
income/ occupation and cardiovascular risk factors (cigarette smoking, BP, and
cholesterol) on people aged 25 to 64 and showed that only education was significantly
associated with these risk factors after adjustment for age and time of survey. Hoeymans
et al 81 also provided evidence that there was inverse association between educational
level and prevalence of smoking, physical inactivity, obesity, hypertension,
hypercholesterolemia and low high density lipoprotein (HDL)-cholesterol, but not with
alcohol. The Minnesota Heart Survey82 conducted in 7781 adults aged 25 to 74 years
shown that education was inversely related to BP, cigarette smoking and BMI in both
men and women while serum cholesterol was inversely related to education in women
only. However, for household income, the results showed less consistency in magnitude
and direction. A study from India83 showed that higher education level was associated
with overweight, physical inactivity, family history of CVD, higher fruit intake and
lower alcohol intake in both men and women, but higher diabetes and hypertension
prevalence was found only in men. BMI and waist circumference were also greater in
those with higher educational level for both sexes. The poorer had less diabetes, less
overweight in both sexes and less likely to have a family or established history of CVD,
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 19
and smoked more in poor men. They concluded that some biological cardiovascular risk
factors were worse in higher SES peoples while some behavioral risk factors were worse
in lower SES peoples with little knowledge about risk factors and screening practice. In
the SESAMI study by Alter et al 32, income was inversely associated with 2-year
mortality rate in unadjusted model (HR=0.45, 95%CI: 0.35, 0.57; P<0.001), but after
adjustment for age and preexisting cardiovascular events or cardiovascular risk factors,
the effect was attenuated (HR=0.77, 95%CI: 0.54, 1.10; P= 0.150).
2.6 Causal pathways between education/income and cardiovascular
diseases Many researchers agreed that socioeconomic indicators should not be used
interchangeably46 because they may effect outcome in different causal pathways and
may represent independent separate important risk factors of CVD. Although education
and income correlated to each other, it is not strong enough to use education and income
as proxies for each other46. They measure different phenomena and tap into different
causal mechanisms. Income can vary at similar education levels, mainly across different
social (eg., age, sex, race, ethnic) groups.
Many researchers have accepted that education/income should be included
and considered alongside with standard risk factors in risk prediction. They recognized
that, even in the well-known Framingham risk score, it was underestimated in low SES
and overestimated in the highest SES groups84, 85. Molshatzki and colleagues86 found
that long-term post-MI prediction model considerably gained improvement when
education/income was considered into the model. Gerber and colleague87 studied
income-by-education interaction in post-MI patients which showed that low income
with low education patients had higher mortality risk, because they failed to attend
cardiac rehabilitation, and did not adhered to post-discharge medication and lifestyle
recommendations.
Win Khaing Literature Review / 20
2.7 Conceptual framework Education may directly affect major cardiovascular events and its effect
may also be indirectly mediated through income effect.
Income may directly affect major cardiovascular events and its effect may
also be indirectly mediated through education effect. (Figure 2.12)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 21
Table 2.1 Characteristics of included studies
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Arrich88 2005 Austria, Europe IHD RR Education
(Medium vs High) 1.02 a (0.82, 1.28) 0 66.7 54.3 25.3 26.5 67.0 NA
(High) Education (Low vs High)
1.29 a (1.01, 1.65)
Income (Medium vs High)
1.07 a (0.92, 1.24) 0 65.5 54.4 27.0 27.6 45.0 NA
Income (Low vs High)
0.79 a (0.41, 1.50)
Rehkopf89 2015 US (High) IHD OR Education
(Medium vs High) 1.01 (1.01, 1.02) 9 47 78 8 NA 24 NA
Income (Medium vs High)
0.99 (0.98,1.00)
Geyer90 2006 Germany, Europe MI RR Education
(Medium vs High) 3.41 (2.18, 5.35) 1 42.5 72.4 NA NA NA NA
(High)
Education (Low vs High)
4.06 (2.14, 7.67)
Income (Medium vs High)
1.48 (1.24, 1.76)
Income (Low vs High)
2.02 (1.83, 2.23)
Honjo91 2008 Japan, Asia
(High) CHD HR Education (Medium vs High)
1.75 b (0.52, 5.88) 11 NA 0 2.36 7.28 14.8 NA
Education (Low vs High)
1.28 b (0.67, 2.35)
W
in Khaing
Literature R
eview / 22
Table 2.1 Characteristics of included studies (continued)
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Rawshani92 2015 Sweden, Europe CHD HR Education
(Medium vs High) 1.16 b, c (1.09, 1.23) 14 39.3 53.8 100 11.8 NA 25.6
(High)
Education (Low vs High)
1.35 b, c (1.10, 1.64)
Income (Medium vs High)
1.39 c (1.04, 1.86) 14 39.2 53.8 100 12.3 NA 25.6
Income (Low vs High)
1.86 c (1.54, 2.24)
Thurston93 2005 US (High) CHD HR Education
(Medium vs High) 1.22 c (0.95, 1.55) 13 47.4 45.6 3.8 38.1 6.54 25.6
Education (Low vs High)
1.40 c (1.10, 1.77)
Income (Medium vs High)
1.24 c (1.05, 1.46)
Income (Low vs High)
1.23 c (1.05, 1.43)
Salomaa30 2000 Finland, Europe (High)
MI RR Education (Low vs High)
1.49 c (1.42, 1.56) 2 NA NA NA NA NA NA
Income (Low vs High)
1.72 c (1.65, 1.79)
Andersen29 2003 Demark, Europe IHD HR Income
(Medium vs High) 1.18 c (1.05, 1.32) 9 52.7 46.4 NA 36.2 NA 25.0
(High)
Income (Low vs High)
1.45 c (1.30, 1.63)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 23
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Hetemaa94 2006 Finland, Europe MI HR Education
(Medium vs High) 0.77 c (0.68, 0.86) 13 67.3 61.8 16.3 NA 32.1 NA
(High) Education (Low vs High)
0.67 c (0.60, 0.75)
Income (Medium vs High)
0.83 c (0.77, 0.90)
Income (Low vs High)
0.67 c (0.61, 0.73)
Peter95 2007 Germany, Europe IHD HR Education
(Medium vs High) 0.29 c (0.24, 0.35) 0 38.9 53.6 NA NA NA NA
(High) Education (Low vs High)
0.61 c (0.49, 0.76)
Income (Medium vs High)
1.81 c (1.39, 2.36)
Income (Low vs High)
2.98 c (2.17, 4.10)
MI HR Education (Medium vs High)
0.25 c (0.18, 0.35)
Education (Low vs High)
0.64 c (0.45, 0.91)
Income (Medium vs High)
2.39 c (1.55, 3.67)
Income (Low vs High)
4.06 c (2.36, 6.97)
Lammintausta31 2012 Finland, Europe MI RR Income
(Medium vs High) 1.90 c (1.69, 2.14) 2 56.7 44.9 NA NA NA NA
(High) Income (Low vs High)
2.82 c (2.56, 3.10)
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 24
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Honjo96 2010 Japan, Asia CHD HR Education
(Medium vs High) 1.03 b, c (0.92, 1.15) 11 54.8 38.8 3.7 23.3 14.4 NA
(High) Education (Low vs High)
0.65 b, c (0.30, 1.40)
Roux97 2001 US (High) CHD RR Education
(Medium vs High) 1.41 c (1.15, 1.73) 1 NA NA NA NA NA NA
Income (Medium vs High)
1.41 c (1.19, 1.68)
Fujino98 2005 Japan, Asia IHD RR Education
(Medium vs High) 0.88 c (0.68, 1.14) 5 66.1 NA NA 21.7 NA NA
(High) Education (Low vs High)
0.85 c (0.68, 1.07)
Andersen78 2005 Denmark, Europe MI HR Income
(Medium vs High) 1.05 (0.84, 1.31) 11 49.5 57.3 NA 36.2 NA NA
(High) Income (Low vs High)
1.17 (0.85, 1.61)
Lynch53 1996 Finland, Europe MI HR Income
(Medium vs High) 1.91 (0.79, 4.63) 23 NA 100 NA NA NA NA
(High) Income (Low vs High)
2.30 (1.21, 4.37)
Lee99 2000 Taiwan,
Asia (High)
CAD OR Education (Low vs High)
1.25 b (0.83, 1.67) 0 NA 47.3 8.6 31.6 28.4 23.84
Weikert100 2008 Germany, Europe MI RR Education
(Medium vs High) 1.18 (0.85, 1.63) 2 54.5 64.5 10.4 22.9 58.5 26.9
(High) Education (Low vs High)
1.22 (0.91, 1.62)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 25
Table 2.1 Characteristics of included studies (continued)
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Hippe101 1999 Demark, Europe MI RR Education
(Medium vs High) 1.36 b, c (1.25, 1.49) 2 66 53.5 NA NA NA NA
(High) Education (Low vs High)
1.71 b, c (1.41, 2.07)
Huisman102 2008 Netherlands,
Europe (High)
MI RR Education (Low vs High)
1.72 (1.06, 2.80) 10 42.3 67.5 NA NA NA NA
Eaker103 1992 US (High) MI HR Education
(Medium vs High) 1.60 (0.70, 3.70) 6 54 0 NA NA NA NA
Education (Low vs High)
2.5 (1.00, 6.10)
Bosma104 1995 Lithuania, Europe MI RR Education
(Medium vs High) 1.40 (0.80, 2.46) 2 51.6 100 NA 72.6 NA 27.2
(High) Education (Low vs High)
1.42 (0.83, 2.45)
Netherlands, Europe Education
(Medium vs High) 0.78 (0.46, 1.31) 2 52.4 100 NA 92.2 NA 25.5
(High) Education (Low vs High)
0.68 (0.38, 1.23)
Chaix105 2007 Sweden, Europe IHD HR Education
(Low vs High) 1.38 (1.24, 1.53) 10 NA NA NA NA NA NA
(High) Income (Medium vs High)
1.30 (1.10, 1.52)
Income (Low vs High)
1.65 (1.38, 1.97)
W
in Khaing
Literature R
eview / 26
Table 2.1 Characteristics of included studies (continued)
Author Year
Country, Setting (income level)
Outcome Risk
measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM
(%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(a) Coronary artery diseases
Kuper106 2006 Sweden, Europe MI HR Education
(Medium vs High) 1.70 (1.10, 2.50) 7 40.2 0 1.3 59.3 9.3 23.5
(High) Education (Low vs High)
1.90 (1.30, 2.80)
Lapidus & Bengtsson107 1986
Sweden, Europe (High)
MI RR Education (Low vs High)
1.50 (0.60, 3.50) 1 NA 0 NA NA NA NA
(b) Cardiovascular events
Braig108 2011 Germany, Europe
CVE (MI with OR Education
(Medium vs High) 1.22 (0.94, 1.59) 5 50 59.6 1.7 24.2 NA 26.1
(High) Stroke) Education (Low vs High)
1.83 (1.52, 2.21)
Jakobsen109 2012 Demark, Europe CVE HR Education
(Medium vs High) 0.94 (0.77, 1.15) 25 NA 75 6.5 48.7 NA NA
(High) Education (Low vs High)
0.87 (0.71, 1.07)
Income (Medium vs High)
1.15 (0.97, 1.36) 25 NA 73.2 9.5 46.3 NA NA
Income (Low vs High)
1.06 (0.87, 1.29)
Panagiotakos25 2016 Greek, Europe CVE HR Education
(Medium vs High) 0.78 (0.41, 1.51) 9 45.5 49.8 9.0 54.6 31.8 26.3
(High) Education (Low vs High)
1.31 (0.63, 2.74)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 27
Table 2.1 Characteristics of included studies (continued)
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(b) Cardiovascular events
Rasmussen110 2007 Demark, Europe RV HR Education
(Medium vs High) 1.04 b (1.02, 1.05) 2 60.8 71.1 4.3 NA NA NA
(High) (CVE) Education (Low vs High)
1.01 b (0.95, 1.06)
Income (Medium vs High)
0.96 b (0.95, 0.97) 2 60.7 71.1 9.5 NA NA NA
Income (Low vs High)
0.89 b (0.85, 0.94)
Senan & Petrosyan111 2014
India, Asia
(Lower- CVE RR Education
(Low vs High) 1.22 b, c (1.19, 1.25) 4 NA 75.1 NA NA NA NA
middle) Income (Low vs High)
1.15 b, c (1.14, 1.16)
Bosma112 2005 Netherla
nds, Europe
CVE HR Education (Medium vs High)
1.15 (0.84, 1.57) 10 69.7 41.7 5.4 NA 20.4 NA
(High) Education (Low vs High)
1.24 (0.92, 1.68)
Income (Medium vs High)
1.30 (0.94, 1.79)
Income (Low vs High)
1.21 (0.89, 1.64)
Masoudkabir113 2012 Iran, Asia
(Upper- CVE
(IHD with HR Education (Medium vs High)
1.14 b (0.88, 1.47) 5 58.8 45.4 21.8 25.7 59.8 27.3
middle) stroke) Education (Low vs High)
0.99 b (0.52, 1.89)
Income (Medium vs High)
1.05 b (0.98, 1.12)
Income (Low vs High)
1.18 b (0.81, 1.70)
W
in Khaing
Literature R
eview / 28
Table 2.1 Characteristics of included studies (continued)
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2) (b) Cardiovascular events
Minh114 2006 Vietnam, Asia (Lower- CVE RR Education
(Low vs High) 4.50 (3.40, 5.80) 2 41.6 53.6 NA NA Na NA
middle) Income (Low vs High)
1.25 b (0.83, 1.67)
Hirokawa115 2006 Japan, Asia (High) CVE HR Education
(Medium vs High) 1.67 c (0.90, 3.09) 13 55.0 61.2 NA 25.0 35.1 23.02
Education (Low vs High)
2.01 c (1.04, 3.91)
Siegel116 1987 US (High) CVE HR Education (Medium vs High)
0.59 (0.26, 1.33) 11 72.9 36.5 NA 11.3 NA NA
He117 2001 US (High) HF RR Education (Medium vs High)
1.22 c (1.05, 1.41) 14 49.8 40.6 3.82 35 28.2 25.6
Christensen11
8 2011 Demark, Europe HF HR Education
(Low vs High) 1.27 b (1.19, 1.36) 12 52.4 45.3 2.87 63.4 6.25 25.15
(High) Income (Medium vs High)
1.13 b, c (1.08, 1.17) 0 52.4 45.3 2.87 63.4 6.25 25.15
Income (Low vs High)
1.51 b, c (1.28, 1.78)
Borne119 2011 Sweden, Europe HF HR Income
(Medium vs High) 0.97 b ,c (0.96, 0.98) 4 60.8 44.4 NA NA NA NA
(High) Income (Low vs High)
1.67 b, c (1.61, 1.73)
Philbin120 2001 US (High) HF OR Income (Medium vs High)
1.08 (1.01, 1.16) 6 74 43 33 NA 45 NA
Income (Low vs High)
1.18 (1.10, 1.26)
Schwarz & Elman121 2003 US (High) HF HR Education
(Medium vs High) 0.51 (0.26, 1.02) 0 78 50 42 NA 33 NA
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 29
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(b) Cardiovascular events
Sui122 2008 US (High) HF HR Education
(Medium vs High) 1.63 (0.94, 2.81) 0 63.7 78 27.5 NA 46.5 NA
CVE HR Education (Medium vs High)
1.55 (1.05, 2.30)
Rosvall123 2006 Sweden, Europe CVE HR Education
(Medium vs High) 1.59 (0.89, 2.80) 2 59.2 44.4 8.1 30.3 18.2 25.4
(High) Education (Low vs High)
2.19 (1.29, 3.73)
Engstrom124 2000 Sweden, Europe CVE HR Education
(Medium vs High) 2.77 b (1.46, 5.27) 6 51 0 13 72 31 24.6
(High) Education (Low vs High)
2.86 b (0.91, 9.09)
Notara28 2016 Greek, Europe CVE HR Education
(Medium vs High) 1.61 (1.23, 2.08) 9 66.1 76.0 31.5 NA 53.6 NA
(High) Education (Low vs High)
1.25 (0.88, 1.78)
(c) Strokes
Weikert100 2008 Germany, Europe Stroke RR Education
(Medium vs High) 1.66 (1.13, 2.45) 2 55.9 64.5 13.8 12.4 63.8 26.8
(High) Education (Low vs High)
1.63 (1.14, 2.33)
Lapidus & Bengtsson107 1986
Sweden, Europe (High)
Stroke RR Education (Low vs High)
1.30 (0.40, 4.10) 1 NA 0 NA NA NA NA
W
in Khaing
Literature R
eview / 30
Table 2.1 Characteristics of included studies (continued)
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI (kg/m2)
(c) Strokes Avendano & Glymour125 2008 US
(High) Stroke HR Education (Medium vs High)
1.07 b, c (0.97, 1.18) 17 67.7 43 21 22 21 27
Education (Low vs High)
0.96 b, c (0.86, 1.08)
Income (Medium vs High)
1.00 b, c (0.99, 1.01)
Income (Low vs High)
1.08 b, c (1.01, 1.16)
Rawshani92 2015 Sweden, Europe Stroke HR Education
(Medium vs High) 1.42 b, c (1.31, 1.55) 14 39.3 53.8 100 11.8 NA 25.6
(High) Education (Low vs High)
1.82 b, c (1.33, 2.50)
Income (Medium vs High)
1.29 c (0.88, 1.69) 14 39.2 53.8 100 12.3 NA 25.6
Income (Low vs High)
2.09 c (1.62, 2.69)
Li126 2008 Sweden, Europe Stroke RR Income
(Medium vs High) 1.41 c (1.21, 1.63) 4 62.7 48.9 NA NA NA NA
(High) Income (Low vs High)
1.45 c (1.24, 1.70)
Rossum127 1999 Netherlands, Europe Stroke RR Education
(Medium vs High) 0.90 b (0.83, 25.0) 12 71 0 4.5 18.7 35.8 26.8
(High) Education (Low vs High)
4.79 b (1.48, 15.5)
Income (Medium vs High)
1.42 b (1.04, 1.96)
Income (Low vs High)
1.75 b (0.81, 3.85)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 31
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
Smoking (%)
HT (%)
Mean BMI
(kg/m2)
(c) Strokes Gillum & Mussolino128 2003 US
(High) Stroke RR Education (Medium vs High)
0.85 b, c (0.84, 0.86) 9 62 47.3 7.48 29.81 19.47 NA
Education (Low vs High)
1.16 b, c (0.90, 1.49)
Kuper129 2007 Sweden, Europe Stroke HR Education
(Medium vs High) 1.20 (0.90, 1.80) 7 40.27 0 1.3 59.42 9.24 23.49
(High) Education (Low vs High)
1.5 (1.00, 2.20)
Jackson130 2014 Australia, Asia Stroke OR Education
(Medium vs High) 1.57 (0.95, 2.61) 11 49.5 0 3.68 16.91 24.53 25.94
(High) Education (Low vs High)
1.52 (0.99, 2.33)
Honjo96 2010 Japan, Asia (High) Stroke HR Education
(Medium vs High) 1.38 b, c (1.24, 1.54) 11 54.78 38.8 3.65 23.32 14.36 NA
Education (Low vs High)
1.04 b, c (0.65, 1.67)
Fujino98 2005 Japan, Asia (High) Stroke RR Education
(Medium vs High) 1.14 c (0.98, 1.33) 5 66.08 NA NA 21.72 NA NA
Education (Low vs High)
1.22 c (1.01, 1.47)
Lee99 2000 Taiwan,
Asia (High)
Stroke OR Education (Low vs High)
1.67 b (0.91, 2.50) 0 NA 47.28 8.55 31.62 28.4 23.84
Honjo91 2008 Japan, Asia (High) Stroke HR Education
(Medium vs High) 0.74 b (0.51, 1.09) 11 NA 0 2.36 7.28 NA NA
Education (Low vs High)
1.10 b (0.96, 1.28)
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 32
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2) (c) Strokes
Arrich88 2005 Austria, Europe Stroke RR Education
(Medium vs High) 1.90 a (1.56, 2.31) 0 66.7 54.3 25.3 26.5 67.0 NA
(High) Education (Low vs High)
2.07 a (1.68, 2.55)
Income (Medium vs High)
1.29 a (1.16, 1.42) 0 65.5 54.4 27.0 27.6 45.0 NA
Income (Low vs High)
1.25 a (0.89, 1.76)
Andersen131 2014 Denmark, Europe Stroke RR Education
(Medium vs High) 1.17 b (1.15, 1.18) 4 71.9 52.5 12.5 27.4 46.9 NA
(High) Education (Low vs High)
1.17 b (1.15, 1.22)
Income (Medium vs High)
1.51 b (1.49, 1.52)
Income (Low vs High)
1.35 b (1.34, 1.37)
Avendano56 2006 US (High) Stroke HR Education (Medium vs High)
0.89 (0.54, 1.47) 12 NA NA NA NA NA NA
Education (Low vs High)
0.73 (0.51, 1.04)
Income (Medium vs High)
0.74 (0.44, 1.26)
Income (Low vs High)
0.70 (0.47, 1.06)
(d) Cardiovascular deaths
Lynch53 1996 Finland, Europe
Death due to HR Income
(Medium vs High) 0.34 (0.13, 0.93) 23 NA 100 NA NA NA NA
(High) CVE Income (Low vs High)
0.72 (0.39, 1.34)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 33
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2) (d) Cardiovascular deaths
Jakovljevic132 2001 Finland, Europe
Death due to stroke HR Education
(Low vs High) 1.19 (1.05, 1.33) 4 NA 71.3 NA NA NA NA
(High) Income (Medium vs High)
1.36 (1.13, 1.64)
Income (Low vs High)
1.72 (1.45, 2.05)
Zhou133 2006 China, Asia (High)
Death due to stroke HR Education
(Medium vs High) 1.02 (0.57, 1.83) 9 77.2 54.8 26.5 27.7 NA NA
Education (Low vs High)
0.88 (0.46, 1.68)
Income (Low vs High)
3.37 (2.34, 4.87)
Beebe-Dimmer22 2004 US (High) Death due
to IHD or HR Education (Medium vs High)
0.99 (0.80, 1.21) 7 44 0 NA 40.3 NA NA
stroke Education (Low vs High)
1 (0.79, 1.28)
Income (Low vs High)
1.45 (1.20, 1.74)
Jakobsen109 2012 Demark, Europe
Death due to CVE HR Education
(Medium vs High) 0.82 (0.57, 1.16) 25 NA 75 6.5 48.7 NA NA
(High) Education (Low vs High)
0.74 (0.52, 1.04)
Income (Medium vs High)
1.05 (0.77, 1.44) 25 NA 73.2 9.5 47.3 NA NA
Income (Low vs High)
1.22 (0.88, 1.69)
Kim134 2005 US (High) Death due to CVE OR Education
(Low vs High) 1.41 (1.28, 1.56) 3 45.0 0 NA NA NA NA
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 34
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(d) Cardiovascular deaths
Geyer90 2006 Sweden, Europe
Death due to MI RR Education
(Medium vs High) 1.22 (1.14, 1.30) 1 47.8 49.3 NA NA NA NA
(High) Education (Low vs High)
1.41 (1.31, 1.50)
Income (Medium vs High)
1.38 (1.27, 1.51)
Income (Low vs High)
2.20 (2.09, 2.31)
Qureshi135 2003 US (High)
Death due to stroke RR Education
(Medium vs High) 1.37 c (1.09, 1.71) 9 50.7 42.4 4.0 26.45 NA 25.82
Death due to MI RR Education
(Medium vs High) 1.36 c (1.18, 1.57)
Pednekar136 2011 India, Asia
(Lower-
Death due to CVE HR Education
(Low vs High) 1.15 b, c (1.14, 1.17) 5 51.7 59.8 NA 9.93 NA NA
middle) Death due to IHD HR Education
(Low vs High) 1.05 b, c (1.04, 1.07)
Death due to stroke HR Education
(Low vs High) 2.31 b, c (1.98, 2.68)
Rawshani92 2015 Sweden, Europe
Death due to CVE HR Education
(Medium vs High) 1.54 c (1.41, 1.68) 11 39.3 53.8 100 11.81 NA 25.61
(High) Education (Low vs High)
1.27 c (0.94, 1.70)
Income (Medium vs High)
1.92 c (1.31, 2.81) 14 39.2 53.8 NA 12.25 NA 25.6
Income (Low vs High)
3.40 c (2.64, 4.37)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 35
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(d) Cardiovascular deaths
Coady137 2014 US (High) Death due to MI HR Education
(Medium vs High) 1.04 c (0.96, 1.13) 9 78 49.6 28.3 NA 55.0 NA
Education (Low vs High)
1.08 c (0.99, 1.17)
Gallo138 2012 Europe (High)
Death due to CVE HR Education
(Medium vs High) 1.21 b, c (1.20, 1.23) 7 52.0 35.9 NA 27.3 NA NA
Education (Low vs High)
1.63 b, c (1.46, 1.82)
Death due to IHD HR Education
(Medium vs High) 1.24 b, c (1.22, 1.26)
Education (Low vs High)
1.87 b, c (1.60, 2.19)
Death due to stroke HR Education
(Medium vs High) 1.16 b, c (1.12, 1.19)
Education (Low vs High)
1.43 b, c (1.12, 1.83)
Rasmussen44 2006 Demark, Europe
Death due to MI RR Education
(Medium vs High) 1.10 c (1.02, 1.19) 6 61.0 70.9 3.6 NA NA NA
(High) Education (Low vs High)
1.15 c (1.06, 1.25)
Income (Medium vs High)
1.14 c (1.08, 1.20) 6 61 60.2 3.6 NA NA NA
Income (Low vs High)
1.42 c (1.35, 1.50)
Salomaa30 2000 Finland, Europe
Death due to MI RR Education
(Low vs High) 1.96 c (1.84, 2.08) 2 NA NA NA NA NA NA
(High) Income (Low vs High)
2.61 c (2.47, 2.76)
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 36
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(d) Cardiovascular deaths Bucher & Ragland139 1995 US
(High) Death due to CHD RR Education
(Medium vs High) 1.54 (1.13, 2.09) 5 46.2 100 NA NA NA NA
Income (Low vs High)
2.07 (0.94, 4.57)
Death due to stroke RR Education
(Medium vs High) 1.27 (0.97, 1.66)
Income (Low vs High)
1.08 (0.56, 2.08)
Tonne140 2005 US (High)
Death due to MI RR Education
(Medium vs High) 1.21 (1.05, 1.39) 13 69 58.1 31 NA 63.5 NA
Education (Low vs High)
1.32 (1.15, 1.52)
Income (Medium vs High)
1.25 (1.04, 1.52)
Income (Low vs High)
1.38 (1.14, 1.67)
Chen39 2015 China, Asia (High)
Death due to stroke HR Education
(Low vs High) 1.88 (1.05, 3.36) 14 73.4 46.1 10.2 22.6 72.5 NA
Income (Low vs High)
1.64 (0.97, 2.78)
Chaix105 2007 Sweden, Europe
Death due to IHD HR Education
(Low vs High) 1.46 (1.24, 1.73) 10 NA NA NA NA NA NA
(High) Income (Medium vs High)
1.85 (1.43, 2.43)
Income (Low vs High)
2.83 (2.16, 3.82)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 37
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2) (d) Cardiovascular deaths
Ito141 2008 Japan, Asia (High)
Death due to CVE HR Education
(Medium vs High) 1.15 c (0.77, 1.70) 14 NA 48.3 NA 28.8 NA NA
Education (Low vs High)
1.33 c (0.90, 1.97)
Lee99 2000 Taiwan, Asia (High)
Death due to CVE OR Education
(Low vs High) 1.25 b (0.83, 2.00) 0 NA 47.3 8.55 31.6 28.
4 23.8
Minh142 2003 Vietnam,
Asia (Lower-middle)
Death due to CVE RR Education
(Low vs High) 1.00 (0.32, 3.13) 4 NA 75.2 NA NA NA NA
Liu143 1982 Chicago, CHA, US
Death due to CHD RR Education
(Medium vs High) 2.12 (1.15, 3.89) 1 48.9 100 NA 40.2 NA NA
(High) Education (Low vs High)
3.6 (1.99, 6.60)
Death due to CVE RR Education
(Medium vs High) 2.49 (1.40, 4.44)
Education (Low vs High)
4.08 (2.31, 7.21)
Chicago, WEPG, US
Death due to CHD RR Education
(Medium vs High) 1.00 (0.63, 1.59) 1 48.6 100 NA 70.2 NA NA
(High) Education (Low vs High)
1.62 (1.08, 2.44)
Death due to CVE RR Education
(Medium vs High) 0.97 (0.62, 1.42)
Education (Low vs High)
1.52 (1.09, 2.11)
Kilander67 2001 Sweden, Europe
Death due to CVE and HR Education
(Medium vs High) 0.78 (0.48, 1.24) 17 NA 100 NA 50.6 NA 25.0
(High) stroke Education (Low vs High)
1.01 (0.67, 1.52)
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 38
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2) (d) Cardiovascular deaths Schwarz & Elman121 2003 US
(High) Death due to
HF HR Education (Medium vs High)
0.58 (0.10, 3.43) 0 78.7 61.9 NA NA NA NA
Sui122 2008 US (High)
Death due to HF HR Education
(Medium vs High) 1.58 (0.66, 3.78) 0 63.7 78 27.
5 NA 46.5 NA
Death due to CVE HR Education
(Medium vs High) 1.60 (0.90, 2.84)
Bosma104 1995 Lithuania, Europe
Death due to CHD RR Education
(Medium vs High) 1.06 (0.60, 1.90) 2 51.6 100 NA 72.6 NA 27.2
(High) Education (Low vs High)
1.08 (0.62, 1.88)
Death due to CVE RR Education
(Medium vs High) 1.12 (0.67, 1.86)
Education (Low vs High)
1.16 (0.72, 1.88)
Netherlands, Europe
Death due to CHD RR Education
(Medium vs High) 1.78 (0.85, 3.70) 2 52.4 100 NA 92.2 NA 25.5
(High) Education (Low vs High)
1.06 (0.46, 2.43)
Death due to CVE RR Education
(Medium vs High) 1.56 (0.88, 2.77)
Education (Low vs High)
1.40 (0.76, 2.58)
Lapidus & Bengtsson107 1986
Sweden, Europe (High)
Death due to CVE RR Education
(Low vs High) 1.2 (0.7, 2.0) 1 NA 0 NA NA NA NA
Notara28 2016 Greece, Europe
Death due to ACS HR Education
(Medium vs High) 1.72 (1.35, 2.22) 9 66.1 76.0 31.
5 NA 53.6 NA
(High) Education (Low vs High)
1.33 (0.93, 1.92)
Table 2.1 Characteristics of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 39
Author Year
Country, Setting (income level)
Outcome Risk measure
Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(d) Cardiovascular deaths
Rawshani144 2016 Sweden, Europe
Death due to CVE HR Education
(Medium vs High) 1.08 b (1.04, 1.10) 16 58.3 60.4 100 17.2 NA 30.3
(High) Education (Low vs High)
1.19 b (1.10, 1.28)
Income (Medium vs High)
1.54 (1.42, 1.68)
Income (Low vs High)
1.87 (1.76, 1.99)
Lammintausta31 2012 Finland, Europe
Death due to MI RR Income
(Medium vs High) 2.68 c (2.12, 3.24) 2 56.7 44.9 NA NA NA NA
(High) Income (Low vs High)
4.78 c (4.13, 5.54)
Li126 2008 Sweden, Europe
Death due to stroke RR Income
(Medium vs High) 1.32 c (0.90, 1.93) 4 62.7 48.9 NA NA NA NA
(High) Income (Low vs High)
1.90 c (1.32, 2.72)
Rosvall145 2008 Sweden, Europe
Death due to MI HR Income
(Medium vs High) 1.09 b, c (1.05, 1.14) 1 70.4 65.4 7.4 NA 8.04 NA
(High) Income (Low vs High)
1.26 b, c (1.22, 1.30)
Khang146 2007 South Korea, Asia
Death due to CVE RR Income
(Low vs High) 1.35 c (1.25, 1.45) 5 43.1 100 NA 57.2 NA NA
(High) Death due to IHD RR Income
(Low vs High) 1.20 c (1.05, 1.36)
Table 2.1 Characteristics of included studies (continued)
W
in Khaing
Literature R
eview / 40
Author Year Country, Setting
(income level) Outcome Risk
measure Study factors (Categories)
Relative Risks (95%CI)
NC Mean Age
(years)
Male (%)
DM (%)
SMK (%)
HT (%)
Mean BMI
(kg/m2)
(d) Cardiovascular deaths
Rosvall147 2008 Sweden, Europe (High)
Death due to MI OR Income
(Low vs High) 1.23 c (1.18, 1.28) 2 63.8 70.5
4 NA NA NA NA
Arrich88 2005 Austria, Europe
Death due to stroke HR
Education (Medium vs High)
0.85 (0.60, 1.19) 9 66.71 54.2
8 25.2
7 26.46 66.98 NA
(High) Education (Low vs High)
0.86 (0.56, 1.32)
Income (Medium vs High)
1.64 (1.23, 2.17) 0 65.5 54.3
5 27.0
3 27.6 44.95 NA
Income (Low vs High)
1.07 (0.26, 4.39)
RR, Relative risk, OR, Odds ratio; HR, Hazard ratio; NC; Number of controlled variable; DM, Diabetes Mellitus; SMK, Smoking; HT, Hypertension; BMI, Body Mass Index; IHD, Ischemic heart disease; MI, Myocardial infarction; CHD, Coronary heart disease; CAD, Coronary artery disease; CVE, Cardiovascular events; RV, Revascularization; HF, Heart failure; ACS, Acute coronary syndrome; NA, US, United States, CHA, Chicago Heart Association Detection Project; WEPG, Chicago Western Electric Company Study and Peoples Gas Company Studies not available (or) not reported a, RR (95%CI) was recalculated based on raw/frequency data reported in original article; b, RR (95%CI) was recalculated by reversing original RR if the middle or lowest category of education or income was used as a reference group; c, RR (95%CI) was recalculated by pooling separate subgroup RRs (weighted by inverse of their variance) to obtain a single estimate from each study
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 41
Table 2.2 Risk of bias assessment of included studies
Authors Year
Selection Comparability Outcome
Total stars Representativeness
of cohort
Selection of non-exposed cohort
Ascertainment of exposure
Outcome of interest
Comparability of cohorts
Assessment of outcome
Adequate duration of follow
up
Adequate follow up of cohort
Andersen et al.29 2003 a(*) a(*) a(*) a(*) a(*), b(*) b(*) a(*) c 8
Andersen et al.78 2005 a(*) a(*) a(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Andersen et al.131 2014 a(*) a(*) a(*) a(*) a(*) b(*) a(*) b(*) 8
Arrich et al.88 2005 b(*) a(*) b(*) b a(*), b(*) b(*) b c 6
Avendano et al.56 2006 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Avendano & Glymour125 2008 a(*) a(*) b(*) a(*) a(*) b(*) a(*) b(*) 8
Beebe-Dimmer et al.22 2004 c (women) a(*) c b a(*), b(*) b(*) a(*) b(*) 6
Borné et al.119 2011 a(*) a(*) c a(*) a(*) b(*) a(*) b(*) 7
Bosma et al.104 1995 c (men) a(*) a(*) b a(*) b(*) a(*) b(*) 6
Bosma et al.112 2005 a(*) a(*) c a(*) a(*), b(*) b(*) a(*) d 7
Braig et al.108 2011 a(*) a(*) c a(*) a(*), b(*) c b b(*) 6
Bucher & Ragland139 1995 c (men) a(*) a(*) a(*) a(*) b(*) a(*) d 6
Chaix et al.105 2007 a(*) a(*) a(*) a(*) a(*) b(*) a(*) c 7
Chen et al.39 2015 a(*) a(*) b(*) b a(*), b(*) c a(*) b(*) 7
Christensen et al.118 2011 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
W
in Khaing
Literature R
eview / 42
Table 2.2 Risk of bias assessment of included studies (continued)
Authors Year
Selection Comparability Outcome
Total stars Representativeness
of cohort
Selection of non-exposed cohort
Ascertainment of exposure
Outcome of interest
Comparability of cohorts
Assessment of outcome
Adequate duration of follow
up
Adequate follow up of cohort
Coady et al.137 2014 a(*) a(*) a(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Eaker et al.103 1992 c (women) a(*) b(*) b a(*), b(*) a(*) a(*) b(*) 7
Engström et al.124 2000 c (women) a(*) a(*) b a(*), b(*) b(*) a(*) d 6
Fujino et al.98 2005 a(*) a(*) b(*) a(*) a(*) b(*) a(*) c 7
Gallo et al.138 2012 a(*) a(*) b(*) b a(*), b(*) b(*) a(*) b(*) 8
Geyer et al.90 2006 a(*) a(*) a(*) b a(*) b(*) a(*) d 6
Gillum & Mussolino128 2003 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
He et al.117 2001 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Hetemaa et al.94 2006 a(*) a(*) a(*) a(*) a(*) b(*) b d 6
Hippe et al.101 1999 a(*) a(*) a(*) a(*) a(*) b(*) a(*) b(*) 8
Hirokawa et al.115 2006 a(*) a(*) c a(*) a(*), b(*) b(*) a(*) c 7
Honjo et al.91 2008 c (women) a(*) c a(*) a(*), b(*) b(*) a(*) b(*) 7
Honjo et al.96 2010 a(*) a(*) c a(*) a(*), b(*) a(*) a(*) c 7
Huisman et al.102 2008 a(*) a(*) c a(*) a(*) b(*) a(*) c 6
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 43
Table 2.2 Risk of bias assessment of included studies (continued)
Authors Year
Selection Comparability Outcome
Total stars Representativeness
of cohort
Selection of non-exposed cohort
Ascertainment of exposure
Outcome of interest
Comparability of cohorts
Assessment of outcome
Adequate duration of follow
up
Adequate follow up of cohort
Ito et al.141 2008 a(*) a(*) c a(*) a(*), b(*) b(*) a(*) c 7
Jackson et al.130 2014 c (women) a(*) b(*) a(*) a(*), b(*) b(*) b c 6
Jakobsen et al.109 2012 a(*) a(*) a(*) b a(*), b(*) b(*) a(*) b(*) 8
Jakovljević et al.132 2001 a(*) a(*) a(*) a(*) a(*) b(*) a(*) a(*) 8
Khang et al.146 2007 c (men) a(*) a(*) a(*) a(*), b(*) b(*) a(*) b(*) 8
Kilander et al.67 2001 c (men) a(*) b(*) b a(*), b(*) b(*) a(*) b(*) 7
Kim et al.134 2005 c (women) a(*) b(*) b a(*) b(*) a(*) b(*) 6
Kuper et al.106 2006 c (women) a(*) c b a(*), b(*) b(*) a(*) b(*) 6
Kuper et al.129 2007 c (women) a(*) c a(*) a(*), b(*) b(*) a(*) c 6
Lammintausta et al.31 2012 a(*) a(*) a(*) a(*) - a(*) a(*) d 6
Lapidus & Bengtsson107 1986 c (women) a(*) b(*) a(*) a(*) a(*) a(*) b(*) 7
Lee et al.99 2000 a(*) a(*) b(*) b - a(*) a(*) b(*) 6
Li et al.126 2008 a(*) a(*) b(*) a(*) a(*) b(*) a(*) b(*) 8
Liu et al.143 1982 c (men) a(*) b(*) b a(*) b(*) a(*) b(*) 6
Lynch et al.53 1996 c (men) a(*) b(*) a(*) a(*), b(*) b(*) a(*) c 7
W
in Khaing
Literature R
eview / 44
Table 2.2 Risk of bias assessment of included studies (continued)
Authors Year
Selection Comparability Outcome
Total stars Representativeness
of cohort
Selection of non-exposed cohort
Ascertainment of exposure
Outcome of interest
Comparability of cohorts
Assessment of outcome
Adequate duration of follow
up
Adequate follow up of cohort
Masoudkabir et al.113 2012 a(*) a(*) b(*) a(*) a(*), b(*) a(*) a(*) b(*) 9
Minh et al.142 2003 a(*) a(*) b(*) b a(*) a(*) b d 5
Minh et al.114 2006 a(*) a(*) b(*) b a(*) a(*) a(*) d 6
Notara et al.28 2016 a(*) a(*) b(*) a(*) a(*), b(*) a(*) a(*) b(*) 9
Panagiotakos et al.25 2016 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) c 8
Pednekar et al.136 2011 a(*) a(*) b(*) a(*) a(*) b(*) a(*) b(*) 8
Peter et al.95 2007 a(*) a(*) a(*) a(*) - b(*) b d 5
Philbin et al.120 2001 a(*) a(*) a(*) b a(*) b(*) b b(*) 6
Qureshi et al.135 2003 a(*) a(*) a(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Rasmussen et al.44 2006 a(*) a(*) a(*) a(*) a(*) b(*) a(*) b(*) 8
Rasmussen et al.110 2007 a(*) a(*) a(*) a(*) a(*) b(*) a(*) b(*) 8
Rawshani et al.92 2015 c (dm) a(*) a(*) a(*) a(*), b(*) b(*) a(*) b(*) 8
Rawshani et al.144 2016 c (dm) a(*) a(*) d a(*) b(*) a(*) b(*) 6
Rehkopf et al.89 2015 b(*) a(*) a(*) b a(*), b(*) b(*) a(*) d 7
Table 2.2 Risk of bias assessment of included studies (continued)
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 45
Authors Year
Selection Comparability Outcome
Total stars Representativeness
of cohort
Selection of non-exposed cohort
Ascertainment of exposure
Outcome of interest
Comparability of cohorts
Assessment of outcome
Adequate duration of follow
up
Adequate follow up of cohort
Rossum et al.127 1999 c (women) a(*) b(*) a(*) a(*), b(*) b(*) b c 6
Rosvall et al.123 2006 a(*) a(*) c a(*) a(*) b(*) a(*) b(*) 7
Rosvall et al.145 2008 a(*) a(*) a(*) a(*) a(*) b(*) a(*) d 7
Rosvall et al.147 2008 a(*) a(*) a(*) a(*) a(*) b(*) a(*) d 7
Roux et al.97 2001 a(*) a(*) b(*) a(*) a(*) b(*) a(*) c 7
Salomaa et al.30 2000 a(*) a(*) a(*) a(*) a(*) b(*) a(*) b(*) 8
Schwarz & Elman121 2003 a(*) a(*) b(*) b - b(*) b b(*) 5
Senan & Petrosyan111 2014 b(*) a(*) c a(*) a(*) c a(*) b(*) 6
Siegel et al.116 1987 c (elderly) a(*) b(*) a(*) a(*), b(*) a(*) b d 6
Sui et al.122 2008 a(*) a(*) a(*) b - b(*) b b(*) 5
Thurston et al.93 2005 a(*) a(*) b(*) a(*) a(*), b(*) b(*) a(*) b(*) 9
Tonne et al.140 2005 a(*) a(*) a(*) b a(*) a(*) a(*) d 6
Weikert et al.100 2008 a(*) a(*) b(*) a(*) a(*) b(*) a(*) b(*) 8
Zhou et al.133 2006 a(*) a(*) b(*) a(*) a(*), b(*) a(*) b a(*) 8
Win Khaing Literature Review / 46
Table 2.3 Estimations of pooled effects of education and income on cardiovascular outcomes (co-variates adjusted studies only)
Coronary Artery Diseases Cardiovascular Events
n RR
(95% CI) Q p-value
I2 (%)
n
RR (95% CI)
Q p-value
I2 (%)
Education Medium vs High 15 1.21
(1.06, 1.40) 0.005 96 12 1.27
(1.09, 1.48) 0.003 83
Low vs High 17 1.36
(1.11, 1.66) 0.002 94 13 1.50
(1.17, 1.92) 0.001 99
Income Medium vs High 10 1.27
(1.10, 1.47) 0.001 95 7 1.05
(0.98, 1.13) 0.131 99
Low vs High 10 1.49
(1.16, 1.91) 0.002 98 6 1.17
(0.96, 1.44) 0.117 97
Strokes Cardiovascular Deaths
n RR (95% CI)
Q p-value
I2 (%)
n
RR (95% CI)
Q p-value
I2 (%)
Education Medium vs High 12 1.17
(1.01, 1.35) 0.034 99 28 1.21
(1.12, 1.30) <0.001 98
Low vs High 13 1.23
(1.06, 1.43) 0.005 83 34 1.39
(1.26, 1.54) <0.001 98
Income Medium vs High 6 1.24
(1.00, 1.53) 0.049 99 12 1.34
(1.17, 1.54) <0.001 96
Low vs High 6 1.30
(0.99, 1.72) 0.061 98 21 1.76
(1.45, 2.14) <0.001 99
n, Number of studies; RR, Relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 47
Table 2.4 Pooled education and income effects on cardiovascular outcomes by regions
Education Income
n RR (95% CI) Q p-value I2 n RR (95% CI)
Q p-value I2
Cardiovascular deaths
Asia M vs H 2 1.12 (0.78, 1.60) 0.540 5 0 NA NA NA
L vs H 8 1.34 (1.04, 1.72) 0.024 99 4 1.69 (1.07, 2.67) 0.024 96
Europe M vs H 15 1.17 (1.06, 1.29) 0.001 99 12 1.40 (1.18, 1.67) <0.001 97
L vs H 19 1.32 (1.17, 1.49) <0.001 91 14 1.89 (1.47, 2.44) <0.001 99
US M vs H 14 1.30 (1.14, 1.49) <0.001 72 1 NA NA NA
L vs H 8 1.69 (1.28, 2.22) <0.001 95 4 NA NA NA
CAD
Asia M vs H 3 1.03 (0.85, 1.25) 0.750 28 0 NA NA NA
L vs H 4 1.03 (0.79, 1.33) 0.839 45 0 NA NA NA
Europe M vs H 11 1.04 (0.72, 1.50) 0.852 99 11 1.39 (1.18, 1.63) <0.001 92
L vs H 15 1.24 (0.97, 1.60) 0.086 96 12 1.74 (1.31, 2.32) <0.001 98
US M vs H 4 1.21 (0.97, 1.51) 0.085 75 3 NA NA NA
L vs H 2 1.51 (0.93, 2.45) 0.099 47 1 NA NA NA
CVE
Asia M vs H 2 1.47 (0.82, 2.63) 0.191 61 2 NA NA NA
L vs H 4 1.85 (0.93, 3.70) 0.081 96 2 NA NA NA
Europe M vs H 8 1.26 (1.06, 1.49) 0.090 76 5 1.05 (0.95, 0.37) 0.368 99
L vs H 9 1.36 (1.07, 1.72) 0.011 95 5 1.24 (0.98, 1.58) 0.080 98
US M vs H 5 1.07 (0.69, 1.66) 0.758 78 1 NA NA NA
L vs H 0 NA NA NA 1 NA NA NA
Strokes
Asia M vs H 4 1.22 (0.91, 1.65) 0.192 87 0 NA NA NA
L vs H 5 1.27 (1.07, 1.50) 0.006 34 0 NA NA NA
Europe M vs H 6 1.46 (1.23, 1.72) <0.001 87 5 1.37 (1.24, 1.52) <0.001 70
L vs H 7 1.61 (1.28, 2.02) <0.001 76 5 1.54 (1.33, 1.79) <0.001 64
US M vs H 3 0.98 (0.81, 1.19) 0.848 89 2 0.89 (0.62, 1.27) 0.514 49
L vs H 3 0.99 (0.83, 1.20) 0.957 53 2 0.91 (0.58, 1.41) 0.661 78 M vs H, Medium vs High; L vs H, Low vs High; n, Number of studies; RR, Relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics (%); CAD, Coronary Artery Diseases; CVE, Cardiovascular Events; US, United States; NA, Not available or insufficient data; Table 2.5 Pooled education and income effect on coronary artery diseases (subgroup analyses)
Win Khaing Literature Review / 48
Education Income
n RR (95% CI) Q p-value I2 n RR (95% CI)
Q p-value I2
Number of adjusted variables ≤ 5 M vs H 10 0.97 (0.65, 1.45) 0.888 97 6 1.57 (1.30, 1.91) <0.001 87 L vs H 12 1.22 (0.94, 1.57) 0.130 93 6 2.12 (1.52, 2.96) <0.001 98 > 5 M vs H 8 1.14 (0.98, 1.32) 0.085 95 8 1.14 (0.99, 1.31) 0.059 92 L vs H 9 1.28 (1.02, 1.61) 0.035 89 7 1.29 (0.98, 1.68) 0.066 95 Age (years) ≤ 60 M vs H 12 1.05 (0.70, 1.57) 0.817 99 9 1.42 (1.32, 1.52) <0.001 95 L vs H 12 1.28 (0.99, 1.65) 0.058 85 8 1.83 (1.82, 1.84) <0.001 97 > 60 M vs H 4 1.00 (0.77, 1.30) 0.999 92 2 0.94 (0.73, 1.20) 0.600 89 L vs H 4 1.05 (0.69, 1.59) 0.821 95 2 0.72 (0.51, 1.01) 0.060 18 Male percentage ≤ 60 M vs H 10 0.94 (0.64, 1.38) 0.759 99 8 1.43 (1.16, 1.76) 0.001 91 L vs H 12 1.25 (0.99, 1.58) 0.060 85 8 1.82 (1.30, 2.56) <0.001 96 > 60 M vs H 6 1.26 (0.85, 1.86) 0.246 97 4 1.16 (0.85, 1.59) 0.356 98 L vs H 6 1.25 (0.78, 2.01) 0.359 92 3 1.28 (0.70, 2.32) 0.419 99 Diabetes percentage ≤ 8 M vs H 5 1.16 (0.95, 1.42) 0.136 83 2 NA NA NA L vs H 4 1.25 (0.83, 1.88) 0.295 64 1 NA NA NA > 8 M vs H 4 1.03 (0.87, 1.22) 0.733 87 2 0.94 (0.73, 1.20) 0.600 89 L vs H 5 1.11 (0.84, 1.46) 0.465 89 2 0.72 (0.51, 1.01) 0.060 18 BMI (kg/m2) < 25 M vs H 1 NA NA NA 1 NA NA NA L vs H 2 NA NA NA 1 NA NA NA ≥ 25 M vs H 5 1.16 (1.10, 1.23) <0.001 0 2 1.31 (1.07, 1.59) 0.007 29 L vs H 5 1.30 (1.15, 1.47) <0.001 0 2 1.50 (1.00, 2.26) 0.050 91 Smoking percentage < 30 M vs H 6 1.07 (0.97, 1.19) 0.166 48 2 1.21 (0.90, 1.62) 0.203 69 L vs H 6 1.13 (0.94, 1.37) 0.192 58 2 1.26 (0.53, 2.98) 0.600 85 ≥ 30 M vs H 4 1.21 (0.89, 1.63) 0.225 61 3 1.17 (1.07, 1.29) 0.001 0 L vs H 5 1.32 (0.98, 1.77) 0.066 68 3 1.32 (1.14, 1.53) <0.001 51
M vs H, Medium vs High; L vs H, Low vs High; n, Number of studies; RR, relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics (%);BMI, Body Mass Index; NA, Not available or insufficient data;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 49
Table 2.6 Pooled education and income effect on cardiovascular events (subgroup analyses)
Education Income
n RR (95% CI) Q p-value I2 n RR (95% CI)
Q p-value I2
Number of adjusted variables ≤ 5 M vs H 7 1.25 (1.03, 1.52) 0.027 71 5 1.05 (0.97, 1.13) 0.249 99 L vs H 6 1.69 (1.07, 2.68) 0.025 99 5 1.31 (1.01, 1.69) 0.039 98 > 5 M vs H 8 1.28 (1.03, 1.60) 0.028 74 3 1.11 (1.00, 1.23) 0.052 9 L vs H 7 1.22 (0.98, 1.51) 0.074 72 3 1.16 (1.05, 1.28) 0.004 7 Age (years) ≤ 60 M vs H 7 1.35 (1.06, 1.70) 0.014 61 2 1.09 (1.01, 1.16) 0.018 68 L vs H 8 1.93 (1.35, 2.76) <0.001 92 3 1.34 (1.10, 1.64) 0.004 38 > 60 M vs H 7 1.17 (0.90, 1.53) 0.248 79 4 1.01 (0.93, 1.09) 0.900 99 L vs H 3 1.09 (0.90, 1.31) 0.393 65 4 1.21 (0.92, 1.58) 0.167 99 Male percentage ≤ 60 M vs H 9 1.21 (0.95, 1.55) 0.128 79 5 1.07 (1.00, 1.14) 0.054 89 L vs H 8 1.61 (1.10, 2.37) 0.015 95 6 1.35 (1.18, 1.55) <0.001 88 > 60 M vs H 6 1.31 (1.17, 1.48) <0.001 81 3 1.08 (0.94, 1.23) 0.282 99 L vs H 5 1.17 (1.00, 1.38) 0.047 94 2 0.99 (0.83, 1.19) 0.955 65 Diabetes percentage ≤ 8 M vs H 6 1.11 (1.02, 1.21) 0.020 51 3 1.05 (0.94, 1.17) 0.394 95 L vs H 6 1.23 (0.99, 1.53) 0.058 95 3 1.18 (0.86, 1.63) 0.308 92 > 8 M vs H 7 1.46 (1.07, 1.99) 0.016 74 3 1.07 (1.02, 1.12) 0.005 0 L vs H 4 1.35 (0.83, 2.19) 0.231 72 3 1.17 (1.10, 1.24) <0.001 0 BMI (kg/m2) < 25 M vs H 2 2.14 (1.26, 3.63) 0.005 29 0 NA NA NA L vs H 2 2.26 (1.17, 4.37) 0.016 8 0 NA NA NA ≥ 25 M vs H 5 1.20 (1.06, 1.35) 0.003 5 2 1.09 (1.01, 1.18) 0.025 73 L vs H 5 1.50 (1.16, 1.93) 0.002 74 2 1.35 (1.01, 1.81) 0.043 54 Smoking percentage < 30 M vs H 4 1.18 (0.98, 1.42) 0.078 8 1 NA NA NA L vs H 3 1.61 (1.06, 2.47) 0.027 44 1 NA NA NA ≥ 30 M vs H 5 1.29 (0.95, 1.75) 0.099 77 1 NA NA NA L vs H 5 1.39 (0.93, 2.09) 0.109 88 1 NA NA NA
M vs H, Medium vs High; L vs H, Low vs High; n, Number of studies; RR, relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics (%);BMI, Body Mass Index; NA, Not available or insufficient data;
Win Khaing Literature Review / 50
Table 2.7 Pooled education and income effect on strokes (subgroup analyses)
Education Income
n RR (95% CI) Q p-value I2 n RR (95% CI)
Q p-value I2
Number of adjusted variables ≤ 5 M vs H 4 1.43 (1.15, 1.77) 0.001 90 3 1.41 (1.27, 1.56) <0.001 76 L vs H 6 1.48 (1.17, 1.87) 0.001 85 3 1.38 (1.24, 1.52) <0.001 5 > 5 M vs H 9 1.13 (0.93, 1.36) 0.221 95 4 1.10 (0.86, 1.40) 0.463 78 L vs H 9 1.23 (0.99, 1.53) 0.055 80 4 1.32 (0.79, 2.20) 0.292 93 Age (years) ≤ 60 M vs H 5 1.41 (1.32, 1.50) <0.001 0 1 NA NA NA L vs H 5 1.54 (1.30, 1.83) <0.001 0 1 NA NA NA > 60 M vs H 6 1.23 (0.93, 1.63) 0.147 99 5 1.31 (1.19, 1.45) <0.001 99 L vs H 6 1.31 (0.97, 1.75) 0.073 95 5 1.26 (1.19, 1.33) <0.001 83 Male percentage ≤ 60 M vs H 10 1.26 (1.05, 1.50) 0.011 99 6 1.32 (1.14, 1.53) <0.001 99 L vs H 12 1.37 (1.15, 1.63) <0.001 87 6 1.40 (1.16, 1.68) <0.001 94 > 60 M vs H 1 NA NA NA 0 NA NA NA L vs H 1 NA NA NA 0 NA NA NA Diabetes percentage ≤ 8 M vs H 6 1.12 (0.83, 1.51) 0.445 93 1 NA NA NA L vs H 6 1.28 (1.05, 1.57) 0.014 33 1 NA NA NA > 8 M vs H 5 1.37 (1.15, 1.63) <0.001 95 4 1.25 (1.06, 1.47) 0.008 99 L vs H 6 1.48 (1.15, 1.89) 0.002 93 4 1.31 (1.30, 1.31) <0.001 98 BMI (kg/m2) < 25 M vs H 1 NA NA NA 0 NA NA NA L vs H 2 NA NA NA 0 NA NA NA ≥ 25 M vs H 5 1.35 (1.12, 1.64) 0.002 76 3 1.19 (0.94, 1.51) 0.139 72 L vs H 5 1.55 (1.07, 2.23) 0.019 80 3 1.61 (1.01, 2.55) 0.044 90 Smoking percentage < 30 M vs H 11 1.26 (1.07, 1.50) 0.007 99 5 1.29 (1.09, 1.54) 0.003 99 L vs H 11 1.35 (1.13, 1.60) 0.001 89 5 1.41 (1.11, 1.79) 0.006 96 ≥ 30 M vs H 1 NA NA NA 0 NA NA NA L vs H 2 NA NA NA 0 NA NA NA
M vs H, Medium vs High; L vs H, Low vs High; n, Number of studies; RR, relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics (%);BMI, Body Mass Index; NA, Not available or insufficient data;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 51
Table 2.8 Pooled education and income effect on cardiovascular deaths (subgroup analyses)
Education Income
n RR (95% CI) Q p-value I2 n RR (95% CI)
Q p-value I2
Number of adjusted variables ≤ 5 M vs H 14 1.29 (1.15, 1.44) <0.001 51 6 1.34 (1.13, 1.60) 0.001 95 L vs H 18 1.53 (1.31, 1.79) <0.001 99 12 1.81 (1.40, 2.34) <0.001 99 > 5 M vs H 17 1.16 (1.07, 1.26) 0.001 98 7 1.37 (1.07, 1.76) 0.013 96 L vs H 17 1.28 (1.14, 1.44) <0.001 88 10 1.73 (1.30, 2.30) <0.001 97 Age (years) ≤ 60 M vs H 18 1.26 (1.16, 1.38) <0.001 98 4 1.43 (1.10, 1.87) 0.008 96 L vs H 18 1.53 (1.31, 1.78) <0.001 99 9 1.94 (1.40, 2.71) <0.001 99 > 60 M vs H 9 1.18 (1.00, 1.39) 0.047 81 5 1.26 (1.08, 1.47) 0.004 93 L vs H 7 1.21 (1.05, 1.40) 0.009 78 8 1.65 (1.30, 2.09) <0.001 98 Male percentage ≤ 60 M vs H 13 1.19 (1.00, 1.29) <0.001 98 6 1.55 (1.25, 1.92) <0.001 88 L vs H 18 1.35 (1.20, 1.53) <0.001 99 9 2.31 (1.72, 3.10) <0.001 95 > 60 M vs H 18 1.25 (1.08, 1.45) 0.002 88 6 1.14 (1.01, 1.28) 0.028 93 L vs H 15 1.41 (1.17, 1.69) <0.001 91 11 1.35 (1.19, 1.54) <0.001 96 Diabetes percentage ≤ 8 M vs H 4 1.15 (0.93, 1.42) 0.205 85 2 1.12 (1.06, 1.18) <0.001 61 L vs H 2 1.22 (0.78, 1.90) 0.375 89 2 1.33 (1.18, 1.50) <0.001 93 > 8 M vs H 9 1.23 (1.05, 1.45) 0.012 94 4 1.47 (1.20, 1.81) <0.001 83 L vs H 9 1.24 (1.11, 1.40) <0.001 65 6 1.76 (1.32, 2.36) <0.001 88 BMI (kg/m2) < 25 M vs H 0 NA NA NA 0 NA NA NA L vs H 1 NA NA NA 0 NA NA NA ≥ 25 M vs H 9 1.27 (1.09, 1.47) 0.002 84 2 1.71 (1.29, 2.27) <0.001 54 L vs H 7 1.21 (1.07, 1.36) 0.002 2 2 2.49 (1.39, 4.47) 0.002 95 Smoking percentage < 30 M vs H 10 1.21 (1.10, 1.33) <0.001 99 3 1.65 (1.39, 1.97) <0.001 34 L vs H 12 1.39 (1.18, 1.65) <0.001 99 5 2.38 (1.71, 3.33) <0.001 83 ≥ 30 M vs H 11 1.18 (0.95, 1.47) 0.129 64 0 NA NA NA L vs H 12 1.42 (1.06, 1.89) 0.018 81 3 1.32 (1.21, 1.44) <0.001 38
M vs H, Medium vs High; L vs H, Low vs High; n, Number of studies; RR, relative risk; CI, Confidence Interval; Q p-value, p value for Q test for heterogeneity, I2, I2 statistics (%);BMI, Body Mass Index; NA, Not available or insufficient data;
Win Khaing Literature Review / 52
PubMed search (n=354) Scopus search (n=1335)
Duplicates (n=115)
Record screened (n=1585)
Full text articles assessed for eligibility (n = 186)
Records excluded based on titles and abstract review
941 Non-CVD374 Not include study factors36 Non-English16 Narrative review14 Systematic review7 Commentary5 No full-text available3 Letter 1 Protocol1 Guidelines1 Book
Studies included in qualitative synthesis
(n = 72)
Full text articles excluded 41 Not cardiovascular outcomes studies 35 Non-cohort design23 No outcome of interest12 Study factors as co-variate/control factors3 Not sufficient for data extraction
References lists (n=11)
Studies included in meta-analysis (n = 72)
Education (n = 62)
Income (n = 39 )
Coronary Diseases (n = 23)
Cardiovascular Events (n = 18)
Cardiovascular Death (n = 42)
Cerebrovascular Diseases (n = 15)
Coronary Diseases (n = 15)
Cardiovascular Events (n = 9)
Cardiovascular Death (n = 22)
Cerebrovascular Diseases (n = 7)
Figure 2.1 Flow diagram for selection of studies
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 53
Figure 2.2 Pooling effects of education on cardiovascular outcomes
Win K
haing
Literature Review
/ 54
Figure 2.3 Funnel plots of relative risks of cardiovascular outcomes among medium versus high education levels
Fac. of Grad. Studies, M
ahidol Univ.
Ph.D
.(Clinical Epidem
iology) / 55
Figure 2.4 Funnel plots of relative risks of cardiovascular outcomes among low versus high education levels
Win K
haing
Literature Review
/ 56
Figure 2.5 Contour-enhanced plots of relative risks of cardiovascular outcomes among medium versus high education levels
Fac. of Grad. Studies, M
ahidol Univ.
Ph.D
.(Clinical Epidem
iology) / 57
Figure 2.6 Contour-enhanced plot of relative risks of cardiovascular outcomes among low versus high education levels
Win Khaing Literature Review / 58
Figure 2.7 Pooling effects of income on cardiovascular outcomes
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 59
Figure 2.8 Funnel plots of relative risks of cardiovascular outcomes among medium versus high income levels
Win K
haing
Literature Review
/ 60
Figure 2.9 Funnel plots of relative risks of cardiovascular outcomes among low versus high income levels
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 61
Figure 2.10 Contour-enhanced plots of relative risks of cardiovascular outcomes among medium versus high income levels
Win K
haing
Literature Review
/ 62
Figure 2.11 Contour-enhanced plots of relative risks of cardiovascular outcomes among low versus high income level
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 63
Figure 2.12 Conceptual framework. Direct effect shown in solid line, and mediated
effect shown in dashed line
Education Income
Confounders • Demographic: age, sex,
marital status, • Physical Examination: height,
weight, waist, hip, SBP, DBP • Behavioral factors: smoking,
alcohol, exercise • Underlying diseases:
hypertension, diabetes, stroke/TIA, CKD, dyslipidemia
• Medication history: hypertension, diabetes, dyslipidemia
• Family history: hypertension, diabetes, dyslipidemia
Major Cardiovascular Events
• Myocardial infraction
• Stroke/TIA
• Cardiovascular death
Win Khaing Methodology / 64
CHAPTER III
METHODOLOGY
3.1 Study design and setting The design of this study was a prospective cohort study of EGAT148. Briefly,
the EGAT cohort was collaboration between Ramathibodi Hospital and EGAT, which
consists of 3 parallel cohorts, i.e., first cohort (called EGAT1), second cohort (called
EGAT2), and third cohort (called EGAT3). A total number of 9,082 subjects were
randomly selected and enrolled in 1985, 1998 and 2009 for EGAT1, EGAT2, EGAT3,
respectively with corresponding sample sizes of 3,499, 2,999, and 2,584. All subjects
gave written informed consent and underwent complete medical examination with self-
administered questionnaires and thorough laboratory tests. Then, they were regularly
resurveyed every 5 years, except for 12 years follow-up between the first (1985) and the
second (1997) survey of EGAT1/1 and EGAT1/2.
This current study used data of the EGAT1 cohort (see, Figure 3.1) because
it mainly covered the detailed information about education, income, and CVRFs. The
data from the second survey of the EGAT1 cohort in 1997 (called EGAT1/2) was used
as baseline data, and they were followed up in 2002 (EGAT1/3), 2007 (EGAT1/4) and
2012 (EGAT1/5). The 15 years duration of follow-up was considered to be long enough
to develop interested outcomes (i.e., MI, stroke/TIA, and CVD death).
3.2 Study subjects All studied subjects from the second to fifth survey of the EGAT1 cohort
(i.e., EGAT1/2, EGAT1/3, EGAT1/4 and EGAT1/5) were included in the study. The
studied subjects were excluded if subjects had developed any of MCVEs (i.e., MI,
ischemic stroke/TIA, and CVD death) before or at the date of enrollment in 1997.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 65
3.3 Study factors and measurements
3.3.1 Education
Education level was extracted from the self-administered questionnaires of
EGAT1/2, EGAT1/3, and EGAT 1/4. In the EGAT1/5 questionnaire, as there was no
question asking about patient’s education, education level of EGAT1/5 was assumed to
be the same as education level of EGAT1/4. Education levels were re-categorized into
3 groups as low (≤ high school), medium (vocational / diploma) and high (bachelor /
master / PhD).
3.3.2 Income
Income was extracted from the self- administered EGAT1/2, EGAT1/3,
EGAT1/4, and EGAT1/5 questionnaires. Income was re-categorized into 3 groups: low
(<20,000 Baht), medium (20,000 – 50,000 Baht), and high (>50,000 Baht).
3.4 Outcome of interest
3.4.1 Interested outcomes
The outcome of interest was incidence of MCVE which were composite
endpoints of death from CVE, MI, ischemic stroke, and TIA.
3.4.2 Definition of outcomes
3.4.2.1 Myocardial Infarction
Myocardial infarction is defined as “a condition of myocardial
necrosis that occurs due to myocardial ischemia”. The third universal definition of
myocardial infarction149 was used for diagnosis of myocardial infarct in this study.
3.4.2.2 Stroke
Stroke is defined as “a neurological deficit attributed to an acute
focal injury of the central nervous system by a vascular cause, including cerebral
infarction, intracerebral hemorrhage and subarachnoid hemorrhage”. AHA/ASA expert
Win Khaing Methodology / 66
consensus document and an updated definition of stroke for 21st century150 was used for
diagnosis of stroke in this study.
3.4.2.3 Transient ischemic attack
Transient ischemic attack is defined as “brief episodes of
neurological dysfunction resulting from focal cerebral ischemia not associated with
permanent cerebral infraction”. AHA/ASA scientific statement and definition and
evaluation of transient ischemic attack151 was used for diagnosis of TIA in this study.
3.4.2.4 Cardiovascular death
Cardiovascular death is defined as “death from coronary artery
disease including myocardial infarction, sudden cardiac death or ischemic stroke”.
3.4.3 Outcomes verification
All outcomes of interest as well as the date of events were verified by a
panel outcome verification team, which consisted of 2 cardiologists, one neurologist,
one oncologist, and one internist. The subjects’ status had been verified until on 31st
December, 2015. The outcomes were verified in a previous study152 with various
methods as follows:
3.4.3.1 MCVE outcomes
All MCVE outcomes and dates of occurrence were detected and
checked from the following sources:
• Documents of EGAT 1/2, 1/3, 1/4 and 1/5 surveys,
which were taken at 5 years intervals. The surveys consisted of thorough subjects’
history with physical examination records and investigation results including
Electrocardiography (EKG), Chest X-ray (CXR), and etcetera.
• The reimbursement information of EGAT cohort
had already been obtained from EGAT office and hospitals.
• The 3 government reimbursement schemes data
from Comptroller General’s Department, the National Health Security Office and the
Social Security Office, which covered >99% of health reimbursement of Thai people
had been obtained by contacting to Central Office for Health Care Information.
• If in doubt about MCVE outcomes, telephone
interview had been made and hospital medical records had been acquired. If essential,
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 67
other necessary investigations (e.g., CT brain, MRI brain, coronary angiography) results
had been retrieved.
3.4.3.1 Mortality outcomes
Death/alive status of all subjects was checked by requesting data
from the Bureau of Policy and Strategy, Ministry of Public Health and death certificate
databases from the Ministry of Interior.
The causes of death were confirmed with patients’ death
certificates and by telephone interview to the passed away patients’ relatives. If the
patient died at hospital, the patients’ medical records were retrieved. The causes of death
were determined by the consensus of the outcome verification team.
3.5 Others risk factors and measurements Other risk factors of MCVE collected have been shown below.
3.5.1 Self-administered data
3.5.1.1 Demographic data
• Completed age was measured in years. If age was
not reported, age was calculated by subtracting survey date from date of birth.
• Sex was categorized as male or female.
• Marital status was recorded as single or married or
divorced or widowed or separate.
3.5.1.2 Behavior risk factors
• History of smoking status was asked on smoking
status (never or current or regular or quit), duration of smoking (in years), quantity of
smoking (in cigarette per day), age at start smoking and last smoking, and then smoking
status was re-categorized as never or ex-smoker or current.
• History of alcohol consumption was asked on
drinking status (never or current or regular or quit), type of alcohol (liquor, beer, wine,
spirits), frequency of drinking (amount in liters per week, month), duration of drinking
(in years), and then alcohol consumption was re-categorized as never or ex-drinker or
current.
Win Khaing Methodology / 68
• History of doing exercise was asked on exercise
(none or exercise done), length of doing exercise time, frequency of doing exercise per
week, and then exercise status was re-categorized as none or 1 – 2 times/week or ≥3
times/week.
3.5.1.3 Underlying diseases
• History of having underlying diseases including
diabetes, hypertension, dyslipidemia, coronary artery disease, stroke/TIA, chronic
kidney diseases was asked.
3.5.1.4 Family history
• Family history of having diabetes, hypertension,
dyslipidemia, coronary artery disease, stroke/TIA, chronic kidney diseases was asked.
3.5.1.5 Medication history
• History of prescribed treatment for diabetes,
hypertension, and dyslipidemia was asked.
3.5.2 Physical examination data
3.5.2.1 Height and weight
• Subjects’ height was measured in centimeters and
weight was measured in kilograms with dressed in normal clothing without shoes.
3.5.2.2 Hip and waist circumferences
• Subjects’ hip and waist circumferences were
measured in centimeters by measuring tape. Waist circumference was then categorized
as: normal and high with cutoff value of >90 cm in male and >80 cm in female153.
3.5.2.3 SBP and DBP
• SBP and DBP were measured after 5 minutes rest
using a calibrated mercury sphygmomanometer and the first and fifth Krokoff sounds
were recorded, respectively. They were measured twice by sitting position.
3.5.3 Laboratory data
- Blood samples were collected after 12-hour overnight fasting. Blood
glucose was measured by plasma samples in mg/dL (Peridochrome, Boehringer
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 69
Mannheim, Mannheim, Germany)154. Serum total cholesterol (TC), triglyceride (TG),
low density lipoprotein (LDL), and HDL were measured in mg/dL using enzymatic-
calorimetric assays (Boehringer Mannheim, Mannheim, Germany)154. Serum creatinine
was measured in mg/dL using a modified kinetic Jaffe reaction155 by the same
laboratory.
3.5.4 Definition and criteria for classifications
3.5.4.1 Diabetes
Diabetes was diagnosed if the participant had history of diabetes
or had fasting blood sugar ≥126 mg/dL, or used anti-diabetes medications156, e.g.,
Metformin, Insulin, Diamicron, Glucophage, Gilbenclamide, Pioglitazone, and etcetera.
3.5.4.2 Hypertension
Hypertension was diagnosed if the participant had history of
hypertension, SBP ≥140 mmHg or DBP ≥90 mmHg, or had been taking prescribed BP
lowering medications157, e.g., Amlodipine, Blopress, Coversyl, Enalapril, Losartan,
Nifidipine, and etcetera.
3.5.4.3 Body mass index
BMI was calculated from the recorded weight in kilograms
divided by squared height in meters. BMI was categorized as: underweight (<18.5
kg/m2), normal (18.5 – 22.9 kg/m2), overweight (23 – 26.9 kg/m2) and obese (≥27
kg/m2)158.
3.5.4.4 Waist-hip ratio
Waist-hip ratio (WHR) was calculated from the recorded waist
in centimeters divided by hip in centimeters. WHR was categorized as: normal and high
with cutoff value of ≥ 0.9 in male or ≥ 0.85 in female159.
3.5.4.5 Hypercholesterolemia
Hypercholesterolemia was diagnosed if the subjects had fasting
serum cholesterol ≥200 mg/dL160, or had history of hypercholesterolemia or used
cholesterol-lowering medications, e.g., atorvastatin (Lipitor), rosuvastatin (Crestor),
simvastatin (Zocor), and etcetera.
3.5.4.6 Hypertriglyceridemia
Win Khaing Methodology / 70
Hypertriglyceridemia was diagnosed if the subjects had fasting
serum triglyceride ≥150 mg/dL160, or had history of hypertriglyceridemia or used
triglyceride-lowering medications, e.g., atorvastatin (Lipitor), rosuvastatin (Crestor),
simvastatin (Zocor), and etcetera.
3.5.4.7 High density lipoprotein
HDL was classified as low/normal or high with the cutoff value
of subjects’ fasting serum HDL ≥50 mg/dL in male or HDL ≥40 mg/dL in female160, or
used any HDL-raising medication.
3.5.4.8 Low density lipoprotein
LDL was classified as low/normal and high with the cutoff value
of subjects’ fasting serum LDL ≥160 mg/dL160, or used any LDL-lowering medication.
3.5.4.9 Dyslipidemia
Dyslipidemia was classified using Guidelines for the
Management of Dyslipidemias160. The subjects were classified as having dyslipidemia
if they had at least four of the following criteria: 1) HDL <50 mg/dL in male or HDL
<40 mg/dL in female; 2) LDL ≥160 mg/dL 3) triglyceride ≥150 mg/dL 4) used any lipid-
lowering medications, e.g., atorvastatin (Lipitor), rosuvastatin (Crestor), simvastatin
(Zocor), and etcetera.
3.5.4.10 Chronic kidney disease
The chronic kidney disease (CKD) was defined and categorized
by the estimated glomerular filtration rate (eGFR). The eGFR was calculated using the
Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI: 2009) equation161.
Using eGFR, severity of CKD was classified into five stages as stage 1, stage 2, stage
3, stage 4, and stage 5 by using cutoff values of >90, 60-89, 30-59, 15-29, and <15
ml/min/1.73 m2, respectively. The CKD stages 1 and 2 were combined and compared
with CKD stages ≥3.
3.6 Data collection From main EGAT1 database, data recorded at baseline in 1997 (EGAT1/2)
and subsequent follow up data in 2002 (EGAT1/3), 2007 (EGAT1/4), and 2012
(EGAT1/5) were retrieved. They were collected by self-administered questionnaires,
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 71
which consisted of general demographic data (age, gender, educational level, income,
marital status), behavioral data (smoking status, alcohol consumption, exercise/physical
activity), family history of illness, underlying diseases (diabetes, hypertension,
stroke/TIA, chronic kidney disease, dyslipidemia), and use of medication for
hypertension, diabetes, or lipids. In addition, data of physical examination done by
clinicians, cardiologists and trained personnel from Ramathibodi Hospital was also
retrieved. These include weight, height, waist circumference, hip circumference, SBP,
DBP. Blood was collected after fasting overnight for 12 hours before blood
examination. Complete blood count, fasting plasma glucose, lipid profile (total
cholesterol, LDL, HDL, triglyceride), creatinine, and uric acid were retrieved. EKG and
CXR data were also be retrieved.
3.7 Sample size estimation Sample size estimation was calculated based on more than two groups of
proportions calculation technique. Our meta-analysis42 showed that the proportion of
CVE in high, medium and low education groups ranged from 2% to 28%, 3% to 27%,
and 5% to 38%, respectively. Exploring distributions of our cohort data indicated that
ratios of subjects with high versus medium education, and high to low education in our
cohort were 1:3 and 1:5, respectively. Type I and Type II errors were set at 5% and 20%,
respectively. Assigning rate of CVE in high education was 2%, minimal detectable sizes
were 1% and 4% for medium and low versus high education, respectively. Sample size
was calculated using Stata version 14.1, with the following commands162:
. artbin, pr(.02 .03 .05) ngroup(3) aratio(1 3 5) alpha(.05) power(.8)
ART - ANALYSIS OF RESOURCES FOR TRIALS (version 1.0.0, 3 March 2004)
------------------------------------------------------------------------------
A sample size program by Abdel Babiker, Patrick Royston & Friederike Barthel,
MRC Clinical Trials Unit, London NW1 2DA, UK.
------------------------------------------------------------------------------
Type of trial Superiority - binary outcome
Statistical test assumed Unconditional comparison of 3
binomial proportions
Number of groups 3
Allocation ratio 1.00:3.00.00:5.00
Anticipated event probabilities 0.020, 0.030, 0.050
Win Khaing Methodology / 72
Alpha 0.050 (two-sided)
Power (designed) 0.800
Total sample size (calculated) 2775
Expected total number of events 111
------------------------------------------------------------------------------
For income, the proportion of CVE in high, medium and low income groups
ranged from 2% to 24%, 3% to 25%, and 5% to 39%, respectively and ratio of subjects
with high to medium income and high to low income were 1:1 and 1:2, respectively.
Type I and Type II errors were set at 5% and 20%, respectively. Assigning rate of CVE
in high income was 2%, minimal detectable sizes were 1% and 4% for medium and low
versus high income, respectively. Sample size was calculated using Stata version 14.1,
with the following commands162:
. artbin, pr(.02 .03 .05) ngroup(3) aratio(1 1 2) alpha(.05) power(.8)
ART - ANALYSIS OF RESOURCES FOR TRIALS (version 1.0.0, 3 March 2004)
------------------------------------------------------------------------------
A sample size program by Abdel Babiker, Patrick Royston & Friederike Barthel,
MRC Clinical Trials Unit, London NW1 2DA, UK.
------------------------------------------------------------------------------
Type of trial Superiority - binary outcome
Statistical test assumed Unconditional comparison of 3
binomial proportions
Number of groups 3
Allocation ratio 1.00:1.00.00:2.00
Anticipated event probabilities 0.020, 0.030, 0.050
Alpha 0.050 (two-sided)
Power (designed) 0.800
Total sample size (calculated) 2061
Expected total number of events 78
------------------------------------------------------------------------------
Comparing estimated sample sizes between the two calculations, the larger
sample size was used. Therefore, a total sample size of at least 2,775 was needed for
this study.
3.8 Data management The flow of data management was shown in Figure 3.2. Cross-sectional
data of cohort EGAT1/2, EGAT1/3, EGAT1/4, and EGAT1/5 were retrieved from the
main EGAT1 databases including
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 73
• General demographic data (age, gender, educational level,
income, marital status),
• Behavioral data (smoking status, alcohol consumption,
exercise/physical activity),
• Family history of illness, underlying diseases (diabetes,
hypertension, stroke/TIA, chronic kidney disease, dyslipidemia),
• Use of medication (hypertension, diabetes, and lipid lowering
drugs), and
• Physical examination data (weight, height, waist circumference,
hip circumference, SBP, DBP).
Each baseline database was then merged with their corresponding
laboratory data which consisted of complete blood count, fasting plasma glucose, total
cholesterol, LDL, HDL, triglyceride, creatinine, and uric acid. Then, these 4 main
databases (baseline plus laboratory data) were merged. Finally, they were merged with
the outcome database which consisted of interested outcomes (i.e., MI, ischemic
stroke/TIA, and CVD death) as well as the date of occurrence of each event.
Databases including individual cross-sectional and outcome data were then
summarized and checked for inconsistency, possible values, missing data and outliers
(<4SD or >4SD) for each variable. Original source documents (i.e., scanned case record
form) were also used to check for validation of data if required.
Age variables were cross validated with the information from EGAT1/1,
1/2, 1/3, 1/4, and 1/5. If age was missing, it was calculated by survey date minus date of
birth. If there was no date of birth recorded, age was calculated by adding 5 years for
forward cohort or by subtracting 5 years for backward cohort.
By assuming sex and height were constant across all cohort period, sex and
height were cross checked with the information of every cohort period. For example, if
participant’s height was inconsistent, missing or out-of-range, it could be corrected or
filled in using data from other cohort database of the same individual as shown in Table
3.1.
For education variables, consistency and possible levels of education were
checked. For instance, if subject had bachelor degree in the current visit, his/her
education could not be lower than bachelor in the next follow up. The same concepts
Win Khaing Methodology / 74
were applied to marital status, smoking status and alcohol consumptions. For marital
status, if subject was married or widow or separate or divorce in the current visit, he/she
could not be single in the next follow up. If subject was current or ex-smoker/ drinker,
he/she could not be none smoker/ drinker in the subsequent visit. Examples of correcting
or filling these variables were shown in Table 3.2.
Other remaining variables (i.e., income, exercise, weight, waist
circumference, hip circumference, SBP, DBP, glucose, TC, TG, LDL, HDL, creatinine,
and uric acid) were also summarized and checked for outliers (<4SD or >4SD). If there
were outliers, their values were checked with the original scanned case record forms.
If those problems (missing or outlier) were unable to be verified by those
methods mentioned above, imputation of data was then performed. After imputation,
data was then summarized again and checked for inconsistency, possible values, missing
data and outliers (<4SD or >4SD) for each variable as described above.
3.9 Imputation
3.9.1 Imputation methods
Missing data were imputed by a simulation-based approach using multiple
imputation with chained equations (MICE) method163, 164 with the assumption that data
were missing at random (MAR). First, patterns or distributions of missing variables
were explored to check for the MAR assumption. Multiple chain imputations for
longitudinal data165 were performed using Stata version 14.2166. Detail of predicted
variables for each chain equation was summarized in Table 3.3, and detail of commands
used for MICE was described in Appendix C.
3.9.2 Imputed and predictive variables
Variables that were used for imputation were: 1) study factors (i.e.,
education, income); 2) outcomes variables (i.e., composite of MCVE and Death due to
MCVE); and 3) covariates (i.e., age, sex, marital status, smoking, alcohol, exercise,
height, weight, waist, hip, SBP, mediation history of lipid and NSAID, family history
of hypertension, diabetes and dyslipidemia, fasting plasma glucose, cholesterol,
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 75
triglyceride, HDL, LDL, uric acid, and creatinine). Some of these variables were
imputed variables, some were predictors, and some were both variables.
3.9.3 Imputation modeling
There were 20 variables needed to impute. The usage of models for
imputations were as follows: the ordinal logit model was used for imputation of ordinal
variable (i.e., income), the multi-nominal logistic regression model was used for
imputation of nominal variable (i.e., exercise), the logistic regression model was used
for imputation of categorical variable (i.e., medication of lipid drug), and the linear
regression model was used for imputation of continuous variables (i.e., height, weight,
waist circumference, hip circumference, SBP, DBP).
In addition, a quality control was also applied to each individual chain
equation depending on type of variable. For example, education level of current visit
could not be lower than the subsequent visit. The same concept was also applied to
marital status, smoking, alcohol. Proper ranges of some lab tests (i.e., blood glucose
level, cholesterol, triglyceride, LDL, HDL, uric acid and creatinine levels) were
assigned by using an interval linear regression model by setting up the possible upper
and lower values for each variable in each cohort period.
MICE were not applied to some composite variables. For example, BMI was
not imputed because BMI can be calculated using imputed height and weight. Also
diabetes status and hypertension status can be filled in later by using the imputed original
continuous variables like fasting plasma glucose, SBP, DBP.
Each missing variable was modelled with separate regression model with
conditionally on the remaining variables in the data set until all missing data were filled.
Finally, the observed data and the final set of imputed data were then used to create one
“complete” data set.
3.9.4 Numbers and performance of imputations
The percentage of missing data were ranged from 1.9% to 65.5%. Therefore,
70 imputations were applied. Then, the performance of imputations were assessed using
the largest fraction of missing information (FMI) of coefficient estimates due to missing
data and the average relative increase in variance (RVI)167 of estimate, because of
Win Khaing Methodology / 76
missing data. In addition, diagnostic plot was constructed by comparing the distributions
between missing versus observed values to examine the performances of imputations.
3.9.5 Data management after multiple imputation
After imputation, imputed data were checked again for meaningfulness and
possibility value for each variable. For example, the current education could not be low
education if the previous education was high. Similarly, quit or current smokers cannot
be turned back to be never smokers.
Subjects who had MCVE outcome before enrollment of EGAT1/2 were
excluded. All edited data were discussed and approved with the data management team.
Further statistical analysis was performed on mi imputed dataset and “mi
estimate” prefix Stata command was used for every statistical analysis to adjusts
coefficients and standard errors for the variability between imputations according to the
combination using Rubin's rules168.
3.10 Data analysis Subjects’ characteristics of the cohort in EGAT1/2, 1/3, 1/4, and 1/5 were
described using mean and standard deviation or median and range where appropriate for
continuous data, and using frequency and percentage for categorical data.
3.10.1 Education → Income → MCVE
A causal model of education → income → MCVE was displayed in Figure
3.3. For this model, education was considered as independent variable, income was a
mediator, and MCVE was the outcome of interest.
A mediation analysis was conducted using rationale and statistical
procedures outlined by Baron and Kenny169 and MacKinnon and his colleagues80, 170.
For mediation analysis with categorical variables, concepts proposed by Iacobucci171
and MacKinnon and Cox172 were used.
𝑌𝑌 = 𝑖𝑖1 + 𝑐𝑐𝑐𝑐 + 𝑒𝑒1 (1)
𝑌𝑌 = 𝑖𝑖2 + 𝑐𝑐′𝑐𝑐 + 𝑏𝑏𝑏𝑏 + 𝑒𝑒2 (2)
𝑏𝑏 = 𝑖𝑖3 + 𝑎𝑎𝑐𝑐 + 𝑒𝑒3 (3)
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 77
Briefly, income is said to be mediated if (a) education has a statistically
significant effect on the income (exposure-mediator model) (b) the income is associated
with the MCVE after controlling for the education effect (exposure-mediator-outcome
model) (c) the mediated effect is statistically significant (Figure 3.3).
3.10.1.1 Mediation model
The mediation model was constructed by fitting education on
income using generalized structural equation modelling (GSEM).173, 174 A univariate
GSEM model with a multinomial logit link function was used to screen each variable
that might associate with income including age, sex, and marital status. These variables
were then considered in the multi-variate model if their p-value <0.1. Forward selection
was applied to select significant variables being kept in the mediation model with
containing education. Finally, four equations from causal pathway were constructed,
called path a11, a12, a21, a22, see equation 4, 5, 6, and 7 (Figure 3.4).
For path a11
ln �𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖1|𝑒𝑒𝑒𝑒𝑒𝑒1)𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖3|𝑒𝑒𝑒𝑒𝑒𝑒3)
� = 𝑎𝑎011 + 𝑎𝑎11𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝜀𝜀𝑚𝑚 (4)
For path a12
ln �𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖2|𝑒𝑒𝑒𝑒𝑒𝑒1)𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖3|𝑒𝑒𝑒𝑒𝑒𝑒3)
� = 𝑎𝑎012+ 𝑎𝑎12𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝜀𝜀𝑚𝑚 (5)
For path a21
ln �𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖1|𝑒𝑒𝑒𝑒𝑒𝑒2)𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖3|𝑒𝑒𝑒𝑒𝑒𝑒3)
� = 𝑎𝑎021 + 𝑎𝑎21𝑒𝑒𝑒𝑒𝑒𝑒2 + 𝜀𝜀𝑚𝑚 (6)
For path a22
ln �𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖2|𝑒𝑒𝑒𝑒𝑒𝑒2)𝑃𝑃(𝑖𝑖𝑖𝑖𝑖𝑖3|𝑒𝑒𝑒𝑒𝑒𝑒3)
� = 𝑎𝑎022 + 𝑎𝑎22𝑒𝑒𝑒𝑒𝑒𝑒2 + 𝜀𝜀𝑚𝑚 (7)
where, 𝑒𝑒𝑒𝑒𝑒𝑒1 = Low education
𝑒𝑒𝑒𝑒𝑒𝑒2 = Medium education
𝑒𝑒𝑒𝑒𝑒𝑒3 = High education
𝑖𝑖𝑖𝑖𝑐𝑐1 = Low income
𝑖𝑖𝑖𝑖𝑐𝑐2 = Medium income
𝑖𝑖𝑖𝑖𝑐𝑐3 = High income
3.10.1.2 Outcome model
Win Khaing Methodology / 78
The outcome pathway was performed using GSEM approach.
MCVE was considered as dichotomous outcome (yes or no), and education was
considered as exposure and income was considered as mediator. In addition, other co-
variables including age, sex, marital status, smoking, alcohol, exercise, BMI, WHR,
waist circumference, family history (hypertension, diabetes, and dyslipidemia),
underlying diseases (hypertension, diabetes, dyslipidemia, and chronic kidney disease)
were also considered. A univariate GSEM model with a family of Bernoulli and a logit
link was used on each of those co-variables which might associate with MCVE. The
multivariate logistic regression model was then fitted by forward including these
variables with p-value < 0.10 in the MCVE model that already contained education and
income.
Two equations from causal pathway were constructed including
path b1, b2, 𝑐𝑐′1, 𝑐𝑐′2 , see equation 8 and 9 (Figure 3.4).
ln � 𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)1−𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)
� = 𝑏𝑏01 + 𝑏𝑏1𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝑐𝑐′1𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝜀𝜀𝑦𝑦 (8)
ln � 𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)1−𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)
� = 𝑏𝑏02 + 𝑏𝑏2𝑖𝑖𝑖𝑖𝑐𝑐2 + 𝑐𝑐′2𝑒𝑒𝑒𝑒𝑒𝑒2 + 𝜀𝜀𝑦𝑦 (9)
3.10.1.3 Estimation of mediation effects
Finally, the two mediation and outcome models were
simultaneously combined using GSEM approach. Estimated coefficients were used to
decompose the total effects of education on MCVE into direct effect (DE) and mediation
or indirect175 as follows:
The mediation effects (ME):176
(a) Low versus high education and low income
𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎11𝑏𝑏1
(b) Low versus high education and medium income
𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎12𝑏𝑏2
(c) Medium versus high education and low income
𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎21𝑏𝑏1
(d) Medium versus high education and medium income
𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎22𝑏𝑏2
The total effects (TE) were estimated as follows:
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 79
(a) Total effect of low education on MCVE through low income
𝑇𝑇𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎11𝑏𝑏1 + 𝑐𝑐′1
(b) Total effect of low education on MCVE through medium
income
𝑇𝑇𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎12𝑏𝑏2 + 𝑐𝑐′1
(c) Total effect of medium education on MCVE through low
income
𝑇𝑇𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎21𝑏𝑏1 + 𝑐𝑐′2
(d) Total effect of medium education on MCVE through
medium income
𝑇𝑇𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎22𝑏𝑏2 + 𝑐𝑐′2
The percent total effects mediated were then estimated as
follows:
(a) Percent total effect of low education mediated through low
income
%𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎11𝑏𝑏1𝑎𝑎11𝑏𝑏1+𝑖𝑖′1
× 100
(b) Percent total effect of low education mediated through
medium income
%𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎12𝑏𝑏2𝑎𝑎12𝑏𝑏2+𝑖𝑖′1
× 100
(c) Percent total effect of medium education mediated through
low income
%𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑎𝑎21𝑏𝑏1𝑎𝑎21𝑏𝑏1+𝑖𝑖′2
× 100
(d) Percent total effect of medium education mediated through
medium income
%𝑏𝑏𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑎𝑎22𝑏𝑏2𝑎𝑎22𝑏𝑏2+𝑖𝑖′2
× 100
The odds ratio (OR) and 95% confidence interval (CI) of
mediation effects177, 178 were then estimated as follows:
(a) Low versus high education and low income
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎11𝑏𝑏1
Win Khaing Methodology / 80
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎11𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1) = 𝑎𝑎112 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏1) + 𝑏𝑏12𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11)
(b) Low versus high education and medium income
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎12𝑏𝑏2
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎12𝑏𝑏2±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12𝑏𝑏2)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12𝑏𝑏2) = 𝑎𝑎122 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏2) + 𝑏𝑏22𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12)
(c) Medium versus high education and low income
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎21𝑏𝑏1
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎21𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21𝑏𝑏1)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21𝑏𝑏1) = 𝑎𝑎212 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏1) + 𝑏𝑏12𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21)
(d) Medium versus high education and medium income
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎22𝑏𝑏2
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎22𝑏𝑏2±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22𝑏𝑏2)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22𝑏𝑏2) = 𝑎𝑎222 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏2) + 𝑏𝑏22𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22)
3.10.2 Income → Education → MCVE
A causal pathway of income education MCVE was constructed (Figure 3.5)
to assess inverse causal effects of income on MCVE through education. Income,
education and MCVE were considered as independent variable, mediator, and outcome
variable, respectively. Mediation and outcome model were constructed as follows:
3.11.2.1 Mediation model
The mediation model was constructed by fitting income on
education using GSEM. A univariate GSEM model with a multinomial logit link
function was used to screen each variable that might associate with education including
age, sex, and marital status. These variables were then considered in the multi-variate
model if their p-value <0.1. Forward selection was applied to select significant variables
being kept in the mediation model with containing income. Finally, four equations from
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 81
causal pathway was constructed, called path a11, a12, a21, a22, see equation 10, 11, 12,
and 13 (Figure 3.6).
For path a11
ln �𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒1|𝑖𝑖𝑖𝑖𝑖𝑖1)𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒3|𝑖𝑖𝑖𝑖𝑖𝑖3)
� = 𝑎𝑎011 + 𝑎𝑎11𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝜀𝜀𝑚𝑚 (10)
For path a12
ln �𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒2|𝑖𝑖𝑖𝑖𝑖𝑖1)𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒3|𝑖𝑖𝑖𝑖𝑖𝑖3)
� = 𝑎𝑎012+ 𝑎𝑎12𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝜀𝜀𝑚𝑚 (11)
For path a21
ln �𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒1|𝑖𝑖𝑖𝑖𝑖𝑖2)𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒3|𝑖𝑖𝑖𝑖𝑖𝑖3)
� = 𝑎𝑎021 + 𝑎𝑎21𝑖𝑖𝑖𝑖𝑐𝑐2 + 𝜀𝜀𝑚𝑚 (12)
For path a22
ln �𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒2|𝑖𝑖𝑖𝑖𝑖𝑖2)𝑃𝑃(𝑒𝑒𝑒𝑒𝑒𝑒3|𝑖𝑖𝑖𝑖𝑖𝑖3)
� = 𝑎𝑎022 + 𝑎𝑎22𝑖𝑖𝑖𝑖𝑐𝑐2 + 𝜀𝜀𝑚𝑚 (13)
where, 𝑖𝑖𝑖𝑖𝑐𝑐1 = Low income
𝑖𝑖𝑖𝑖𝑐𝑐2 = Medium income
𝑖𝑖𝑖𝑖𝑐𝑐3 = High income
𝑒𝑒𝑒𝑒𝑒𝑒1 = Low education
𝑒𝑒𝑒𝑒𝑒𝑒2 = Medium education
𝑒𝑒𝑒𝑒𝑒𝑒3 = High education
3.10.2.2 Outcome model
The outcome pathway was performed using GSEM approach.
MCVE was considered as dichotomous outcome (yes or no), and income was considered
as exposure and education was considered as mediator. In addition, other co-variables
including age, sex, marital status, smoking, alcohol, exercise, BMI, WHR, waist
circumference, family history (hypertension, diabetes, and dyslipidemia), underlying
diseases (hypertension, diabetes, dyslipidemia, and chronic kidney disease) were also
considered. A univariate GSEM model with a family of Bernoulli and a logit link was
used on each of these co-variables that might associate with MCVE. The multivariate
logistic regression model was then fitted by forward including these variables with p-
value < 0.10 in the MCVE model that already contained income and education.
Two equations from causal pathway were constructed including
path b1, b2, 𝑐𝑐′1, 𝑐𝑐′2 , see equation 14 and 15 (Figure 3.6).
Win Khaing Methodology / 82
ln � 𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)1−𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)
� = 𝑏𝑏01 + 𝑏𝑏1𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝑐𝑐′1𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝜀𝜀𝑦𝑦 (14)
ln � 𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)1−𝑃𝑃(𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀+)
� = 𝑏𝑏02 + 𝑏𝑏2𝑒𝑒𝑒𝑒𝑒𝑒2 + 𝑐𝑐′2𝑖𝑖𝑖𝑖𝑐𝑐2 + 𝜀𝜀𝑦𝑦 (15)
3.10.1.3 Estimation of mediation effects
Finally, the two mediation and outcome models were
simultaneously combined using GSEM approach. Estimated coefficients were used to
decompose the total effects of education on MCVE into DE and mediation or indirect175
as follows:
(a) Low versus high income and low education
𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎11𝑏𝑏1
(b) Low versus high income and medium education
𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎12𝑏𝑏2
(c) Medium versus high income and low education
𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎21𝑏𝑏1
(d) Medium versus high income and medium education
𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎22𝑏𝑏2
The total effects (TE) were estimated as follows:
(a) Total effect of low income on MCVE through low education
𝑇𝑇𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎11𝑏𝑏1 + 𝑐𝑐′1
(b) Total effect of low income on MCVE through medium
education
𝑇𝑇𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎12𝑏𝑏2 + 𝑐𝑐′1
(c) Total effect of medium income on MCVE through low
education
𝑇𝑇𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎21𝑏𝑏1 + 𝑐𝑐′2
(d) Total effect of medium income on MCVE through medium
education
𝑇𝑇𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎22𝑏𝑏2 + 𝑐𝑐′2
The percent total effects mediated was then estimated as
follows:
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 83
(a) Percent total effect of low income mediated through low
education
%𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎11𝑏𝑏1𝑎𝑎11𝑏𝑏1+𝑖𝑖′1
× 100
(b) Percent total effect of low income mediated through medium
education
%𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎12𝑏𝑏2𝑎𝑎12𝑏𝑏2+𝑖𝑖′1
× 100
(c) Percent total effect of medium income mediated through low
education
%𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑎𝑎21𝑏𝑏1𝑎𝑎21𝑏𝑏1+𝑖𝑖′2
× 100
(d) Percent total effect of medium income mediated through
medium education
%𝑏𝑏𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑎𝑎22𝑏𝑏2𝑎𝑎22𝑏𝑏2+𝑖𝑖′2
× 100
The odds ratio (OR) and 95% confidence interval (CI) of
mediation effects were then estimated as follows:
(a) Low versus high income and low education
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎11𝑏𝑏1
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎11𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1) = 𝑎𝑎112 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏1) + 𝑏𝑏12𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11)
(b) Low versus high income and medium education
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖1𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎12𝑏𝑏2
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎12𝑏𝑏2±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12𝑏𝑏2)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12𝑏𝑏2) = 𝑎𝑎122 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏2) + 𝑏𝑏22𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎12)
(c) Medium versus high income and low education
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎21𝑏𝑏1
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎21𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21𝑏𝑏1)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21𝑏𝑏1) = 𝑎𝑎212 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏1) + 𝑏𝑏12𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21)
(d) Medium versus high income and medium education
𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒2 = 𝑒𝑒𝑒𝑒𝑒𝑒𝑎𝑎22𝑏𝑏2
Win Khaing Methodology / 84
95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒2𝑖𝑖𝑖𝑖𝑖𝑖2 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎22𝑏𝑏2±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22𝑏𝑏2)�
𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22𝑏𝑏2) = 𝑎𝑎222 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏2) + 𝑏𝑏22𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎22)
3.10.3 Bootstrapping
A bootstrap analysis with 1,000 replications was then applied to all
coefficients, OR of mediated effect, its standard error and its confidence limit without
requiring the assumption of normality. For each bootstrap, the mediated effect was
estimated, averaged across 1000 replications, and its corresponding 95% CI was then
determined using bias-corrected bootstrap technique179, 180.
All analyses were performed using Stata version 14.2 with “mi estimate”
prefix command. Stata codes for mediation analysis and bootstrapping were shown in
Appendix C. P-value of less than 0.05 was considered as a threshold for statistical
significance.
3.11 Ethics considerations This retrospective study was used the demographic, medical, laboratory data
from EGAT project. The permission to access this database was obtained from the
principal investigator (PI) of EGAT project at the Faculty of Medicine, Ramathibodi
Hospital, Mahidol University. They were clearly informed about the objectives, benefits
and methodology of this study before making a decision about the permission.
3.11.1 Respect for human rights and autonomy
EGAT cohort subjects were absolutely voluntary and had already given
written informed consent including blood analysis. All subjects had the right to ask for
further information or withdraw their participation from this study at any time.
This study was used the answered questionnaires, received physical
examinations and blood test data which were performed in 1997. The outcome data was
extracted from the EGAT cohort outcome monitoring. No additional blood tests or
additional investigation was performed for this study apart from validation of unclear
information in the original EGAT protocol.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 85
This study was approved by Institutional Review Board of Ramathibodi’s
Ethical Committee on January 11, 2017 with EC_600219 (Appendix E).
3.11.2 Confidentiality
The information provided by patients was kept confidential. Their personal
information was concealed and only authorized persons were able to see this
information.
3.11.3 Beneficence
The result of this study may not contain benefit directly to individual
subjects. However, subjects who join to this project received health education and health
advice throughout the study period. They were allowed to participate in any health
activities arranged by the EGAT investigators team. The result of this study provided
benefits to their society.
3.11.4 Non-maleficence
This study was an observational study and no additional potential
intervention was provided to subjects. In addition to this, they can withdraw from this
study at any time. Therefore, there was no more than minimal risk.
3.11.5 Justice
Standard hospital operational procedures were provided to all subjects the
same as those that are provided to those who did not participate or patients who
withdrew from this research.
Win Khaing Methodology / 86
Table 3.1 Example of 3 subjects who had inconsistent, missing data and out-of-range
data for sex and height
For Sex
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A Male Male Female Male Male
Participant B Female Female - (missing) Female Loss
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A Male Male Male Male Male
Participant B Female Female Female Female Loss
For Height
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A 160 160 16 (outlier) 160 160
Participant B 145 115 (outlier) 145 Death Death
Participant C 150 150 - (missing) 151 Loss
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A 160 160 160 160 160
Participant B 145 145 145 Death Death
Participant C 150 150 150 151 Loss
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 87
Table 3.2 Example of 3 subjects who had inconsistent data for education, marital status,
smoking status and alcohol consumption
For education
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4
Participant A ≤High school Bachelor Bachelor ≤High school
Participant B Bachelor Vocational/
Diploma
Bachelor - (missing)
Participant C Bachelor ≤High school ≤High school Death
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4
Participant A ≤High school Bachelor Bachelor Bachelor
Participant B Bachelor Bachelor Bachelor - (missing)
Participant C Bachelor Bachelor Bachelor Death
For marital status
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A Single Married Divorce Single Single
Participant B Married Single Married - (missing) Divorce
Participant C Separate Married Single Divorce Single
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A Single Married Divorce Divorce Divorce
Participant B Married Married Married - (missing) Divorce
Participant C Separate Married Married Divorce Divorce
Win Khaing Methodology / 88
Table 3.2 Example of 3 subjects who had inconsistent data for education, marital status,
smoking status and alcohol consumption (Continued)
For smoking status
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A None Current Ex- None None
Participant B Current None Current - (missing) Ex-
Participant C Ex- Current None Ex- None
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A None Current Ex- Ex- Ex-
Participant B Current Current Current - (missing) Ex-
Participant C Ex- Current Current Ex- Ex-
For alcohol consumption
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A None Current Ex- None None
Participant B Current None Current - (missing) Ex-
Participant C Ex- Current None Ex- None
EGAT1/1 EGAT1/2 EGAT1/3 EGAT1/4 EGAT1/5
Participant A None Current Ex- Ex- Ex-
Participant B Current Current Current - (missing) Ex-
Participant C Ex- Current Current Ex- Ex-
Fac. of G
rad. Studies, Mahidol U
niv.
Ph.D.(C
linical Epidemiology) / 89
Table 3.3 Multiple imputation model per variable with their selected variables
Imputed variable
Predictors Outcome
No.
pre
dict
ors
Equa
tion
Sex
Age
Educ
atio
n
Inco
me
Mst
atus
Smok
ing
Alc
ohol
Exer
cise
Wei
ght
Hei
ght
Wai
st
Hip
SBP
NSA
IDs
Lipi
d D
rug
Glu
cose
Cho
lest
erol
HD
L
LDL
Trig
lyce
ride
Uric
Aci
d
Cre
atin
ine
FM-h
t
FM-d
m
FM-li
pid
E_C
ompo
sit
D_C
ompo
sit
Education √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 23 intreg Income √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 ologit Marital status √ √ √ √ √ √ √ √ √ √ √ √ √ √ 14 intreg Smoking √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 16 intreg Alcohol √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 16 intreg Exercise √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 16 mlogit Height √ √ √ √ √ √ √ √ 8 regress Weight √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 19 regress Waist √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 19 regress Hip √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 19 regress SBP √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 regress DBP √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 regress Lipid Drug √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 logit Glucose √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 intreg Cholesterol √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 intreg HDL √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 intreg LDL √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 20 intreg Triglyceride √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 22 intreg Uric Acid √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 20 intreg Creatinine √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ 21 intreg Mstatus, Marital status; FM-ht, Family history of hypertension; FM-dm, Family history of diabetes; FM-lipid, Family history of dyslipidemia; E_composit, composite outcome of MI and Stroke, D_composit, composite outcome of Death due to MI, Stroke and sudden cardiac death
Win Khaing Methodology / 90
Figure 3.1 Time frame of first EGAT cohort and follow-up
EGAT1/1 1985
n = 3499
EGAT1/2 1997
n = 2967
EGAT1/3 2002
n = 2360
EGAT1/4 2007
n = 1958
EGAT1/5 2012
n = 1609
Time-frame of this study
Baseline 5-years 5-years 5-years
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 91
Subjects in EGAT1/2n = 2,960
Baseline data
n = 2,960Lab datan = 2,728
Subjects in EGAT1/3n = 2,360
Baseline data
n = 2,324Lab datan = 2,331
Subjects in EGAT1/4n = 1,958
Baseline data
n = 1,958Lab datan = 1,885
Subjects in EGAT1/5n = 1,609
Baseline data
n = 1,587Lab datan = 1,609
Setup Baseline and Lab datan = 3,025
Merge Data (n = 3,025)• Check & Clean Data• Identify outliers • Recheck with CRF• Cross check within EGAT1
Outcome ascertainment
n = 3,025
Multiple imputation for missing datan = 3,025
Imputed Data (n = 3,025)• Check & Clean Data• Identify outliers• Cross check within EGAT1
Generate composite variables(BMI, WHR, Dyslipidemia,etc)
n = 2,997
Analyze datan = 2,997
Exclude existing MCVE subjects from EGAT1/2
n = 28
Figure 3.2 Data management flow diagram
Win Khaing Methodology / 92
Figure 3.3 A causal effect pathway of education → income → MCVE
Income (M) b
Cardiovascular Events (Y)
Education (X) c
A
B
a
Cardiovascular Events (Y)
e2
e3
Education (X) c'
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 93
Figure 3.4 Generalized structure equation model of causal effect pathway of education
→ income → MCVE
Income1
(Low vs High)
Income2
(Medium vs High)
a11
a21
Education1
(Low vs High)
a12 b1
b2
a22
Education2
(Medium vs High)
c'1 Major Cardiovascular
Events c'2
Win Khaing Methodology / 94
Figure 3.5 A causal effect pathway of income → education → MCVE
Education (M) b
Cardiovascular Events (Y)
Income (X) c
A
B
a
Cardiovascular Events (Y)
e2
e3
Income (X) c'
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 95
Figure 3.6 Generalized structure equation model of causal effect pathway of income
→ education → MCVE
Education1
(Low vs High)
Education2
(Medium vs High)
a11
a21
Income1
(Low vs High)
a12 b1
b2
a22
Income2
(Medium vs High)
c'1 Major Cardiovascular
Events c'2
Win Khaing Results / 96
CHAPTER IV
RESULTS
4.1 Characteristics of subjects A total of 2,967, 2,360, 1,958 and 1,609 subjects enrolled in the EGAT1/2,
1/3, 1/4 and 1/5, respectively, with a total of 3,025 subjects who enrolled in either
EGAT1/2 or 1/3 or 1/4 or 1/5. Among them, 28 subjects had MCVEs before enrolment
of EGAT1/2, thus they were excluded leaving 2,997 subjects include in this study.
Baseline characteristics of subjects in survey were shown in Table 4.1. For baseline
EGAT1/2, mean ages of subjects were 54.3 ± 4.8, majority of subjects were males
(75%). About a half of subjects was smokers and alcohol drinkers, and most subjects
exercised ≥3 times per week in all EGAT1. Prevalence of hypertension and diabetes
were about 50.1% and 10.6%, respectively. More than a half of subjects (76.3%) were
educated at vocational/diploma or higher and had incomes 20,000 Baht/month or higher
(85.1%).
Among 2,997 subjects, there were 238 MCVEs with the incidence rate of
5.3/1000/year. The cumulative incidences of MCVE at 2002, 2007, and 2012 were
estimated to be 2.2% (95%CI: 1.7%, 2.8%), 3.1% (95%CI: 2.5%, 3.8%), and 2.9%
(95%CI: 2.3%, 3.6%), respectively.
Most subjects in EGAT1/2 were educated at vocation/diploma level
(44.6%), and later their education levels were changed to bachelor or higher about
42.8%, 46.2%, and 49.9% in EGAT1/3, EGAT1/4, and EGAT1/5, respectively (see
Table 4.1).
For income, 43% of subjects in EGAT1/2 had medium income (20,000 to
50,000 Baht/month), while about one-third of the subjects’ in EGAT1/3 and EGAT1/4
where their incomes increased to more than 50,000 Baht/month. However, a half of the
subjects in EGAT1/5 had decreased their incomes to less than 20,000 Baht per month
(see Table 4.1).
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 97
4.2 Imputation results Among the interested variables, only three variables, i.e., age, gender and
outcome status data were completed, whereas 20 other variables (i.e., education,
income, marital status, smoking, alcohol, exercise, height, weight, waist, hip, SBP,
DBP, Medication history of lipid lowering, FBS, cholesterol, triglyceride, LDL, HDL,
uric acid, and creatinine level) were missing data, and percentage of missing ranged
from 1.9% to 65.5%, as summarized in Table 4.2. The 5 most frequent missing variables
were uric acid in EGAT1/4 (65.5%), exercise in EGAT1/2 (57%), alcohol in EGAT1/5
(53.4%), income in EGAT1/5 (52.6%), smoking in EGAT1/5 (50%) and exercise in
EGAT 1/4 (50%). Distribution of these missing values was explored and the results
showed arbitrary patterns, thus, MAR was assumed. Multiple chains imputation with 70
imputations were applied to impute those missing data variables using 27 predictors, in
which both completed data and outcome data were included. Summary of imputed data
compared with original data were shown in Table 4.3. Performances of imputations
were assessed by estimated FMI and RVI, see Table 4.2. The average FMI and RVI
value were 0.3317 and 0.30537, with maximum value of 0.691 and 0.6883, respectively.
Therefore 70 imputations might be enough for this dataset. Bias from imputation was
examined using the “midiagplots” command in Stata, by comparing the distributions of
missing to observed values (Appendix D).
4.3 Education → Income → MCVE pathway Mediation analysis was performed by using GSEM by constructing two
models, mediation model and outcome model using data of 2,997 subjects.
Education was considered as exposure, which was categorized in 3 levels
(i.e., low, medium, high). Income was the mediator, which was also categorized as low,
medium, and high. MCVE was considered as dichotomous outcome (yes/no).
4.3.1 Mediation model
Univariate GSEM was performed to assess whether education along with
other three co-variables (i.e., age, sex, and marital status) were significantly associated
Win Khaing Results / 98
with income, see Table 4.4. Results indicated that education (F = 96.08, p<0.0001), age
(F = 448.22, p<0.0001), and marital status (F = 8.85, p<0.0001) were significantly
associated with income, whereas sex was not (F = 1.72, p = 0.1788). Therefore, age and
marital status were simultaneously considered in multivariate GSEM that already
contained education. Effects of education on income was estimated (i.e., a11, a12, a21,
a22), see Table 4.5. For low versus high education, the effects on low and medium
incomes were 1.99 (95% CI: 1.79, 2.19) and 1.00 (95% CI: 0.85, 1.16), respectively.
For medium versus high education, the effects on low and medium incomes were 1.13
(95% CI: 0.96, 1.30) and 0.49 (95% CI: 0.37, 0.62), respectively.
4.3.2 Outcome model
A univariate GSEM with logit link was used to assess whether income and
education along with other co-variables (i.e., age, sex, marital status, smoking, alcohol,
exercise, BMI, WHR, having underlying diseases of hypertension, diabetes,
dyslipidemia, and chronic kidney disease) were associated with MCVE, see Table 4.6.
For our studied factors, only income (F = 12.79, p<0.0001) was significantly associated
with MCVE whereas education was not (F = 2.26, p = 0.1042). A total of 7 out of 12
co-variables were also significantly associated with MCVE including age (F = 126.81,
p< 0.0001), WHR (F = 3.26, p = 0.0712), exercise (F = 3.51, p = 0.0304), having
underlying diseases of hypertension (F = 18.61, p < 0.0001), diabetes (F = 11.69, p =
0.0006), dyslipidemia (F = 11.19, p = 0.0009), and CKD stages (F = 3.44, p = 0.0638).
Therefore, these co-variables were simultaneously considered in the outcome model
which already contained education and income. A forward selection was applied and
indicated that only hypertension, diabetes, and dyslipidemia should be kept in the final
outcome model, see Table 4.7.
Effects of low and medium income on MCVE controlling for education
(paths b1, b2) and effects of low and medium educations on MCVE controlling for
income (𝑐𝑐′1, 𝑐𝑐′2), were assessed, see Table 4.7. Effects of low and medium incomes on
MCVE after controlling for education and other co-variables were 1.10 (95% CI: 0.61,
1.59) and 0.77 (95% CI: 0.29, 1.25), respectively. Effects of education on MCVE were
not significant with the coefficients of 0.06 (95% CI: -0.44, 0.56) and 0.25 (95% CI: -
0.20, 0.69) for low and medium versus high educations, respectively.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 99
4.3.3 Estimations of DE, ME, and TE
Potential causal relationship between education and MCVE mediating
through income was assessed by mediation analysis with GSM, Figure 4.1. A bootstrap
with 1000 replications was performed for both final models (see Table 4.8), which
yielded the effects of low and medium versus high education on MCVE through income.
Estimated model coefficients were used to decompose the total effects of education on
MCVE into mediation (indirect) and direct effects. As a result, potential causal effects
of low education on MCVE mediated through low and medium incomes were 2.19 (95%
CI: 1.43, 2.97) and 0.77 (95% CI: 0.40, 1.16), respectively (see Tables 4.9 and 4.10).
However, the direct effect of low education on MCVE was not significant (coefficient
= 0.06, 95% CI: -0.42, 0.59). The percentages of low education effects contributed
through low and medium incomes (a11b1 and a12b2 path) were 97.35% and 92.83%,
respectively.
For medium education, the coefficient of medium education on MCVE
through low and medium incomes were 1.24 (95% CI: 0.81, 1.72) and 0.38 (95% CI:
0.20, 0.59), respectively (see Tables 4.11 and 4.12). However, the direct effect of this
was not significant (coefficient = 0.25, 95% CI: -0.19, 0.69). The percentages of medium
education effect contributed through low and medium income mediators (a21b1 and a22b2
path) were 83.47% and 60.64%, respectively.
4.3.4 Estimation of OR of ME
The OR and their 95% CI of ME were then estimated and summarized in
Figure 4.2. Subjects with low education would have low income and thus increased odds
of having MCVE about 8.95 (95% CI: 4.19, 19.56) times higher than subjects with high
education and high income. The odds of MCVE was decreased if low educated subjects,
had medium income with the OR of 2.17 (95%CI: 1.50, 3.20). For medium education,
the odds of developing MCVE was 3.47 (95%CI: 2.24, 5.61) and 1.46 (95%CI: 1.22,
1.80) times higher in subjects with low and medium incomes than those with high
income, respectively.
Win Khaing Results / 100
4.4 Income → Education → MCVE pathway The effect of income on MCVE through education was also considered by
setting income as exposure, MCVE as outcome, and education as mediator. Income and
education were categorized in 3 levels (i.e., low, medium, high) and MCVE was
considered as dichotomous outcome (yes/no).
4.4.1 Mediation model
Univariate GSEM was performed to screen whether income and other three
co-variables (i.e., age, sex, and marital status) were significantly associated with
education. Results showed that income (F = 96.08, p<0.0001), sex (F = 12.51,
p<0.0001), and marital status (F = 4.69, p=0.0009) were significantly associated with
education, whereas age (F = 0.13, p = 0.8787) was not. Therefore, sex and marital status
were simultaneously considered in multivariate GSEM which already contained
income. Effects of income on education were estimated (i.e., a11, a12, a21, a22). For low
versus high income, effects on low and medium educations were 1.60 (95% CI: 1.43,
1.78) and 0.92 (95% CI: 0.77, 1.07), respectively. For medium versus high income,
effects on low and medium educations were 0.86 (95% CI: 0.72, 1.01) and 0.42 (95%
CI: 0.30, 0.54), respectively.
4.4.2 Outcome model
A univariate GSEM with logit link was used to assess whether education
and income along with other co-variables (i.e., age, sex, marital status, smoking,
alcohol, exercise, BMI, WHR, waist circumference, family history (hypertension,
diabetes, and dyslipidemia), having underlying diseases of hypertension, diabetes,
dyslipidemia, and chronic kidney disease) were associated with MCVE.
Since, effect of low and medium education on MCVE controlling for
income (paths b1, b2) was not significant (F = 0.89, p = 0.4093), the causal effect of
income on MCVE through education could not be determined because one of the
assumptions of mediation analysis (i.e., mediator must be associated with outcome) was
violated.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 101
Table 4.1 Baseline characteristics of the studied subjects by EGAT periods
Characteristics EGAT1/2
n = 2,939
EGAT1/3
n = 2,333
EGAT1/4
n = 1,935
EGAT1/5
n = 1,593
Age (years), (mean ± SD) 54.3 ± 4.8 59.1 ± 4.9 64.1 ± 4.7 68.8 ± 4.6
Sex
Male 2252 (75.9) 1773 (75.1) 1449 (74.0) 1173 (72.9)
Female 715 (24.1) 587 (24.9) 509 (26.0) 436 (27.1)
Education
≤ High School 703 (23.7) 613 (26.0) 447 (22.8) 300 (18.6)
Vocational/Diploma 1324 (44.6) 737 (31.2) 606 (31.0) 506 (31.5)
≥Bachelor 940 (31.7) 1010 (42.8) 905 (46.2) 803 (49.9)
Income (Baht)
<20,000 441 (14.9) 488 (20.7) 658 (33.6) 835 (51.9)
20,000 – 49,999 1275 (43.0) 792 (33.5) 579 (29.6) 589 (36.6)
≥50,000 1251 (42.1) 1080 (35.8) 721 (36.8) 185 (11.5)
Marital status
Single 146 (4.9) 127 (5.38) 116 (5.9) 89 (5.5)
Married 2517 (84.8) 2011 (85.2) 1608 (82.1) 1295 (80.5)
Widowed/
separate/divorce 304 (10.3) 222 (9.4) 234 (12.0) 225 (14.0)
Height (cm), (mean ± SD) 163.1 ± 7.2 163.1 ± 7.2 163.1 ± 7.2 163.1 ± 7.2
Body weight (kg),
(mean ± SD) 64.9 ± 9.0 65.1 ± 10.4 65.3 ± 10.8 64.0 ± 10.8
BMI (kg/m2), (mean ± SD) 24.5 ± 3.6 24.4 ± 3.4 24.5 ± 3.5 24.0 ± 3.5
Waist circumference (cm),
(mean ± SD) 90.8 ± 8.1 89.3 ± 8.7 89.7 ± 9.4 89.2 ± 9.4
Hip circumference (cm),
(mean ± SD) 97.3 ± 6.1 96.4 ± 6.3 95.8 ± 6.7 95.7 ± 6.8
SBP (mmHg), (mean ± SD) 135.9 ± 19.6 129.0 ± 18.7 134.7 ± 19.0 133.3 ± 18.1
DBP (mmHg), (mean ± SD) 81.9 ± 11.7 83.0 ± 11.2 80.8 ± 10.4 76.7. ± 10.2
Win Khaing Results / 102
Table 4.1 Baseline characteristics of the studied subjects by EGAT periods (continued)
Characteristics EGAT1/2
n = 2,939
EGAT1/3
n = 2,333
EGAT1/4
n = 1,935
EGAT1/5
n = 1,593
Smoking
Non-smoker 1341 (45.2) 1089 (46.2) 945 (48.3) 769 (47.8)
Ex-smoker 641 (21.6) 935 (39.6) 807 (41.2) 713 (44.3)
Current smoker 985 (33.2) 336 (14.2) 206 (10.5) 127 (7.9)
Alcohol
Non-drinker 1500 (50.5) 1207 (51.1) 992 (50.7) 718 (44.6)
Ex-drinker 370 (12.5) 525 (22.3) 508 (25.9) 731 (45.4)
Current drinker 1097 (37.0) 628 (26.6) 458 (23.4) 160 (10.0)
Exercise
None 159 (5.4) 462 (19.6) 161 (8.2) 464 (28.8)
1 – 2 times/week 161 (5.4) 425 (18.0) 300 (15.3) 191 (11.9)
≥3 times/week 2647 (89.2) 1473 (62.4) 1497 (76.5) 954 (59.3)
Hypertension
Yes 1487 (50.1) 1170 (49.6) 1180 (60.3) 1082 (67.3)
No 1480 (49.9) 1190 (50.4) 778 (39.7) 527 (32.7)
Diabetes Mellitus
Yes 314 (10.6) 559 (23.7) 465 (23.8) 276 (17.2)
No 2653 (89.4) 1801 (76.3) 1493 (76.2) 1333 (82.8)
FPG (mg/dL), (mean ± SD) 96.3 ± 28.6 109.2 ± 34.9 102.5 ± 27.2 99.5 ± 22.9
TC (mg/dL), (mean ± SD) 239.0 ± 40.0 240.1 ± 43.4 210.8 ± 41.0 203.4 ± 43.1
Triglyceride (mg/dL),
(mean ± SD)
164.1±104.7 151.5±106.0 135.1±70.6 120.9±60.9
LDL (mg/dL), (mean ± SD) 155.1 ± 38.6 156.7 ± 39.9 138.1 ± 37.9 131.3 ± 39.4
HDL (mg/dL), (mean ± SD) 52.7 ± 10.7 54.2 ± 14.7 57.7 ± 15.7 59.6 ± 16.5
Creatinine (mg/dL),
(mean ± SD)
1.2 ± 0.5 1.0 ± 0.4 0.9 ± 0.4 1.0 ± 0.4
Uric acid (mg/dL),(mean±SD) 6.3 ±1.3 6.0 ± 1.3 6.2 ± 1.3 6.1 ± 1.5
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 103
Table 4.2 Report of number of missing data
Variables EGAT
period Observed
%
missing Imputed FMI RVI
Education EGAT1/2 2196 27.4 829 0.2494 0.1404
EGAT1/3 2501 17.3 524 0.0828 0.0811
EGAT1/4 1998 34.0 1027 0.0794 0.0868
Income EGAT1/2 2758 8.8 267 0.0999 0.0615
EGAT1/3 2109 30.3 916 0.3444 0.2759
EGAT1/4 1689 44.2 1336 0.5788 0.6792
EGAT1/5 1434 52.6 1591 0.6755 0.6062
Marital EGAT1/2 1938 35.9 1087 0.3587 0.2766
Status EGAT1/3 2309 23.7 716 0.3178 0.2397
EGAT1/4 1941 35.8 1084 0.3035 0.2369
EGAT1/5 1585 47.6 1440 0.2849 0.3690
Smoking EGAT1/2 2660 12.1 365 0.0846 0.0526
EGAT1/3 2308 23.7 717 0.1429 0.1372
EGAT1/4 1943 35.8 1082 0.1967 0.1669
EGAT1/5 1512 50.0 1513 0.3490 0.3550
Alcohol EGAT1/2 2482 18.0 543 0.2008 0.1034
EGAT1/3 2267 25.1 758 0.1789 0.1535
EGAT1/4 1818 39.9 1207 0.1982 0.1724
EGAT1/5 1410 53.4 1615 0.5020 0.4934
Exercise EGAT1/2 1300 57.0 1725 0.6183 0.6813
EGAT1/3 2193 27.5 832 0.3403 0.2921
EGAT1/4 1514 50.0 1511 0.6386 0.5653
EGAT1/5 1581 47.7 1444 0.6462 0.6507
Height 2547 15.8 478 0.0993 0.0550
Weight EGAT1/2 2901 4.1 124 0.0285 0.0147
EGAT1/3 2302 23.9 723 0.0912 0.0501
EGAT1/4 1866 38.3 1159 0.1274 0.0729
EGAT1/5 1568 48.2 1457 0.4183 0.3558
Win Khaing Results / 104
Table 4.2 Report of number of missing data (continued)
Variables EGAT
period Observed
%
missing Imputed FMI RVI
Waist EGAT1/2 2316 23.4 709 0.1539 0.0910
EGAT1/3 2339 22.7 686 0.1296 0.0743
EGAT1/4 1906 37.0 1119 0.2420 0.1603
EGAT1/5 1560 48.4 1465 0.6238 0.6192
Hip EGAT1/2 2316 23.4 709 0.1628 0.0971
EGAT1/3 2763 8.7 262 0.0857 0.0467
EGAT1/4 2468 18.4 557 0.0891 0.0488
EGAT1/5 1560 48.4 1465 0.5662 0.6444
SBP EGAT1/2 2429 19.7 596 0.2051 0.1283
EGAT1/3 2307 23.7 718 0.2677 0.1826
EGAT1/4 1859 38.5 1166 0.5276 0.5539
EGAT1/5 1583 47.7 1442 0.6027 0.6037
DBP EGAT1/2 2427 19.8 598 0.1838 0.1120
EGAT1/3 2307 23.7 718 0.2907 0.2038
EGAT1/4 1858 38.6 1167 0.5541 0.6129
EGAT1/5 1577 47.9 1448 0.6227 0.6883
Medication EGAT1/2 2967 1.9 58 0.0017 0.0009
(Lipid EGAT1/3 2361 22.0 664 0.2087 0.1316
lowering) EGAT1/4 1958 35.3 1067 0.5201 0.5401
EGAT1/5 1839 39.2 1186 0.6030 0.6070
FBS EGAT1/2 2762 8.7 263 0.0042 0.0023
EGAT1/3 2351 22.3 674 0.1617 0.0960
EGAT1/4 1922 36.5 1103 0.3776 0.3025
EGAT1/5 1606 46.9 1419 0.6598 0.6557
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 105
Table 4.2 Report of number of missing data (continued)
Variables EGAT
period Observed
%
missing Imputed FMI RVI
Cholesterol EGAT1/2 2769 8.5 256 0.0071 0.0036
EGAT1/3 2351 22.3 674 0.2678 0.1816
EGAT1/4 1922 36.5 1103 0.4790 0.4560
EGAT1/5 1606 46.9 1419 0.6288 0.6819
Triglyceride EGAT1/2 2769 8.5 256 0.0063 0.0032
EGAT1/3 2351 22.3 674 0.1968 0.1219
EGAT1/4 1922 36.5 1103 0.5075 0.5095
EGAT1/5 1606 46.9 1419 0.6910 0.6028
LDL EGAT1/2 2695 10.9 330 0.0428 0.023
EGAT1/3 2332 22.9 693 0.2235 0.1431
EGAT1/4 1922 36.5 1103 0.5283 0.5545
EGAT1/5 1606 46.9 1419 0.6701 0.6484
HDL EGAT1/2 2769 8.5 256 0.0095 0.0048
EGAT1/3 2351 22.3 674 0.2521 0.1675
EGAT1/4 1922 36.5 1103 0.5450 0.5920
EGAT1/5 1606 46.9 1419 0.6478 0.6697
Uric acid EGAT1/2 1884 37.7 1141 0.5048 0.5060
EGAT1/3 1567 48.2 1458 0.4479 0.4012
EGAT1/4 1045 65.5 1980 0.6071 0.6613
EGAT1/5 1584 47.6 1441 0.6688 0.6366
Creatinine EGAT1/2 2233 26.2 792 0.1021 0.0567
EGAT1/3 2351 22.3 674 0.2122 0.1339
EGAT1/4 1922 36.5 1103 0.2374 0.1549
EGAT1/5 1606 46.9 1419 0.6447 0.6603
Win Khaing Results / 106
Table 4.3 Comparison of characteristics of subjects between original dataset and
imputed dataset
Characteristics Original Imputed
percentage percentage
Education
≤ High School 24.5 26.0
Vocational/Diploma 30.3 29.8
≥Bachelor 45.2 44.2
Income (Baht)
<20,000 26.0 28.0
20,000 – 49,999 35.9 34.5
≥50,000 38.1 37.5
Marital status
Single 5.8 5.8
Married 82.6 82.1
Widowed/ separate/
divorce
11.6 12.1
Smoking
Non-smoker 48.0 47.7
Ex-smoker 33.4 33.4
Current smoker 18.6 18.9
Alcohol
Non-drinker 51.1 50.5
Ex-drinker 22.2 22.8
Current drinker 26.8 26.7
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 107
Table 4.3 Comparison of characteristics of subjects between original dataset and
imputed dataset (continued)
Characteristics Original Imputed
percentage percentage
Exercise
None 18.3 19.4
1 – 2 times/week 15.9 15.8
≥3 times/week 66.1 64.8
Medication
(Lipid lowering)
Yes 48.2 47.3
No 51.8 52.7
Characteristics Original Imputed
mean SD mean SD
Height (cm) 163.0 7.2 163.0 7.2
Body weight (kg) 64.9 10.2 64.9 10.4
Waist circumference (cm) 89.8 9.0 90.0 9.2
Hip circumference (cm) 96.4 6.5 96.3 6.7
SBP (mmHg) 133.1 19.8 133.9 20.0
DBP (mmHg) 81.0 11.6 80.6 11.7
FPG (mg/dL) 101.7 30.0 101.8 30.7
TC (mg/dL) 226.4 45.1 226.0 45.2
Triglyceride (mg/dL) 145.6 93.6 146.0 95.4
LDL (mg/dL) 147.3 41.0 146.8 41.1
HDL (mg/dL) 55.5 14.5 55.7 14.6
Creatinine (mg/dL) 6.1 1.5 6.1 1.5
Uric acid(mg/dL) 1.0 0.4 1.0 0.5
Win Khaing Results / 108
Table 4.4 Mediation analysis of education and income (Univariate Analysis) E
quat
ions
Factors b SE t P 95%CI
Low
Inco
me
Education
Low 1.62 0.09 18.75 <0.0001 1.45, 1.79
Medium 0.93 0.08 12.01 <0.0001 0.78, 1.08
Age 0.16 0.01 30.86 <0.0001 0.15, 0.17
Sex (Male) 0.12 0.08 1.46 0.145 -0.04, 0.28
Marital status
Married -0.03 0.15 -0.20 0.844 -0.32, 0.26
Widowed/
Separate/
Divorce
0.53 0.18 3.04 0.002 0.19, 0.88
Med
ium
Inco
me
Education
Low 0.87 0.07 11.81 <0.0001 0.72, 1.01
Medium 0.42 0.06 6.91 <0.0001 0.30, 0.54
Age 0.06 0.00 13.87 <0.0001 0.05, 0.07
Sex (Male) -0.02 0.07 -0.26 0.793 -0.15, 0.11
Marital status
Married 0.03 0.12 0.20 0.839 -0.22, 0.27
Widowed/
Separate/
Divorce
0.38 0.15 2.56 0.011 0.09, 0.67
b, coefficient; SE, standard error; t, t-test; P, p-value; CI, confidence interval;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 109
Table 4.5 Mediation analysis of education and income (Multivariate Analysis) E
quat
ions
Factors b SE t P 95%CI
Low
Inco
me
Education
Low 1.99 0.10 19.80 <0.0001 1.79, 2.19
Medium 1.13 0.09 13.19 <0.0001 0.96, 1.30
Age 0.17 0.01 31.56 <0.0001 0.16, 0.18
Marital status
Married -0.47 0.17 -2.79 0.005 -0.80, -0.14
Widowed/
Separate/
Divorce
-0.15 0.20 -0.73 0.463 -0.54, 0.25
Med
ium
Inco
me
Education
Low 1.00 0.08 13.03 <0.0001 0.85, 1.16
Medium 0.49 0.06 7.84 <0.0001 0.37, 0.62
Age 0.06 0.00 14.96 <0.0001 0.06, 0.07
Marital status
Married -0.16 0.13 -1.24 0.215 -0.40, 0.09
Widowed/
Separate/
Divorce
0.12 0.15 0.77 0.440 -0.18, 0.41
b, coefficient; SE, standard error; t, t-test; P, p-value; CI, confidence interval;
Win Khaing Results / 110
Table 4.6 Mediation analysis of MCVE (Univariate Analysis)
Factors b SE t P 95%CI
Education
Low 0.40 0.25 1.59 0.112 -0.09, 0.88
Medium 0.45 0.22 2.03 0.042 0.02, 0.89
Income
<20,000 1.20 0.24 5.01 <0.0001 0.73, 1.67
20,000 – 49,999 0.81 0.24 3.38 0.001 0.34, 1.28
Age 0.11 0.01 11.26 <0.0001 0.09, 0.13
Sex (Male) 0.33 0.25 1.33 0.185 -0.16, 0.81
Marital status
Married 0.34 0.56 0.60 0.546 -0.76, 1.44
Widowed/
Separate/ Divorce
0.59 0.60 0.97 0.330 -0.59, 1.77
BMI (kg/m2)
<18.5 -1.63 0.98 -1.66 0.097 -3.55, 0.29
≥23.0 0.18 0.20 0.91 0.362 -0.21, 0.57
WHR (High) 0.42 0.23 1.81 0.071 -0.04, 0.88
Smoking (Smoker) 0.24 0.20 1.18 0.237 -0.16, 0.64
Alcohol (Drinker) 0.18 0.19 0.95 0.344 -0.20, 0.56
Exercise
1-2 times/week -0.79 0.32 -2.48 0.014 -1.42, -0.16
≥3 times/week -0.34 0.21 -1.61 0.108 -0.74, 0.07
Dyslipidemia (Yes) 0.98 0.29 3.35 0.001 0.40, 1.55
Hypertension (Yes) 0.79 0.18 4.31 <0.0001 0.43, 1.15
Diabetes (Yes) 0.66 0.19 3.42 0.001 0.28, 1.04
CKD Stages (≥3) 0.33 0.18 1.86 0.064 -0.02, 0.68 b, coefficient; SE, standard error; t, t-test; P, p-value; CI, confidence interval;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 111
Table 4.7 Mediation analysis of MCVE (Multivariate Analysis)
Factors b SE t P 95%CI
Education
Low 0.06 0.26 0.23 0.817 -0.44, 0.56
Medium 0.25 0.23 1.09 0.276 -0.20, 0.69
Income
<20,000 1.10 0.25 4.45 <0.0001 0.61, 1.59
20,000 – 49,999 0.77 0.24 3.16 0.002 0.29, 1.25
Dyslipidemia (Yes) 0.88 0.30 2.96 0.003 0.30, 1.46
Hypertension (Yes) 0.63 0.19 3.33 0.001 0.26, 0.99
Diabetes (Yes) 0.52 0.20 2.60 0.009 0.13, 0.90 b, coefficient; SE, standard error; t, t-test; P, p-value; CI, confidence interval;
Win Khaing Results / 112
Table 4.8 Mediation analysis of Education on MCVE that was mediated by income
(bias-corrected bootstrapped) E
quat
ions
Factors b SE z P 95%CI
MC
VE
Education
Low 0.06 0.26 0.23 0.821 -0.42, 0.59
Medium 0.25 0.23 1.09 0.275 -0.19, 0.69
Income
<20,000 1.10 0.19 5.78 <0.0001 0.73, 1.48
20,000 – 49,999 0.77 0.19 4.08 <0.0001 0.39, 1.11
Dyslipidemia (Yes) 0.88 0.22 4.05 <0.0001 0.45, 1.32
Hypertension (Yes) 0.63 0.14 4.54 <0.0001 0.37, 0.91
Diabetes (Yes) 0.52 0.16 3.24 0.001 0.20, 0.83
Low
Inco
me
Education
Low 1.99 0.08 26.00 <0.0001 1.86, 2.16
Medium 1.13 0.07 17.11 <0.0001 1.01, 1.26
Age 0.17 0.00 40.90 <0.0001 0.16, 0.18
Marital status
Married -0.47 0.14 -3.45 0.001 -0.74, -0.22
Widowed/
Separate/ Divorce
-0.15 0.15 -0.97 0.332 -0.45, 0.14
Med
ium
Inco
me
Education
Low 1.00 0.06 17.06 <0.0001 0.89, 1.12
Medium 0.49 0.047 10.50 <0.0001 0.40, 0.59
Age 0.06 0.00 19.88 <0.0001 0.06, 0.07
Marital status
Married -0.16 0.09 -1.69 0.091 -0.33, 0.03
Widowed/
Separate/ Divorce
0.12 0.11 1.08 0.280 -0.08, 0.34
b, coefficient; SE, standard error; z, z-test; P, p-value; CI, confidence interval;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 113
Table 4.9 Causal effects of low education on MCVE through low income (bias-
corrected bootstrapped)
Effects Pathway b SE Bias 95%CI
Indirect Edu𝐿𝐿→ Inc𝐿𝐿 → MCVE 2.19 0.40 0.04 1.43, 2.97
Direct Edu𝐿𝐿 → MCVE 0.06 0.26 -0.00 -0.42, 0.59
Percent of total effects mediated 97.35
Percent of direct effect 2.65 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduL, Low education; IncL, Low
income;
Win Khaing Results / 114
Table 4.10 Causal effects of low education on MCVE through medium income (bias-
corrected bootstrapped)
Effects Pathway b SE Bias 95%CI
Indirect Edu𝐿𝐿→ Inc𝑀𝑀 → MCVE 0.77 0.20 0.01 0.40, 1.16
Direct Edu𝐿𝐿 → MCVE 0.06 0.26 -0.00 -0.42, 0.59
Percent of total effects mediated 92.83
Percent of direct effect 7.17 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduL, Low education; IncM, Medium
income;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 115
Table 4.11 Causal effects of medium education on MCVE through low income (bias-
corrected bootstrapped)
Effects Pathway b SE Bias 95%CI
Indirect Edu𝑀𝑀→ Inc𝐿𝐿 → MCVE 1.24 0.23 0.02 0.81, 1.72
Direct Edu𝑀𝑀 → MCVE 0.25 0.23 0.00 -0.19, 0.69
Percent of total effects mediated 83.47
Percent of direct effect 16.53 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduM, Medium education; IncL,
Low income;
Win Khaing Results / 116
Table 4.12 Causal effects of medium education on MCVE through medium income
(bias-corrected bootstrapped)
Effects Pathway b SE Bias 95%CI
Indirect Edu𝑀𝑀→ Inc𝑀𝑀 → MCVE 0.38 0.10 0.01 0.20, 0.59
Direct Edu𝑀𝑀 → MCVE 0.25 0.23 0.00 -0.19, 0.69
Percent of total effects mediated 60.64
Percent of direct effect 39.36 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduM, Medium education; IncM,
Medium income;
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 117
Figu
re 4
.1 C
ausa
l med
iatio
n pa
thw
ay d
iagr
am o
f the
rela
tions
hip
amon
g ed
ucat
ion,
inco
me
and
MC
VE
usin
g ge
nera
lized
stru
ctur
al e
quat
ion
mod
ellin
g *
indi
cate
s sta
tistic
al si
gnifi
cant
; Val
ues s
how
n re
pres
ent u
nsta
ndar
dize
d co
effic
ient
s afte
r bia
s-co
rrec
ted
boot
stra
p te
chni
que;
outc
omec
ve, M
CV
E ou
tcom
e; 1
.iinc
, low
inc
ome;
2.ii
nc, m
ediu
m i
ncom
e; 1
.iedu
gr3,
low
edu
catio
n; 2
.iedu
gr3,
med
ium
edu
catio
n; 2
.imst
atus
,
mar
ried;
3.im
stat
us, w
idow
ed/s
epar
ate/
divo
rce;
dm
, dia
bete
s; h
t, hy
perte
nsio
n; d
yslip
id1,
dys
lipid
emia
;
Win Khaing Results / 118
Figure 4.2 Direct effects of education on MCVE and mediated effects of and percentage
of mediation from effect of education on MCVE through low income. Direct and
mediate effects were on odds ratio scale. Confidence interval that include 1 indicate no
statistical significance. OR, odds ratio; CI, confidence interval; MCVE, major
cardiovascular events.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 119
CHAPTER V
DISCUSSION
5.1 Main findings The causal relationship between education and MCVE through income was
assessed using data from EGAT 1997 to 2012 cohort. The GSEM was applied to assess
causal pathway of education → income → MCVE. Our results indicated that education
did not directly affect MCVE, but mainly affected MCVE through income. Low
educated subjects with low income were approximately 9 times higher odds of having
MCVE than high educated subjects. The odds of MCVE occurrence was decreased
approximately a half if education of these subjects increased to medium level, although
income was still in low level. However, the odds of MCVE was not much different
between low and medium education given the medium income, i.e., approximately 2
times higher odds than high education and high income level. .
5.2 Income measurement In the EGAT1 survey, income was measured by self-reported individual’s
income, which was classified as low, medium, and high income. Three common
methods of income measurement were commonly used, i.e., individual,
household/family and area-based/neighborhood levels. Income provided individual and
families necessary material resources, their purchasing power and contribute to
resources needed in maintaining good health. Several studies compared these income
measurements yielding conflicting results. Many studies181-183 reported that household-
level and area-based level data had poor approximations of individual-level income
data, with weak correlations especially in rural regions, while other studies184-186
suggested both measures are comparable in terms of ability to identify variations in
outcomes. Our study used individual-level income by individual self-reported income
Win Khaing Discussion / 120
as categorical data. The actual income can be measured as continuous data, but it is quite
sensitive data in which subjects may not want to tell the truth about their incomes. Thus
validity of continuous income data is less than categorical income data.
5.3 Education measurement Educational level of one person is relatively easier to obtain than income
and it seems to have less recall bias as people tends to remember their education easily.
However, different societies have different complex educational systems which also
changed over time. Complete survey on education attainment should contain (a) the
years of schooling, (b) their school grading, (c) how much education they have
completed, (d) the title or nature of their qualifications and (e) the type of institution that
they attended post-school. The most common measure of education is a number of years
of schooling, which could be considered as continuous data or categorize it as
categorical data. Connelly et al187 reported that year of schooling as continuous data was
particularly useful in statistical modelling, but it was less useful as a proxy of
educational attainment because qualifications with very different levels often require a
similar amount of time in education due to the structure and organization of the
educational year in some countries. They suggested that qualification-based
categorization provide more detailed information on educational attainment. In the
EGAT1 survey, education was measured as self-reported individual’s education
attainment by category, which was an appropriate measure in this study.
5.4 Causal relationship pathway between education and MCVE
through income The relationship between education and health has been well-established in
recent decades. Many studies24-28, 37 and our meta-analysis42 found an inverse
relationship between education and cardiovascular outcome, i.e., low education was
more likely to have metabolic diseases and MCVE; or in other words, high education
was less likely to develop these disease conditions. These might be explained as
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 121
follows188: First, low educated subjects were more likely to be unemployed thus less
income, whereas well-educated subjects were more likely to be employed and thus good
income. Second, regarding social-psychological resources, the better-educated subjects
have a greater sense of control over their lives and their health, and they have higher
levels of social support. Finally, higher educated subjects were less likely to have risk
behaviors (smoking, alcohol drinking), more likely to have better life-style (more leisure
time for exercise, eat healthier foods), and utilize better healthcare resources with
regular medical check-ups.
Our results showed no direct association between education and MCVE.
Although only about an quarter of our EGAT subjects were educated at high school or
lower level, the EGAT enterprise has regularly provided the excellent health-care
scheme (e.g., regular annual medical check-up, health education) for all staff. This
regular screening procedure might impact on subjects’ health condition and might mask
the effect of education on MCVE outcome.
Another explanation could be related with patient’s health literacy, which is
a set of abilities to locate, understand and use health-related information and services
needed to make appropriate health decisions189. It reflects more ‘updated’ practical skills
and capabilities which are indispensable for better health outcomes. One’s health needs
and necessary health information are not static over the life course and life-long or life-
wide learning is usually ignored after the achievement of formal education190. Bennett
et al.191 reported that health literacy mediated significantly between educational
attainment and health outcomes in senior adults of America. Schillinger et al.,192 also
found that health literacy was a significant mediator between education and glycemic
control among low-income diabetes patients. Friis et al.,193 also confirmed that
educational attainment was mediated by health literacy in the relationship with physical
activity, poor diet and obesity. As a result, most researchers accepted that health literacy
may be a more current indicator of education status and it is sensible to be take into
account in education-health related outcome research to fill the time gap in between
formal education and health outcomes especially for middle-age to older cohorts194.
Unfortunately, we did not collect health literacy in EGAT1 survey, its causal pathway
between education and MCVE could not be explored.
Win Khaing Discussion / 122
Variation of income might also play a role on CVD. In the study by Kubota
et al195, income change appeared to speed up the lifetime CVD cumulative risk. They
found that in the joint association of educational attainment and income change, there
was a steep rise in lifetime CVD risk in those low educated with high income change.
Subramanian and Kawachi196 also pointed out that early childhood education policy
could mediate the relationship between income inequality and health outcomes, and
income-based inequality and its effects on health outcomes would continue to be
relevant for social epidemiology.
5.5 Education, income and sustainable development goals Education and income are essential issues to achieve sustainable
development goals (SDG), which should be able to reduce poverty and hunger, improve
health, economic growth, and etcetera. To fulfill SDG goal 1 of no poverty, education
is the one of the most effective ways to reduce poverty. United Nations Educational,
Scientific and Culture Organization197 reported that if all students in low-income
countries left schools with basic reading skills, 171 million people could be lifted out of
poverty. They also found that individual’s earning was increased up to 10% when
schooling was increased by one extra year. The SDG goal 4 is about quality education;
investing and strengthening in education can provide greater quality education to all
others with the skills needed to be responsible global citizens and nurture a culture that
values learning. In addition, education is also one of the strongest drivers of SDG goal
8, which is about good jobs and economic growth. Each additional year of schooling
raised annual gross domestic product growth by 0.37%197. Finally, concerning SDG goal
10 for reducing inequality within and among countries, evidence showed that improving
a country’s education equality about 0.1% can raise its per capita income by 23% higher
after forty years198.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 123
5.6 Multiple imputation Missing data could not be avoided in observational studies particularly for
a cohort like our study. Nowadays, MICE is increasingly recommended in
epidemiological studies because of its advantage over other techniques in terms of its
flexibility and validity163, 164. Previously, imputation was recommended only in
longitudinal studies, but a recent review by Enders199 suggested that imputation should
be considered in all missing data by using MI method instead of complete case analysis,
and the choice of MI method should be based on context and assumption behind of the
missing data. Furthermore, it was confirmed that MI methods have more valid
variability than longitudinal single imputation200, which is more superior than single
imputation methods201.
Percentages of missing data for our studied factors (i.e., education and
income) ranged from about 27% to 52%, and about 1.9% to 65.5% for other co-
variables. Multiple chain imputation with 70 replications was applied to fill in these
missing data. Performance of imputation should be determined and assessed after
imputations, which depend on number of missing subjects and number of imputations.
This could be checked using FMI and RVI, which ranged from 0.0017 to 0.691, 0.0009
to 0.6883, respectively. Also for a rule of thumb163 suggested that the number of
imputation should be larger than FMI × 100. For example, if FMI = 0.65, a number of
imputation larger than 65 imputations are required. Therefore, a number of 70
imputations would be insufficient for this dataset.
5.7 The use of GSEM model Structural equation modelling (SEM) is very powerful multivariate
statistical method for continuous data, which uses a conceptual model or a causal
pathway diagram by applying regression-style linked equation to capture complex and
dynamic relationships within a network of observed and unobserved variables. Although
SEM is similar with regression model, sometimes it does not require to clearly define
dependent and independent variables. In addition, dependent variables in one model can
Win Khaing Discussion / 124
be considered as independent variables in another model. SEM infers causal relationship
by applying these variables in reciprocal role.
The original “causal steps” approach by Baron and Kenny169 used a series
of regression equations, in which both causes and effects have to be assigned priori
either a cause or an effect. In mediation analysis, mediation assumes both causality and
a temporal ordering among exposure, mediator and outcome variables. Therefore,
Iacobucci et al202 pointed out that SEM approaches dominate because SEM can estimate
all these simultaneously instead of assuming series of equations.
There are many advantages to using the SEM framework in the context of
mediation analysis. First, SEM simplifies testing of mediation hypotheses by designing
complicated mediation model in a single analysis, and allows for ease of interpretation
and estimation. SEM can also extend a mediation process from multiple independent
with single mediator to multiple mediators pathways. Regression method approach can
apply to only a simple causal pathway with only one or a few mediators considered.
Estimation of indirect/mediation effects and variances from this approach is easier, if
there are more than two or three pathways.
Second, the GSEM, an extension of SEM can easily handle family of data
distributions (e.g., gaussian, bernoulli, exponential, ordinal, poisson, Weibull, etcetera)
and link functions (e.g., identity, log, logit, probit, cloglog, etcetera) for cross-sectional,
longitudinal data, and even multilevel data. A mediator can be repeatedly
examined/measured, which can capture or identify temporal relationship with the
outcome over time better than measured at once.
Third, Bollen and Pearl203 approved that the result from SEM and standard
regression analysis showed the different results even for the same equation used because
SEM implies a functional relationship expressed via conceptual model, path diagram,
and mathematical equation, while regression analysis implies a statistical relationship
based on a conditional expected value. Therefore, SEM is more appropriate for
discovering the simultaneous nature of the indirect and direct effects, and the causal
relationships in a hypothesized mediation process in which a mediator plays the dual
role of both a cause for the outcome and an effect of the exposure.
Selection of variables for the mediation and outcome models are affected on
results of casual inferences. Traditional stepwise variable selections used general
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 125
predictive models might be not applicable to the causal inference model204, 205.
Generally, stepwise variable selection is an optimal method for statistical modelling,
which aims to limit or minimize a set of predictors but still can well explain variation of
dependent variable. Conversely, selection of variables for the casual pathway models is
aimed to get as many as variables that can explain all causal pathway models, which
consist of at least two models (i.e., medication and outcome models) or more.
Constructing these causal pathway model is much required background knowledge in
pathophysiology and also mechanism of event/disease occurrences for drawing causal
pathway204, 205. Although, stepwise variable selection is extremely useful in prediction
models, it might be useful if selection is performed based on clearly defined causal
pathway. Although some methodologies206-209 for selection of variables have been
purposed, there is no standard guideline to solve this problem because each method has
its own limitations which depend on the context in which it is applied. Our study had
therefore firstly considered conceptual frame work of the mediation and outcome
models. This led to gather and collect relevant variables that might explain mediator and
outcome variables. Selection of variables was then performed separately by mediation
and outcome model using forward selections based on the casual pathways.
Finally, model fit information from SEM analysis is also another advantage
over standard regression method. After constructing the mediation model, the model can
be verified about the consistency of the hypothesized mediational model to the data and
can be tested for the evidence for the plausibility of the causality assumptions.
5.8 Strengths of this study There are some strengths in this study. First, the study design was
prospective cohort with repeated measurements every 5 years. Data were longitudinal,
which could assess temporal relationships between studied factors and outcome of
interest. The robust results can be obtained from such comprehensive studies. Second,
EGAT1 cohort survey is still an ongoing survey with a very good documentary systems.
The characteristics of the subjects with complete information of cardiovascular risk
factors were considered, evaluated, systematically collected and then computerized.
Win Khaing Discussion / 126
Furthermore, the follow up survey was conducted every 5 years since 1997 (EGAT1/2),
and therefore, resulted in excellent detection of outcomes. The detection of the outcomes
of interest was quite complete. The vital status of all subjects was 100% accurate
because all possible vital event databases from Ministry of Interior, Bureau Policy and
Strategy, Ministry of Public Health had been confirmed. Additionally, outcome
verification process was also very strong because all medical records, hospital records,
report for health care coverage schemes, reimbursement for medical expenses were
checked for outcome assessment. The causes of deaths were also verified and confirmed
by an outcome ascertainment team, which consisted of many specialty experts.
This study also applied appropriate MICE method in longitudinal format for
interested missing variables. This study applied a mediation analysis with GSEM to
determine the direct and mediated effects of education through income on MCVE. This
method is claimed to be a more appropriate statistical method to fulfil the objective of
this study as described above.
5.9 Limitations of this study This study also had some limitations. First, both education and income were
self-reported, which may lead to un-avoidable information bias in this study. Although
these data could be verified from the Department of Human Resources of EGAT, we
could not access to the data. We considered only income which does not reflect wealthy
and socio-economic status. Thus, retried subjects might have low income but they were
still wealthy.
Another drawback of this study was that all the subjects of EGAT1 cohort
were employees of EGAT enterprises, which is said to be one of the largest enterprises
in Thailand. Although a wide range of socio-demographic backgrounds with more than
30 different strata of job positions were included, it could not be said to be representative
of Thailand as a whole. The socio-economic status of EGAT employees, and baseline
CVD risk factor profile may not be comparable with those economically disadvantaged
Thais in the entire country. Because severely ill and disabled are ordinarily excluded
from employment, the ‘healthy worker effect’ would come into play and the death rate
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 127
was likely to be lower than those in the general population. The EGAT1 study
population, although large in sample size, included subjects’ lifestyle and CVD risk
factors may not align with those in the entire country. Due to excellent health coverage
of the EGAT employees, it was possible that MCVE outcomes were likely to be lower
than those in the general population.
5.10 Clinical application To the best of our knowledge, this is the first study to examine causal
pathway analysis between education, income and MCVE using a contemporary
statistical methodology. Certain clinical implications may arise from our results.
Education and income are not routinely collected in clinical settings and are often
neglected as they are outside the health care system and are not under physicians’
control. Findings from this study provide additional clinical value to assessing a
patient’s education and income level to improve clinical decision making. In addition,
this study confirmed that effect of education on MCVE were prominently mediated by
income pathway. This results should raise the questions regarding the benefit of
considering education or income in predicting risk of MCVE in complex pathways
linking socioeconomic, behavioral, and biological causes. Fiscella et al210 had evaluated
the impact of adding SDH into predictive models. They found that there was small
improvement in the overall prediction with modest contribution and value especially in
the subgroup of subjects who are most socially disadvantaged. Definitely, the study
results still needs to be confirmed before they can be practiced in clinical setting, but in
general, physicians should reflect on the patients’ background characteristics especially
focus on socioeconomic factors together with other risk factors when clinical decision-
making in both preventive and curative aspects of CVD.
5.11 Suggestion for further studies This study suggested that the majority of the effects of education on MCVE
were mediated through income. By considering the dynamic concept of the education
Win Khaing Discussion / 128
or income, and viewing them in biological and epidemiological contexts, we can better
understanding about dynamic relationships that underlie the association between
education, income and CVD risk factors and events in order to design more effective
primary and secondary preventions. WHO global action plan for the prevention and
control of NCD (2013–2020)211 was targeted to reduce premature mortality of CVD to
25% relative to 2010 by the year 2025 (called “25 × 25 target”). In order to reach this
target, we need insight understanding of how education and income translates into
differential risks within and between population groups, especially in disadvantaged
groups. Majority of previous studies examined the combined effect of education and
income with CVD risk factors and events. Few have examined the interrelated or
separated effects of education or income with CVD outcome with little understanding
of the pathways through which education and income are related to CVD risk factors
and events. Therefore, further studies should not rely on any single socioeconomic
indicators and ignore the others. If the complex pathways between socioeconomic
indicators are often neglected, the risk of being fruitless in some research may occur
even in well design research.
The association between education and MCVE needs to be confirmed in a
large cohort with sufficient sample size and adequate number of outcomes of interest in
a general population with different ethnicities, different age stratifications, and different
working environments. If possible, detailed information about education and income
should be collected as discussed above.
Although this study increases the insight in some important causal pathway
between education, income and MCVE, many important SDH dimensions (i.e., health
literacy, working environment, unemployment, job instability, social isolation and
discrimination, etcetera) are still lack examination. If there is possibility of available
data available, future mediation analysis studies with these variables are encouraged.
5.12 Conclusion This study provided evidence of causal relationship among education and
MCVE through income. The effects of education on MCVE were largely mediated by
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 129
income. The mediated effect of income was greater than the direct effect of education
on MCVE.
Win Khaing References / 130
REFERENCES
1. Mendis S, Armstrong T, Bettcher D, Branca F, Lauer J, Mace C, et al. Chapter 1.
Global target 1: A 25% relative reduction in the overall mortality from
cardiovascular diseases, cancer, diabetes, or chronic respiratory diseases.
World Health Organization: Global status report on noncommunicable
diseases 2014. 1st ed. Switzerland: World Health Organization; 2014. p. 9.
2. Kannel WB, McGee DL. Diabetes and cardiovascular disease: the Framingham study.
Jama. 1979;241(19):2035-8.
3. Keil U, Kuulasmaa K. WHO MONICA Project: risk factors. Int J Epidemiol.
1989;18(3 Suppl 1):S46-55.
4. Yusuf S, Hawken S, Ôunpuu S, Dans T, Avezum A, Lanas F, et al. Effect of
potentially modifiable risk factors associated with myocardial infarction in
52 countries (the INTERHEART study): case-control study. The Lancet.
2004;364(9438):937-52.
5. Murray CJ, Lopez AD. Measuring the global burden of disease. New England Journal
of Medicine. 2013;369(5):448-57.
6. Feigin VL, Roth GA, Naghavi M, Parmar P, Krishnamurthi R, Chugh S, et al. Global
burden of stroke and risk factors in 188 countries, during 1990–2013: a
systematic analysis for the Global Burden of Disease Study 2013. The
Lancet Neurology. 2016.
7. Lim SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair-Rohani H, et al. A
comparative risk assessment of burden of disease and injury attributable to
67 risk factors and risk factor clusters in 21 regions, 1990–2010: a
systematic analysis for the Global Burden of Disease Study 2010. The
lancet. 2013;380(9859):2224-60.
8. Hemingway H, Philipson P, Chen R, Fitzpatrick NK, Damant J, Shipley M, et al.
Evaluating the quality of research into a single prognostic biomarker: a
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 131
systematic review and meta-analysis of 83 studies of C-reactive protein in
stable coronary artery disease. PLoS Med. 2010;7(6):e1000286.
9. Collaboration L-PS. Lipoprotein-associated phospholipase A 2 and risk of coronary
disease, stroke, and mortality: collaborative analysis of 32 prospective
studies. The Lancet. 2010;375(9725):1536-44.
10. Collaboration HS. Homocysteine and risk of ischemic heart disease and stroke: a
meta-analysis. Jama. 2002;288(16):2015-22.
11. Hoogeveen RC, Gaubatz JW, Sun W, Dodge RC, Crosby JR, Jiang J, et al. Small
Dense LDL Cholesterol Concentrations Predict Risk for Coronary Heart
Disease: the Atherosclerosis Risk in Communities (ARIC) Study.
Arteriosclerosis, thrombosis, and vascular biology. 2014;34(5):1069.
12. Ernst E, Resch KL. Fibrinogen as a cardiovascular risk factor: a meta-analysis and
review of the literature. Annals of Internal Medicine. 1993;118(12):956-63.
13. Yusuf S, Reddy S, Ôunpuu S, Anand S. Global burden of cardiovascular diseases
part I: general considerations, the epidemiologic transition, risk factors, and
impact of urbanization. Circulation. 2001;104(22):2746-53.
14. Lang T, Lepage B, Schieber A-C, Lamy S, Kelly-Irving M. Social determinants of
cardiovascular diseases. Public Health Reviews. 2012;33(2):601-22.
15. McKee M, Chow CK. In: Yusuf S, Cairns JA, Camm AJ, Fallen EL, BJ G, editors.
Evidence-Based Cardiology, Third Edition2010. p. 211-20.
16. Rose G, Marmot M. Social class and coronary heart disease. British heart journal.
1981;45(1):13-9.
17. Smith GD, Bartley M, Blane D. The Black report on socioeconomic inequalities in
health 10 years on. BMJ: British Medical Journal. 1990;301(6748):373.
18. Marmot MG, Stansfeld S, Patel C, North F, Head J, White I, et al. Health inequalities
among British civil servants: the Whitehall II study. The Lancet.
1991;337(8754):1387-93.
19. Johnson JL, Heineman EF, Heiss G, Hames CG, Tyroler HA. Cardiovascular disease
risk factors and mortality among black women and white women aged 40–
64 years in Evans County, Georgia. American journal of epidemiology.
1986;123(2):209-20.
Win Khaing References / 132
20. Lin CC, Rogot E, Johnson NJ, Sorlie PD, Arias E. A further study of life expectancy
by socioeconomic factors in the National Longitudinal Mortality Study.
Ethnicity & disease. 2002;13(2):240-7.
21. Nietert PJ, Sutherland SE, Keil JE, Bachman DL. Demographic and biologic
influences on survival in whites and blacks: 40 years of follow-up in the
Charleston heart study. International journal for equity in health.
2006;5(1):8 - 16.
22. Beebe-Dimmer J, Lynch JW, Turrell G, Lustgarten S, Raghunathan T, Kaplan GA.
Childhood and Adult Socioeconomic Conditions and 31-Year Mortality
Risk in Women. American Journal of Epidemiology. 2004;159(5):481-90.
23. Jacobsen BK, Thelle DS. Risk factors for coronary heart disease and level of
education the tromsø heart study. American Journal of Epidemiology.
1988;127(5):923-32.
24. Cirera L, Tormo M-J, Chirlaque M-D, Navarro C. Cardiovascular risk factors and
educational attainment in Southern Spain: a study of a random sample of
3091 adults. European Journal of Epidemiology. 1998;14(8):755-63.
25. Panagiotakos D, Georgousopoulou E, Notara V, Pitaraki E, Kokkou E, Chrysohoou
C, et al. Education status determines 10-year (2002-2012) survival from
cardiovascular disease in Athens metropolitan area: the ATTICA study,
Greece. Health & social care in the community. 2016;24(3):334-44.
26. Mackenbach JP, Cavelaars A, Kunst AE, Groenhof F. Socioeconomic inequalities
in cardiovascular disease mortality. An international study. European heart
journal. 2000;21(14):1141-51.
27. Gallo LC, Matthews KA, Kuller LH, Sutton-Tyrrell K, Edmundowicz D.
Educational attainment and coronary and aortic calcification in
postmenopausal women. Psychosomatic medicine. 2001;63(6):925-35.
28. Notara V, Panagiotakos D, Kogias Y, Stravopodis P, Antonoulas A, Zombolos S, et
al. The impact of education status on the 10-year (2004-2014)
cardiovascular disease incidence and all cause mortality, among Acute
Coronary Syndrome patients: the GREECS longitudinal study. Journal of
Preventive Medicine and Public Health. 2016.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 133
29. Andersen I, Osler M, Petersen L, Grønbæk M, Prescott E. Income and risk of
ischaemic heart disease in men and women in a Nordic welfare country.
International Journal of Epidemiology. 2003;32(3):367-74.
30. Salomaa V, Niemelä M, Miettinen H, Ketonen M, Immonen-Räihä P, Koskinen S,
et al. Relationship of socioeconomic status to the incidence and prehospital,
28-day, and 1-year mortality rates of acute coronary events in the
FINMONICA myocardial infarction register study. Circulation.
2000;101(16):1913-8.
31. Lammintausta A, Immonen-Räihä P, Airaksinen JKE, Torppa J, Harald K, Ketonen
M, et al. Socioeconomic Inequalities in the Morbidity and Mortality of
Acute Coronary Events in Finland: 1988 to 2002. Annals of Epidemiology.
2012;22(2):87-93.
32. Alter DA, Chong A, Austin PC, Mustard C, Iron K, Williams JI, et al.
Socioeconomic status and mortality after acute myocardial infarction.
Annals of internal medicine. 2006;144(2):82-93.
33. Rao SV, Schulman KA, Curtis LH, Gersh BJ, Jollis JG. Socioeconomic status and
outcome following acute myocardial infarction in elderly patients. Archives
of internal medicine. 2004;164(10):1128-33.
34. Fernald LC, Adler NE. Blood pressure and socioeconomic status in low-income
women in Mexico: a reverse gradient? J Epidemiol Community Health.
2008;62(5):e8.
35. Xu F, Tse LA, Yin X, Yu IT-s, Griffiths S. Impact of socio-economic factors on
stroke prevalence among urban and rural residents in Mainland China. BMC
Public Health. 2008;8(1):1.
36. Cox AM, McKevitt C, Rudd AG, Wolfe CD. Socioeconomic status and stroke. The
Lancet Neurology. 2006;5(2):181-8.
37. Kaplan GA, Keil JE. Socioeconomic factors and cardiovascular disease: a review of
the literature. Circulation. 1993;88(4):1973-98.
38. Marshall IJ, Wang Y, Crichton S, McKevitt C, Rudd AG, Wolfe CD. The effects of
socioeconomic status on stroke risk and outcomes. The Lancet Neurology.
2015;14(12):1206-18.
Win Khaing References / 134
39. Chen R, Hu Z, Chen R-L, Zhang D, Xu L, Wang J, et al. Socioeconomic deprivation
and survival after stroke in China: a systematic literature review and a new
population-based cohort study. BMJ open. 2015;5(1):e005688.
40. Manrique-Garcia E, Sidorchuk A, Hallqvist J, Moradi T. Socioeconomic position
and incidence of acute myocardial infarction: a meta-analysis. Journal of
Epidemiology and Community Health. 2011;65(4):301-9.
41. Vathesatogkit P, Batty GD, Woodward M. Socioeconomic disadvantage and
disease-specific mortality in Asia: systematic review with meta-analysis of
population-based cohort studies. J Epidemiol Community Health.
2014;68(4):375-83.
42. Khaing W, Vallibhakara SA, Attia J, McEvoy M, Thakkinstian A. Effects of
education and income on cardiovascular outcomes: A systematic review and
meta-analysis. European Journal of Preventive Cardiology.
2017;24(10):1032-42.
43. Gerber Y, Goldbourt U, Drory Y. Interaction between income and education in
predicting long-term survival after acute myocardial infarction. European
journal of cardiovascular prevention and rehabilitation : official journal of
the European Society of Cardiology, Working Groups on Epidemiology &
Prevention and Cardiac Rehabilitation and Exercise Physiology.
2008;15(5):526-32.
44. Rasmussen JN, Rasmussen S, Gislason GH, Buch P, Abildstrom SZ, Kober L, et al.
Mortality after acute myocardial infarction according to income and
education. J Epidemiol Community Health. 2006;60(4):351-6.
45. Lemstra M, Rogers M, Moraros J. Income and heart disease Neglected risk factor.
Canadian Family Physician. 2015;61(8):698-704.
46. Braveman PA, Cubbin C, Egerter S, Chideya S, Marchi KS, Metzler M, et al.
Socioeconomic status in health research: one size does not fit all. Jama.
2005;294(22):2879-88.
47. Levenson JW, Skerrett PJ, Gaziano JM. Reducing the global burden of
cardiovascular disease: the role of risk factors. Preventive cardiology.
2002;5(4):188-99.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 135
48. Olshansky SJ, Ault AB. The fourth stage of the epidemiologic transition: the age of
delayed degenerative diseases. The Milbank Quarterly. 1986:355-91.
49. World Health Organization. WHO | What are social determinants of health? : World
Health Organization; 2015
(http://www.who.int/social_determinants/sdh_definition/en/).
50. Winkleby MA, Jatulis DE, Frank E, Fortmann SP. Socioeconomic status and health:
how education, income, and occupation contribute to risk factors for
cardiovascular disease. Am J Public Health. 1992;82(6):816-20.
51. Havranek EP, Mujahid MS, Barr DA, Blair IV, Cohen MS, Cruz-Flores S, et al.
Social Determinants of Risk and Outcomes for Cardiovascular Disease A
Scientific Statement From the American Heart Association. Circulation.
2015;132(9):873-98.
52. Winkleby MA, Jatulis DE, Frank E, Fortmann SP. Socioeconomic status and health:
how education, income, and occupation contribute to risk factors for
cardiovascular disease. American journal of public health. 1992;82(6):816-
20.
53. Lynch JW, Kaplan GA, Cohen RD, Tuomilehto J, Salonen JT. Do cardiovascular
risk factors explain the relation between socioeconomic status, risk of all-
cause mortality, cardiovascular mortality, and acute myocardial infarction?
American journal of epidemiology. 1996;144(10):934-42.
54. Hallqvist J, Lundberg M, Diderichsen F, Ahlbomb A. Socioeconomic differences in
risk of myocardial infarction 1971–1994 in Sweden: time trends, relative
risks and population attributable risks. International Journal of
Epidemiology. 1998;27(3):410-5.
55. Hart CL, Hole DJ, Smith GD. The contribution of risk factors to stroke differentials,
by socioeconomic position in adulthood: the Renfrew/Paisley Study.
American Journal of Public Health. 2000;90(11):1788.
56. Avendano M, Kawachi I, Van Lenthe F, Boshuizen HC, Mackenbach JP, Van den
Bos GA, et al. Socioeconomic status and stroke incidence in the US elderly:
the role of risk factors in the EPESE study. Stroke. 2006;37(6):1368-73.
Win Khaing References / 136
57. Berkman LF, Glass TA. Social integration, social networks, social support and
health. In: Berkman LF, Kawachi I, editors. Soical Epidemiology. New
York: Oxford University Press; 2000. p. 137 - 73.
58. Pele L. Currency converter in the past with official exchange rates from 1953 2016
(http://fxtop.com/en/currency-converter-past.php). (Accessed 31 August
2016).
59. White IR. Multivariate random-effects meta-regression: updates to mvmeta. Stata
Journal. 2011;11(2):255-70.
60. World Bank. World Bank Country and Lending Groups World Bank Data Help
Desk 2016
(https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-
world-bank-country-and-lending-groups). (Accessed 31 August 2016).
61. StataCorp. Stata Statistical Software: Release 14. College Station, TX: StataCorp
LP. 2015.
62. Naska A, Katsoulis M, Trichopoulos D, Trichopoulou A. The root causes of
socioeconomic differentials in cancer and cardiovascular mortality in
Greece. European Journal of Cancer Prevention. 2012;21(5):490-6.
63. Bostock S, Steptoe A. Association between low functional health literacy and
mortality in older adults: longitudinal cohort study. Bmj. 2012;344:e1602.
64. Meijer A, Conradi HJ, Bos EH, Thombs BD, van Melle JP, de Jonge P. Prognostic
association of depression following myocardial infarction with mortality
and cardiovascular events: a meta-analysis of 25 years of research. General
hospital psychiatry. 2011;33(3):203-16.
65. Myers V, Gerber Y, Benyamini Y, Goldbourt U, Drory Y. Post-myocardial
infarction depression: increased hospital admissions and reduced adoption
of secondary prevention measures—a longitudinal study. Journal of
psychosomatic research. 2012;72(1):5-10.
66. Nielsen KM, Faergeman O, Foldspang A, Larsen ML. Cardiac rehabilitation: health
characteristics and socio-economic status among those who do not attend.
The European Journal of Public Health. 2008;18(5):479-83.
67. Kilander L, Berglund L, Boberg M, Vessby B, Lithell H. Education, lifestyle factors
and mortality from cardiovascular disease and cancer. A 25-year follow-up
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 137
of Swedish 50-year-old men. International Journal of Epidemiology.
2001;30(5):1119-26.
68. Steptoe A, Marmot M. The role of psychobiological pathways in socio-economic
inequalities in cardiovascular disease risk. European heart journal.
2002;23(1):13-25.
69. Suadicani P, Hein HO, Gyntelberg F. Strong mediators of social inequalities in risk
of ischaemic heart disease: a six-year follow-up in the Copenhagen Male
Study. International Journal of Epidemiology. 1997;26(3):516-22.
70. Blair AS, Lloyd-Williams F, Mair FS. What do we know about socioeconomic status
and congestive heart failure? A review of the literature. The Journal of
family practice. 2002;51(2):169-.
71. Wilkinson RG, Pickett KE. Income inequality and population health: a review and
explanation of the evidence. Social science & medicine. 2006;62(7):1768-
84.
72. Lahelma E, Martikainen P, Laaksonen M, Aittomäki A. Pathways between
socioeconomic determinants of health. Journal of Epidemiology and
Community Health. 2004;58(4):327-32.
73. Ahmed AA, Zhang Y, Bourge RC, Kilgore ML, Williams B, Sawyer P, et al.
Abstract 12064: Low Income, Regardless of Education Level, is a
Significant Independent Predictor of Incident Heart Failure in Community-
Dwelling, Medicare-Eligible Older Adults. Circulation. 2011;124(Suppl
21):A12064-A.
74. Shavers VL. Measurement of socioeconomic status in health disparities research.
Journal of the national medical association. 2007;99(9):1013.
75. Marmot M, Ryff CD, Bumpass LL, Shipley M, Marks NF. Social inequalities in
health: next questions and converging evidence. Social science & medicine.
1997;44(6):901-10.
76. Marmot MG, Rose G, Shipley M, Hamilton PJ. Employment grade and coronary
heart disease in British civil servants. Journal of epidemiology and
community health. 1978;32(4):244-9.
Win Khaing References / 138
77. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for
systematic reviews and meta-analyses: the PRISMA statement. Annals of
internal medicine. 2009;151(4):264-9.
78. Andersen I, Gamborg M, Osler M, Prescott E, Diderichsen F. Income as mediator
of the effect of occupation on the risk of myocardial infarction: does the
income measurement matter? Journal of epidemiology and community
health. 2005;59(12):1080-5.
79. Lu Y, Hajifathalian K, Rimm EB, Ezzati M, Danaei G. Mediators of the effect of
body mass index on coronary heart disease: decomposing direct and indirect
effects. Epidemiology. 2015;26(2):153-62.
80. MacKinnon DP, Fairchild AJ, Fritz MS. Mediation analysis. Annual review of
psychology. 2007;58:593.
81. Hoeymans N, Smit H, Verkleij H, Kromhout D. Cardiovascular risk factors in
relation to educational level in 36 000 men and women in The Netherlands.
European Heart Journal. 1996;17(4):518-25.
82. Luepker RV, Rosamond WD, Murphy R, Sprafka JM, Folsom AR, McGovern PG,
et al. Socioeconomic status and coronary heart disease risk factor trends.
The Minnesota Heart Survey. Circulation. 1993;88(5):2172-9.
83. Zaman MJ, Patel A, Jan S, Hillis GS, Raju PK, Neal B, et al. Socio-economic
distribution of cardiovascular risk factors and knowledge in rural India.
International journal of epidemiology. 2012:dyr226.
84. Ramsay SE, Morris RW, Whincup PH, Papacosta AO, Thomas MC, Wannamethee
SG. Prediction of coronary heart disease risk by Framingham and SCORE
risk assessments varies by socioeconomic position: results from a study in
British men. European Journal of Cardiovascular Prevention &
Rehabilitation. 2011;18(2):186-93.
85. Tunstall-Pedoe H, Woodward M. By neglecting deprivation, cardiovascular risk
scoring will exacerbate social gradients in disease. Heart. 2006;92(3):307-
10.
86. Molshatzki N, Drory Y, Myers V, Goldbourt U, Benyamini Y, Steinberg DM, et al.
Role of socioeconomic status measures in long-term mortality risk
prediction after myocardial infarction. Medical care. 2011;49(7):673-8.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 139
87. Gerber Y, Goldbourt U, Drory Y. Interaction between income and education in
predicting long-term survival after acute myocardial infarction. European
Journal of Cardiovascular Prevention & Rehabilitation. 2008;15(5):526-32.
88. Arrich J, Lalouschek W, Müllner M. Influence of socioeconomic status on mortality
after stroke: Retrospective cohort study. Stroke. 2005;36(2):310-4.
89. Rehkopf DH, Eisen EA, Modrek S, Mokyr Horner E, Goldstein B, Costello S, et al.
Early-Life State-of-Residence Characteristics and Later Life Hypertension,
Diabetes, and Ischemic Heart Disease. Am J Public Health.
2015;105(8):1689-95.
90. Geyer S, Hemström Ö, Peter R, Vågerö D. Education, income, and occupational
class cannot be used interchangeably in social epidemiology. Empirical
evidence against a common practice. Journal of epidemiology and
community health. 2006;60(9):804-10.
91. Honjo K, Iso H, Inoue M, Tsugane S, Group JS. Education, Social Roles, and the
Risk of Cardiovascular Disease Among Middle-Aged Japanese Women The
JPHC Study Cohort I. Stroke. 2008;39(10):2886-90.
92. Rawshani A, Svensson AM, Rosengren A, Eliasson B, Gudbjörnsdottir S. Impact of
socioeconomic status on cardiovascular disease and mortality in 24,947
individuals with type 1 diabetes. Diabetes Care. 2015;38(8):1518-27.
93. Thurston RC, Kubzansky LD, Kawachi I, Berkman LF. Is the association between
socioeconomic position and coronary heart disease stronger in women than
in men? American Journal of Epidemiology. 2005;162(1):57-65.
94. Hetemaa T, Manderbacka K, Reunanen A, Koskinen S, Keskimäki I. Socioeconomic
inequities in invasive cardiac procedures among patients with incident
angina pectoris or myocardial infarction. Scandinavian Journal of Public
Health. 2006;34(2):116-23.
95. Peter R, Gässler H, Geyer S. Socioeconomic status, status inconsistency and risk of
ischaemic heart disease: A prospective study among members of a statutory
health insurance company. Journal of Epidemiology and Community
Health. 2007;61(7):605-11.
96. Honjo K, Tsutsumi A, Kayaba K, Group JMSCS. Socioeconomic indicators and
cardiovascular disease incidence among Japanese community residents: the
Win Khaing References / 140
Jichi Medical School Cohort Study. International journal of behavioral
medicine. 2010;17(1):58-66.
97. Roux AVD, Merkin SS, Arnett D, Chambless L, Massing M, Nieto FJ, et al.
Neighborhood of residence and incidence of coronary heart disease. New
England Journal of Medicine. 2001;345(2):99-106.
98. Fujino Y, Tamakoshi A, Iso H, Inaba Y, Kubo T, Ide R, et al. A nationwide cohort
study of educational background and major causes of death among the
elderly population in Japan. Preventive medicine. 2005;40(4):444-51.
99. Lee Y-T, Lin RS, Sung FC, Yang C-Y, Chien K-L, Chen W-J, et al. Chin-Shan
Community Cardiovascular Cohort in Taiwan–baseline data and five-year
follow-up morbidity and mortality. Journal of clinical epidemiology.
2000;53(8):838-46.
100. Weikert C, Stefan N, Schulze MB, Pischon T, Berger K, Joost H-G, et al. Plasma
fetuin-a levels and the risk of myocardial infarction and ischemic stroke.
Circulation. 2008;118(24):2555-62.
101. Hippe M, Vestbo J, Hein HO, Borch-Johnsen K, Jensen G, Sørensen T. Familial
predisposition and susceptibility to the effect of other risk factors for
myocardial infarction. Journal of epidemiology and community health.
1999;53(5):269-76.
102. Huisman M, Van Lenthe F, Avendano M, Mackenbach J. The contribution of job
characteristics to socioeconomic inequalities in incidence of myocardial
infarction. Social science & medicine. 2008;66(11):2240-52.
103. Eaker ED, Pinsky J, Castelli WP. Myocardial infarction and coronary death among
women: psychosocial predictors from a 20-year follow-up of women in the
Framingham Study. American Journal of Epidemiology. 1992;135(8):854-
64.
104. Bosma H, Appels A, Sturmans F, Grabauskas V, Gostautas A. Educational level
of spouses and risk of mortality: the WHO Kaunas-Rotterdam Intervention
Study (KRIS). International Journal of Epidemiology. 1995;24(1):119-26.
105. Chaix B, Rosvall M, Merlo J. Neighborhood socioeconomic deprivation and
residential instability: effects on incidence of ischemic heart disease and
survival after myocardial infarction. Epidemiology. 2007;18(1):104-11.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 141
106. Kuper H, Adami H-O, Theorell T, Weiderpass E. Psychosocial determinants of
coronary heart disease in middle-aged women: a prospective study in
Sweden. American Journal of Epidemiology. 2006;164(4):349-57.
107. Lapidus L, Bengtsson C. Socioeconomic factors and physical activity in relation to
cardiovascular disease and death. A 12 year follow up of participants in a
population study of women in Gothenburg, Sweden. British heart journal.
1986;55(3):295-301.
108. Braig S, Peter R, Nagel G, Hermann S, Rohrmann S, Linseisen J. The impact of
social status inconsistency on cardiovascular risk factors, myocardial
infarction and stroke in the EPIC-Heidelberg cohort. BMC Public Health.
2011;11.
109. Jakobsen L, Niemann T, Thorsgaard N, Thuesen L, Lassen JF, Jensen LO, et al.
Dimensions of socioeconomic status and clinical outcome after primary
percutaneous coronary intervention. Circulation Cardiovascular
interventions. 2012;5(5):641-8.
110. Rasmussen JN, Rasmussen S, Gislason GH, Abildstrom SZ, Schramm TK, Torp-
Pedersen C, et al. Persistent socio-economic differences in revascularization
after acute myocardial infarction despite a universal health care system - A
Danish study. Cardiovascular Drugs and Therapy. 2007;21(6):449-57.
111. Senan M, Petrosyan A. The relationship between socioeconomic status and
cardiovascular events. Georgian medical news. 2014(227):42-7.
112. Bosma H, Van Jaarsveld C, Tuinstra J, Sanderman R, Ranchor A, Van Eijk JTM,
et al. Low control beliefs, classical coronary risk factors, and socio-
economic differences in heart disease in older persons. Social science &
medicine. 2005;60(4):737-45.
113. Masoudkabir F, Toghianifar N, Talaie M, Sadeghi M, Sarrafzadegan N,
Mohammadifard N, et al. Socioeconomic status and incident cardiovascular
disease in a developing country: Findings from the Isfahan cohort study
(ICS). International Journal of Public Health. 2012;57(3):561-8.
114. Van Minh H, Huong DL, Wall S, Byass P, Chuc NTK. Peer Reviewed:
Cardiovascular Disease Mortality and Its Association With Socioeconomic
Win Khaing References / 142
Status: Findings From a Population-based Cohort Study in Rural Vietnam,
1999–2003. Preventing chronic disease. 2006;3(3).
115. Hirokawa K, Tsutusmi A, Kayaba K. Impacts of educational level and employment
status on mortality for Japanese women and men: the Jichi Medical School
cohort study. European journal of epidemiology. 2006;21(9):641-51.
116. Siegel D, Kuller L, Lazarus NB, Black D, Feigal D, Hughes G, et al. Predictors of
cardiovascular events and mortality in the Systolic Hypertension in the
Elderly Program pilot project. American journal of epidemiology.
1987;126(3):385-99.
117. He J, Ogden LG, Bazzano LA, Vupputuri S, Loria C, Whelton PK. Risk factors for
congestive heart failure in US men and women: NHANES I epidemiologic
follow-up study. Archives of internal medicine. 2001;161(7):996-1002.
118. Christensen S, Mogelvang R, Heitmann M, Prescott E. Level of education and risk
of heart failure: a prospective cohort study with echocardiography
evaluation. European heart journal. 2011;32(4):450-8.
119. Borné Y, Engström G, Essén B, Sundquist J, Hedblad B. Country of birth and risk
of hospitalization due to heart failure: a Swedish population-based cohort
study. European journal of epidemiology. 2011;26(4):275-83.
120. Philbin EF, Dec GW, Jenkins PL, DiSalvo TG. Socioeconomic status as an
independent risk factor for hospital readmission for heart failure. The
American journal of cardiology. 2001;87(12):1367-71.
121. Schwarz KA, Elman CS. Identification of factors predictive of hospital
readmissions for patients with heart failure. Heart & Lung: The Journal of
Acute and Critical Care. 2003;32(2):88-99.
122. Sui X, Gheorghiade M, Zannad F, Young JB, Ahmed A. A propensity matched
study of the association of education and outcomes in chronic heart failure.
International journal of cardiology. 2008;129(1):93-9.
123. Rosvall M, Engström G, Hedblad B, Janzon L, Göran B. The role of preclinical
atherosclerosis in the explanation of educational differences in incidence of
coronary events. Atherosclerosis. 2006;187(2):251-6.
124. Engström G, Tydén P, Berglund G, Hansen O, Hedblad B, Janzon L. Incidence of
myocardial infarction in women. A cohort study of risk factors and
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 143
modifiers of effect. Journal of epidemiology and community health.
2000;54(2):104-7.
125. Avendano M, Glymour MM. Stroke disparities in older Americans: is wealth a
more powerful indicator of risk than income and education? Stroke.
2008;39(5):1533-40.
126. Li C, Hedblad B, Rosvall M, Buchwald F, Khan FA, Engström G. Stroke incidence,
recurrence, and case-fatality in relation to socioeconomic position a
population-based study of middle-aged swedish men and women. Stroke.
2008;39(8):2191-6.
127. van Rossum CT, van de Mheen H, Breteler MM, Grobbee DE, Mackenbach JP.
Socioeconomic differences in stroke among Dutch elderly women the
Rotterdam Study. Stroke. 1999;30(2):357-62.
128. Gillum R, Mussolino ME. Education, poverty, and stroke incidence in whites and
blacks: the NHANES I Epidemiologic Follow-up Study. Journal of clinical
epidemiology. 2003;56(2):188-95.
129. Kuper H, Adami H-O, Theorell T, Weiderpass E. The socioeconomic gradient in
the incidence of stroke a prospective study in middle-aged women in
Sweden. Stroke. 2007;38(1):27-33.
130. Jackson CA, Jones M, Mishra GD. Educational and homeownership inequalities in
stroke incidence: a population-based longitudinal study of mid-aged
women. The European Journal of Public Health. 2014;24(2):231-6.
131. Andersen KK, Steding-Jessen M, Dalton SO, Olsen TS. Socioeconomic position
and incidence of ischemic stroke in denmark 2003-2012. A nationwide
hospital-based study. Journal of the American Heart Association. 2014;3(4).
132. Jakovljević D, Sarti C, Sivenius J, Torppa J, Mähönen M, Immonen-Räihä P, et al.
Socioeconomic status and ischemic stroke: The FINMONICA stroke
register. Stroke. 2001;32(7):1492-8.
133. Zhou G, Liu X, Xu G, Liu X, Zhang R, Zhu W. The effect of socioeconomic status
on three-year mortality after first-ever ischemic stroke in Nanjing, China.
BMC Public Health. 2006;6.
Win Khaing References / 144
134. Kim C, Eby E, Piette JD. Is education associated with mortality for breast cancer
and cardiovascular disease among black and white women? Gender
Medicine. 2005;2(1):13-8.
135. Qureshi AI, Suri MFK, Saad M, Hopkins LN. Educational attainment and risk of
stroke and myocardial infarction. Medical Science Monitor.
2003;9(11):CR466-CR73.
136. Pednekar MS, Gupta R, Gupta PC. Illiteracy, low educational status, and
cardiovascular mortality in India. BMC Public Health. 2011;11(1):1.
137. Coady SA, Johnson NJ, Hakes JK, Sorlie PD. Individual education, area income,
and mortality and recurrence of myocardial infarction in a Medicare cohort:
the National Longitudinal Mortality Study. BMC Public Health.
2014;14:705.
138. Gallo V, Mackenbach JP, Ezzati M, Menvielle G, Kunst AE, Rohrmann S, et al.
Social inequalities and mortality in Europe–results from a large multi-
national cohort. PLoS One. 2012;7(7):e39013.
139. Bucher HC, Ragland DR. Socioeconomic indicators and mortality from coronary
heart disease and cancer: a 22-year follow-up of middle-aged men. Am J
Public Health. 1995;85(9):1231-6.
140. Tonne C, Schwartz J, Mittleman M, Melly S, Suh H, Goldberg R. Long-term
survival after acute myocardial infarction is lower in more deprived
neighborhoods. Circulation. 2005;111(23):3063-70.
141. Ito S, Takachi R, Inoue M, Kurahashi N, Iwasaki M, Sasazuki S, et al. Education
in relation to incidence of and mortality from cancer and cardiovascular
disease in Japan. The European Journal of Public Health. 2008;18(5):466-
72.
142. Van Minh H, Byass P, Wall S. Mortality from cardiovascular diseases in Bavi
District, Vietnam. Scandinavian Journal of Public Health. 2003;31(6
suppl):26-31.
143. Liu K, Cedres LB, Stamler J, Dyer A, Stamler R, Nanas S, et al. Relationship of
education to major risk factors and death from coronary heart disease,
cardiovascular diseases and all causes, Findings of three Chicago
epidemiologic studies. Circulation. 1982;66(6):1308.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 145
144. Rawshani A, Svensson A-M, Zethelius B, Eliasson B, Rosengren A,
Gudbjörnsdottir S. Association Between Socioeconomic Status and
Mortality, Cardiovascular Disease, and Cancer in Patients With Type 2
Diabetes. JAMA Internal Medicine. 2016.
145. Rosvall M, Chaix B, Lynch J, Lindström M, Merlo J. The association between
socioeconomic position, use of revascularization procedures and five-year
survival after recovery from acute myocardial infarction. BMC Public
Health. 2008;8(1):1.
146. Khang Y-H, Lynch J, Jung-Choi K, Cho H-J. Explaining age specific inequalities
in mortality from all causes, cardiovascular disease and ischaemic heart
disease among South Korean male public servants: relative and absolute
perspectives. Heart. 2007.
147. Rosvall M, Gerward S, Engström G, Hedblad B. Income and short-term case
fatality after myocardial infarction in the whole middle-aged population of
Malmö, Sweden. The European Journal of Public Health. 2008;18(5):533-
8.
148. Vathesatogkit P, Woodward M, Tanomsup S, Ratanachaiwong W, Vanavanan S,
Yamwong S, et al. Cohort profile: the electricity generating authority of
Thailand study. Int J Epidemiol. 2012;41(2):359-65.
149. Thygesen K, Alpert JS, Jaffe AS, Simoons ML, Chaitman BR, White HD. Third
universal definition of myocardial infarction. Circulation.
2012;126(16):2020-35.
150. Sacco RL, Kasner SE, Broderick JP, Caplan LR, Culebras A, Elkind MS, et al. An
updated definition of stroke for the 21st century a statement for healthcare
professionals from the American Heart Association/American Stroke
Association. Stroke. 2013;44(7):2064-89.
151. Easton JD, Saver JL, Albers GW, Alberts MJ, Chaturvedi S, Feldmann E, et al.
Definition and Evaluation of Transient Ischemic Attack A Scientific
Statement for Healthcare Professionals From the American Heart
Association/American Stroke Association Stroke Council; Council on
Cardiovascular Surgery and Anesthesia; Council on Cardiovascular
Radiology and Intervention; Council on Cardiovascular Nursing; and the
Win Khaing References / 146
Interdisciplinary Council on Peripheral Vascular Disease: The American
Academy of Neurology affirms the value of this statement as an educational
tool for neurologists. Stroke. 2009;40(6):2276-93.
152. Nakarin Sansanayudh. The association between mean platelet volume and risk of
cardiovascular events: Mahidol University; 2015.
153. Alberti G, Zimmet P, Shaw J, Grundy S. International Diabetes Federation. The
IDF consensus worldwide definition of the Metabolic Syndrome.
International Diabetes Fundation publication 2006: 2-24.
154. Sritara P, Cheepudomwit S, Chapman N, Woodward M, Kositchaiwat C,
Tunlayadechanont S, et al. Twelve-year changes in vascular risk factors and
their associations with mortality in a cohort of 3499 Thais: the Electricity
Generating Authority of Thailand Study. International journal of
epidemiology. 2003;32(3):461-8.
155. Kitiyakara C, Yamwong S, Cheepudomwit S, Domrongkitchaiporn S,
Unkurapinun N, Pakpeankitvatana V, et al. The metabolic syndrome and
chronic kidney disease in a Southeast Asian cohort. Kidney international.
2007;71(7):693-700.
156. Gavin III JR, Alberti K, Davidson MB, DeFronzo RA. Report of the expert
committee on the diagnosis and classification of diabetes mellitus. Diabetes
care. 1997;20(7):1183.
157. Chalmers J, MacMahon S, Mancia G, Whitworth J, Beilin L, Hansson L, et al.
1999 World Health Organization-International Society of Hypertension
Guidelines for the management of hypertension. Guidelines sub-committee
of the World Health Organization. Clinical and experimental hypertension
(New York, NY: 1993). 1998;21(5-6):1009-60.
158. Barba C, Cavalli-Sforza T, Cutter J, Darnton-Hill I. Appropriate body-mass index
for Asian populations and its implications for policy and intervention
strategies. The lancet. 2004;363(9403):157.
159. WHO Expert Consultation. Waist circumference and waist-hip ratio. Report of a
WHO Expert Consultation Geneva: World Health Organization. 2008:8-11.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 147
160. Catapano AL, Graham I, De Backer G, Wiklund O, Chapman MJ, Drexel H, et al.
2016 ESC/EAS Guidelines for the Management of Dyslipidaemias.
European heart journal. 2016;37(39):2999-3058.
161. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF, Feldman HI, et al. A
new equation to estimate glomerular filtration rate. Annals of internal
medicine. 2009;150(9):604-12.
162. Barthel FM-S, Royston P, Babiker A. A menu-driven facility for complex sample
size calculation in randomized controlled trials with a survival or a binary
outcome: update. Stata J. 2005;5(1):123-9.
163. White IR, Royston P, Wood AM. Multiple imputation using chained equations:
issues and guidance for practice. Statistics in medicine. 2011;30(4):377-99.
164. Rubin DB, Schenker N. Multiple imputation in health ‐are databases
and some applications. Statistics in medicine. 1991;10(4):585-98.
165. Biering K, Hjollund NH, Frydenberg M. Using multiple imputation to deal with
missing data and attrition in longitudinal studies with repeated measures of
patient-reported outcomes. Clinical epidemiology. 2015;7:91.
166. StataCorp. Base Reference Manual: Stata Mulitple-Imputation Reference Manual.
Stata Statistical Software: Release 14. College Station, TX: StataCorp LP;
2015.
167. Van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood
pressure covariates in survival analysis. Statistics in medicine.
1999;18(6):681-94.
168. Rubin D. Multiple imputation for nonresponse in surveys. New York: John Wiley.
1987.
169. Baron RM, Kenny DA. The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations.
Journal of personality and social psychology. 1986;51(6):1173.
170. MacKinnon DP. Analysis of mediating variables in prevention and intervention
research. NIDA research monograph. 1994;139:127-.
171. Iacobucci D. Mediation analysis and categorical variables: The final frontier.
Journal of Consumer Psychology, Forthcoming. 2012.
Win Khaing References / 148
172. MacKinnon DP, Cox MC. Commentary on “Mediation analysis and categorical
variables: The final frontier” by Dawn Iacobucci. Journal of consumer
psychology: the official journal of the Society for Consumer Psychology.
2012;22(4):600.
173. Gunzler D, Chen T, Wu P, Zhang H. Introduction to mediation analysis with
structural equation modeling. Shanghai archives of psychiatry.
2013;25(6):390.
174. VanderWeele TJ. Invited commentary: structural equation models and
epidemiologic analysis. American journal of epidemiology. 2012:kws213.
175. MacKinnon DP, Warsi G, Dwyer JH. A simulation study of mediated effect
measures. Multivariate behavioral research. 1995;30(1):41-62.
176. MacKinnon DP, Lockwood CM, Hoffman JM, West SG, Sheets V. A comparison
of methods to test mediation and other intervening variable effects.
Psychological methods. 2002;7(1):83.
177. VanderWeele TJ, Vansteelandt S. Odds ratios for mediation analysis for a
dichotomous outcome. American journal of epidemiology.
2010;172(12):1339-48.
178. Sobel ME. Asymptotic confidence intervals for indirect effects in structural
equation models. Sociological methodology. 1982;13:290-312.
179. Williams J, MacKinnon DP. Resampling and distribution of the product methods
for testing indirect effects in complex models. Structural Equation
Modeling. 2008;15(1):23-51.
180. Comulada WS. Model specification and bootstrapping for multiply imputed data:
An application to count models for the frequency of alcohol use. The Stata
journal. 2015;15(3):833.
181. Demissie K, Hanley JA, Menzies D, Joseph L, Ernst P. Agreement in measuring
socio-economic status: area-based versus individual measures. Chronic
diseases and injuries in Canada. 2000;21(1):1.
182. Sin DD, Svenson LW, Man SP. Do area-based markers of poverty accurately
measure personal poverty? Canadian Journal of Public Health.
2001;92(3):184.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 149
183. Southern DA, McLaren L, Hawe P, Knudtson ML, Ghali WA, Investigators A.
Individual-level and neighborhood-level income measures: agreement and
association with outcomes in a cardiac disease cohort. Medical care.
2005;43(11):1116-22.
184. Krieger N. Women and social class: a methodological study comparing individual,
household, and census measures as predictors of black/white differences in
reproductive history. Journal of Epidemiology and Community Health.
1991;45(1):35-42.
185. Krieger N. Overcoming the absence of socioeconomic data in medical records:
validation and application of a census-based methodology. American
journal of public health. 1992;82(5):703-10.
186. Locker D, Ford J. Using Area ‐based Measures of Socioeconomic Status in Dental
Health Services Research. Journal of public health dentistry. 1996;56(2):69-
75.
187. Connelly R, Gayle V, Lambert PS. A review of educational attainment measures
for social survey research. Methodological Innovations.
2016;9:2059799116638001.
188. Ross CE, Wu C-l. The links between education and health. American sociological
review. 1995:719-45.
189. Berkman ND, Davis TC, McCormack L. Health literacy: what is it? Journal of
health communication. 2010;15(S2):9-19.
190. Wister AV, Malloy-Weir LJ, Rootman I, Desjardins R. Lifelong educational
practices and resources in enabling health literacy among older adults.
Journal of aging and health. 2010;22(6):827-54.
191. Bennett IM, Chen J, Soroui JS, White S. The contribution of health literacy to
disparities in self-rated health status and preventive health behaviors in older
adults. The Annals of Family Medicine. 2009;7(3):204-11.
192. Schillinger D, Barton LR, Karter AJ, Wang F, Adler N. Does literacy mediate the
relationship between education and health outcomes? A study of a low-
income population with diabetes. Public health reports. 2006;121(3):245-
54.
Win Khaing References / 150
193. Friis K, Lasgaard M, Rowlands G, Osborne RH, Maindal HT. Health literacy
mediates the relationship between educational attainment and health
behavior: A Danish population-based study. Journal of Health
Communication. 2016;21(sup2):54-60.
194. Rootman I, Ronson B. Literacy and health research in Canada: where have we been
and where should we go? Canadian Journal of Public Health/Revue
Canadienne de Sante'e Publique. 2005:S62-S77.
195. Kubota Y, Heiss G, MacLehose RF, Roetker NS, Folsom AR. Association of
Educational Attainment With Lifetime Risk of Cardiovascular Disease: The
Atherosclerosis Risk in Communities Study. JAMA Internal Medicine.
2017.
196. Subramanian SV, Kawachi I. Income inequality and health: what have we learned
so far? Epidemiologic reviews. 2004;26(1):78-91.
197. UNESCO. Education Counts: towards the Millennium Development Goals.
France: UNESCO; 2013.
198. UNESCO. Education transforms lives. France: UNESCO; 2013.
199. Enders CK. Analyzing longitudinal data with missing values. Rehabilitation
Psychology. 2011;56(4):267.
200. Twisk J, de Vente W. Attrition in longitudinal studies: how to deal with missing
data. Journal of clinical epidemiology. 2002;55(4):329-37.
201. Fielding S, Fayers P, Ramsay C. Predicting missing quality of life data that was
later recovered: an empirical comparison of approaches. Clinical Trials.
2010.
202. Iacobucci D, Saldanha N, Deng X. A meditation on mediation: Evidence that
structural equations models perform better than regressions. Journal of
Consumer Psychology. 2007;17(2):139-53.
203. Bollen KA, Pearl J. Eight Myths About Causality and Structural Equation Models.
In: Morgan SL, editor. Handbook of Causal Analysis for Social Research:
Springer Netherlands; 2013. p. 301.
204. Kozak M, Azevedo RA. Does using stepwise variable selection to build sequential
path analysis models make sense? Physiologia plantarum. 2011;141(3):197-
200.
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 151
205. Vansteelandt S, Bekaert M, Claeskens G. On model selection and model
misspecification in causal inference. Statistical methods in medical research.
2012;21(1):7-30.
206. Cefalu M, Dominici F, Arvold N, Parmigiani G. Model averaged double robust
estimation. Biometrics. 2017;73(2):410-21.
207. Ghosh D, Zhu Y, Coffman DL. Penalized regression procedures for variable
selection in the potential outcomes framework. Stat Med.
2015;34(10):1645-58.
208. Koch B, Vock DM, Wolfson J. Covariate selection with group lasso and doubly
robust estimation of causal effects. Biometrics. 2017.
209. Shortreed SM, Ertefaie A. Outcome-adaptive lasso: Variable selection for causal
inference. Biometrics. 2017.
210. Fiscella K, Tancredi D, Franks P. Adding socioeconomic status to Framingham
scoring to reduce disparities in coronary risk assessment. American heart
journal. 2009;157(6):988-94.
211. World Health Organization. Global action plan for the prevention and control of
noncommunicable diseases 2013-2020. 2013.
Win Khaing Appendices / 152
APPENDICES
Appendix A
Search terms and search strategy used
PubMed Search
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 153
((((((((((((((((("Cardiovascular Diseases"[Mesh])) OR ("cardiovascular events")) OR
("Myocardial Infarction"[Mesh])) OR ("Heart Failure"[Mesh])) OR ("Ventricular
Function, Left"[Mesh])) OR ("Coronary Restenosis"[Mesh])) OR (restenos*)) OR (re-
stenos*)) OR ("Coronary Disease"[Mesh])) OR ("coronary flow")) OR ("coronary
blood flow")) OR ("ejection fraction")) OR ("stroke")) OR ("cardiovascular death")) OR
("cardiovascular mortality"))) AND (((((("Education"[Mesh])) OR ("Educational
Status"[Mesh])) OR ("education level"))) OR ("Income"[Mesh]))
Scopus Search
( ( ( TITLE-ABS-KEY ( "cardiovascular disease*" ) ) OR ( TITLE-ABS-KEY (
"cardiovascular event*" ) ) OR ( TITLE-ABS-KEY ( "myocardial infarction" ) ) OR
( TITLE-ABS-KEY ( "coronary restenosis" ) ) OR ( TITLE-ABS-KEY ( restenos* ) )
OR ( TITLE-ABS-KEY ( re-stenos* ) ) OR ( TITLE-ABS-KEY ( "cardiovascular
death" ) ) OR ( TITLE-ABS-KEY ( "cardiovascular mortality" ) ) OR ( TITLE-ABS-
KEY ( "heart failure" ) ) ) OR ( ( TITLE-ABS-KEY ( "left ventricular function" ) ) OR
( TITLE-ABS-KEY ( "ejection fraction" ) ) OR ( TITLE-ABS-KEY ( "coronary flow"
) ) OR ( TITLE-ABS-KEY ( "coronary blood flow" ) ) OR ( TITLE-ABS-KEY (
"stroke" ) ) ) ) AND ( ( TITLE-ABS-KEY ( education ) ) OR ( TITLE-ABS-KEY (
income ) ) )
Win Khaing Appendices / 154
Domain Terms MEDLINE SCOPUS
1
Cardiovascular
"Cardiovascular
Diseases"[Mesh]
2,117,055 305,263
2 "cardiovascular events" 26,307 31,550
3 "Myocardial
Infarction"[Mesh]
159,476 250,482
4 "Heart Failure"[Mesh] 101,991 259,016
5 "Ventricular Function,
Left"[Mesh]
31,342 23,419
6 "Coronary Restenosis"[Mesh] 7,183 7,546
7 restenos* 23,008 33,738
8 re-stenos* 588 777
9 "Coronary Disease"[Mesh] 199,791 139,689
10 "coronary flow" 9,362 11,965
11 "coronary blood flow" 6,531 8,334
12 "ejection fraction" 52,155 85,789
13 "stroke" 252,078 348,242
14 "cardiovascular death" 4,707 6,239
15 "cardiovascular mortality" 10,646 26,591
16
1 OR 2 OR 3 OR 4 OR 5 OR
6 OR 7 OR 8 OR 9 OR 10 OR
11 OR 12 OR 13 OR 14 OR
15
2,225,476 1,189,423
17
Education
"Education"[Mesh] 652,766 1,656,179
18 "Educational Status"[Mesh] 45,577 72,983
19 "education level" 9,630 22,398
20 17 OR 18 OR 19 695,843 1,697,405
21 Income "Income"[Mesh] 56,358 267,143
22 16 AND 20 AND 21 354 1,335
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 155
Appendix B
Newcastle - Ottawa Quality Assessment Scale (Cohort studies)
Note: A study can be awarded a maximum of one star for each numbered item within
the Selection and Outcome categories. A maximum of two stars can be given for
Comparability
Selection
1) Representativeness of the exposed cohort
a. truly representative of the average in the community*
b. somewhat representative of the average in the community*
c. selected group of users e.g. nurses, volunteers
d. no description of the derivation of the cohort
2) Selection of the non-exposed cohort
a. drawn from the same community as the exposed cohort*
b. drawn from a different source*
c. no description of the derivation of the non-exposed cohort
3) Ascertainment of exposure
a. secure record (e.g. surgical records, medical records, census
registration)*
b. structured interview*
c. written self-report
d. no description
4) Demonstration that outcome of interest was not present at start of study
In the case of mortality studies, outcome of interest is still the presence of a
disease/incident, rather than death. That is to say that a statement of no history
of disease or incident earns a star.
a. yes*
b. no
Win Khaing Appendices / 156
Comparability
1) Comparability of cohorts on the basis of the design or analysis. A maximum
of 2 stars can be allotted in this category.
a. study controls for age/sex *
b. study controls for any three of the following cardiovascular risk factors:
Diabetes, BMI, Obesity, Physical activity, Hypertension, Smoking,
Alcohol drinking, Dyslipidemia and Chronic Kidney Disease *
Outcome
1. Assessment of outcome
a. independent or blind assessment stated in the paper, or confirmation of
the outcome by reference to secure records (x-rays, medical records,
etc.)*
b. record linkage (e.g. identified through ICD codes on database records)*
c. self-report (i.e. no reference to original medical records or x-rays to
confirm the outcome)
d. no description.
2. Was follow-up long enough for outcomes to occur
Minimum required follow-up period is ≥ 5 years.
a. yes*
b. no
If the follow-up period is reported with a mean and a range, and the mean is
longer than the required minimum, rate it as ‘yes.’
3. Adequacy of follow-up of cohorts
a. complete follow-up, all subjects accounted for*
b. subjects lost to follow-up are unlikely to introduce bias – small number
lost <20%
c. follow-up rate <80% and no description of those lost
d. no description or unclear
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 157
Appendix C
Commands used for multiple imputation
set more off **********Set lower and upper for interval regression*************************** **************************************************** ******* education *edu12 gen zedu12_l = edu12 gen zedu12_u = edu12 replace zedu12_l = 1 if edu12 == . replace zedu12_u = edu13 if edu12 == . & edu13!=. replace zedu12_u = edu14 if edu12 == . & edu13==. & edu14!=. replace zedu12_u = 4 if edu12 == . & edu13==. & edu14==. *edu13 gen zedu13_l = edu13 gen zedu13_u = edu13 replace zedu13_l = edu12 if edu13== . & edu12!=. replace zedu13_l = 1 if edu13== . & edu12==. replace zedu13_u = edu14 if edu13==. & edu14!=. replace zedu13_u = 4 if edu13==. & edu14==. *edu14 gen zedu14_l = edu14 gen zedu14_u = edu14 replace zedu14_l = edu13 if edu14 == . & edu13!=. replace zedu14_l = edu12 if edu14 == . & edu13==. & edu12!=. replace zedu14_l = 1 if edu14 == . & edu13==. & edu12==. replace zedu14_u = 4 if edu14 == . **************************************************** ***** Marital status *mstatus12 gen zmstatus12_l = mstatus12 gen zmstatus12_u = mstatus12 replace zmstatus12_l = 1 if mstatus12 ==. replace zmstatus12_u = 3 if mstatus12 ==. *mstatus13 gen zmstatus13_l = mstatus13 gen zmstatus13_u = mstatus13 *if mstatus2 is married or divorce, mstatus13 cannot be single replace zmstatus13_l = 2 if mstatus13 ==. & inlist(mstatus12,2,3) replace zmstatus13_u = 3 if mstatus13 ==. & inlist(mstatus12,2,3) replace zmstatus13_l = 1 if mstatus13 ==. & inlist(mstatus12,1,.) replace zmstatus13_u = 3 if mstatus13 ==. & inlist(mstatus12,1,.) *mstatus14 gen zmstatus14_l = mstatus14 gen zmstatus14_u = mstatus14 *if mstatus2/3 is married or divorce, mstatus14 cannot be single replace zmstatus14_l = 2 if mstatus14 ==. & (inlist(mstatus12,2,3) |
inlist(mstatus13,2,3)) replace zmstatus14_u = 3 if mstatus14 ==. & (inlist(mstatus12,2,3) |
inlist(mstatus13,2,3))
Win Khaing Appendices / 158
replace zmstatus14_l = 1 if mstatus14 ==. & (inlist(mstatus12,1,.) | inlist(mstatus13,1,.))
replace zmstatus14_u = 3 if mstatus14 ==. & (inlist(mstatus12,1,.) | inlist(mstatus13,1,.))
*mstatus15 gen zmstatus15_l = mstatus15 gen zmstatus15_u = mstatus15 *if mstatus2/3/4 is married or divorce, mstatus15 cannot be single replace zmstatus15_l = 2 if mstatus15 ==. & (inlist(mstatus12,2,3) |
inlist(mstatus13,2,3) | inlist(mstatus14,2,3)) replace zmstatus15_u = 3 if mstatus15 ==. & (inlist(mstatus12,2,3) |
inlist(mstatus13,2,3) | inlist(mstatus14,2,3)) replace zmstatus15_l = 1 if mstatus15 ==. & (inlist(mstatus12,1,.) |
inlist(mstatus13,1,.) | inlist(mstatus14,1,.)) replace zmstatus15_u = 3 if mstatus15 ==. & (inlist(mstatus12,1,.) |
inlist(mstatus13,1,.) | inlist(mstatus14,1,.)) **************************************************** ******** smoking **** smk12 [0=non smoker,1=quit smoker,2=current smoker] gen zsmk12_l = smk12 gen zsmk12_u = smk12 replace zsmk12_l = 0 if smk12== . replace zsmk12_u = 2 if smk12== . **** smk13 [0=non smoker,1=quit smoker,2=current smoker] gen zsmk13_l = smk13 gen zsmk13_u = smk13 *if smk12 is current/quit smoker, smk13 should not be non (should be 1curr 2ex-) replace zsmk13_l = 1 if smk13== . & inlist(smk12,1,2) replace zsmk13_u = 2 if smk13== . & inlist(smk12,1,2) *if smk12 is non/missing, smk13 can be non, ex-, curr replace zsmk13_l = 0 if smk13== . & inlist(smk12,0,.) replace zsmk13_u = 2 if smk13== . & inlist(smk12,0,.) **** smk14 [0=non smoker,1=quit smoker,2=current smoker] gen zsmk14_l = smk14 gen zsmk14_u = smk14 *if smk12, smk13 is current/quit smoker, smk14 should not be non (should be 1curr 2ex-) replace zsmk14_l = 1 if smk14== . & (inlist(smk12,1,2) | inlist(smk13,1,2)) replace zsmk14_l = 2 if smk14== . & (inlist(smk12,1,2) | inlist(smk13,1,2)) *if smk12, smk13 is non/missing, smk13 can be non, ex-, curr replace zsmk14_l = 0 if smk14== . & (inlist(smk12,0,.) | inlist(smk13,0,.)) replace zsmk14_u = 2 if smk14== . & (inlist(smk12,0,.) | inlist(smk13,0,.)) **** smk15 [0=non smoker,1=quit smoker,2=current smoker] gen zsmk15_l = smk15 gen zsmk15_u = smk15 *if smk12, smk13 is current/quit smoker, smk14 should not be non (should be 1curr 2ex-) replace zsmk15_l = 1 if smk15== . & (inlist(smk12,1,2) | inlist(smk13,1,2) |
inlist(smk14,1,2)) replace zsmk15_l = 2 if smk15== . & (inlist(smk12,1,2) | inlist(smk13,1,2) |
inlist(smk14,1,2)) *if smk12, smk13 is non/missing, smk13 can be non, ex-, curr replace zsmk15_l = 0 if smk15== . & (inlist(smk12,0,.) | inlist(smk13,0,.) |
inlist(smk14,0,.)) replace zsmk15_u = 2 if smk15== . & (inlist(smk12,0,.) | inlist(smk13,0,.) |
inlist(smk14,0,.)) **************************************************** ******** Alcohol Drinking **** alc12 [0=non ,1=quit, 2=current]
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 159
gen zalc12_l = alc12 gen zalc12_u = alc12 replace zalc12_l = 0 if alc12== . replace zalc12_u = 2 if alc12== . **** alc13 [0=non ,1=quit, 2=current] gen zalc13_l = alc13 gen zalc13_u = alc13 *if alc12 is current/quit, alc13 should not be non (should be 1curr 2ex-) replace zalc13_l = 1 if alc13== . & inlist(alc12,1,2) replace zalc13_u = 2 if alc13== . & inlist(alc12,1,2) *if alc12 is non/missing, alc13 can be non, ex-, curr replace zalc13_l = 0 if alc13== . & inlist(alc12,0,.) replace zalc13_u = 2 if alc13== . & inlist(alc12,0,.) **** alc14 [0=non ,1=quit, 2=current] gen zalc14_l = alc14 gen zalc14_u = alc14 *if alc12, alc13 is current/quit, alc14 should not be non (should be 1curr 2ex-) replace zalc14_l = 1 if alc14== . & (inlist(alc12,1,2) | inlist(alc13,1,2)) replace zalc14_l = 2 if alc14== . & (inlist(alc12,1,2) | inlist(alc13,1,2)) *if alc12, alc13 is non/missing, alc13 can be non, ex-, curr replace zalc14_l = 0 if alc14== . & (inlist(alc12,0,.) | inlist(alc13,0,.)) replace zalc14_u = 2 if alc14== . & (inlist(alc12,0,.) | inlist(alc13,0,.)) **** alc15 [0=non ,1=quit, 2=current] gen zalc15_l = alc15 gen zalc15_u = alc15 *if alc12, alc13 is current/quit, alc14 should not be non (should be 1curr 2ex-) replace zalc15_l = 1 if alc15== . & (inlist(alc12,1,2) | inlist(alc13,1,2) |
inlist(alc14,1,2)) replace zalc15_l = 2 if alc15== . & (inlist(alc12,1,2) | inlist(alc13,1,2) |
inlist(alc14,1,2)) *if alc12, alc13 is non/missing, alc13 can be non, ex-, curr replace zalc15_l = 0 if alc15== . & (inlist(alc12,0,.) | inlist(alc13,0,.) |
inlist(alc14,0,.)) replace zalc15_u = 2 if alc15== . & (inlist(alc12,0,.) | inlist(alc13,0,.) |
inlist(alc14,0,.)) **************************************************** ***** glucose gen zglu12_l = glu12 gen zglu12_u = glu12 gen zglu13_l = glu13 gen zglu13_u = glu13 gen zglu14_l = glu14 gen zglu14_u = glu14 gen zglu15_l = glu15 gen zglu15_u = glu15 ** set min/max within limit range (50 / 600) replace zglu12_l = 50 if glu12 == . replace zglu13_l = 50 if glu13 == . replace zglu14_l = 50 if glu14 == . replace zglu15_l = 50 if glu15 == . replace zglu12_u = 600 if glu12 == . replace zglu13_u = 600 if glu13 == . replace zglu14_u = 600 if glu14 == . replace zglu15_u = 600 if glu15 == . **************************************************** **** Cholesterol gen zchol12_u = chol12 gen zchol12_l = chol12 gen zchol13_u = chol13 gen zchol13_l = chol13 gen zchol14_u = chol14
Win Khaing Appendices / 160
gen zchol14_l = chol14 gen zchol15_u = chol15 gen zchol15_l = chol15 ** set min/max within limit range (70 / 600) replace zchol12_l = 70 if chol12 == . replace zchol13_l = 70 if chol13 == . replace zchol14_l = 70 if chol14 == . replace zchol15_l = 70 if chol15 == . replace zchol12_u = 600 if chol12 == . replace zchol13_u = 600 if chol13 == . replace zchol14_u = 600 if chol14 == . replace zchol15_u = 600 if chol15 == . **************************************************** **** HDL gen zhdl12_u = hdl12 gen zhdl12_l = hdl12 gen zhdl13_u = hdl13 gen zhdl13_l = hdl13 gen zhdl14_u = hdl14 gen zhdl14_l = hdl14 gen zhdl15_u = hdl15 gen zhdl15_l = hdl15 ** set min/max within limit range range (7 / 185) replace zhdl12_l = 15 if hdl12 == . replace zhdl13_l = 15 if hdl13 == . replace zhdl14_l = 15 if hdl14 == . replace zhdl15_l = 15 if hdl15 == . replace zhdl12_u = 180 if hdl12 == . replace zhdl13_u = 180 if hdl13 == . replace zhdl14_u = 180 if hdl14 == . replace zhdl15_u = 180 if hdl15 == . **************************************************** ***** tri (Triglyceride) gen ztri12_u = tri12 gen ztri12_l = tri12 gen ztri13_u = tri13 gen ztri13_l = tri13 gen ztri14_u = tri14 gen ztri14_l = tri14 gen ztri15_u = tri15 gen ztri15_l = tri15 ** set min/max within limit range range (15 / 2100) replace ztri12_l = 15 if tri12 == . replace ztri13_l = 15 if tri13 == . replace ztri14_l = 15 if tri14 == . replace ztri15_l = 15 if tri15 == . replace ztri12_u = 2100 if tri12 == . replace ztri13_u = 2100 if tri13 == . replace ztri14_u = 2100 if tri14 == . replace ztri15_u = 2100 if tri15 == . **************************************************** **** LDL gen zldl12_u = ldl12 gen zldl12_l = ldl12 gen zldl13_u = ldl13 gen zldl13_l = ldl13 gen zldl14_u = ldl14 gen zldl14_l = ldl14 gen zldl15_u = ldl15 gen zldl15_l = ldl15 ** set min/max within limit range (20 / 500)
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 161
replace zldl12_l = 20 if ldl12 == . replace zldl13_l = 20 if ldl13 == . replace zldl14_l = 20 if ldl14 == . replace zldl15_l = 20 if ldl15 == . replace zldl12_u = 500 if ldl12 == . replace zldl13_u = 500 if ldl13 == . replace zldl14_u = 500 if ldl14 == . replace zldl15_u = 500 if ldl15 == . **************************************************** **** Uric acid gen zuric12_u = uric12 gen zuric12_l = uric12 gen zuric13_u = uric13 gen zuric13_l = uric13 gen zuric14_u = uric14 gen zuric14_l = uric14 gen zuric15_u = uric15 gen zuric15_l = uric15 ** set min/max within limit range (1.1 / 14) replace zuric12_l = 1.1 if uric12 == . replace zuric13_l = 1.1 if uric13 == . replace zuric14_l = 1.1 if uric14 == . replace zuric15_l = 1.1 if uric15 == . replace zuric12_u = 14 if uric12 == . replace zuric13_u = 14 if uric13 == . replace zuric14_l = 1.1 if uric14 == . replace zuric15_l = 1.1 if uric15 == . **************************************************** **** Creatinine gen zcre12_u = cre12 gen zcre12_l = cre12 gen zcre13_u = cre13 gen zcre13_l = cre13 gen zcre14_u = cre14 gen zcre14_l = cre14 gen zcre15_u = cre15 gen zcre15_l = cre15 ** set min/max within limit range (0.2 / 15) replace zcre12_l = 0.2 if cre12 == . replace zcre13_l = 0.2 if cre13 == . replace zcre14_l = 0.2 if cre14 == . replace zcre15_l = 0.2 if cre15 == . replace zcre12_u = 15 if cre12 == . replace zcre13_u = 15 if cre13 == . replace zcre14_u = 15 if cre14 == . replace zcre15_u = 15 if cre15 == . ********************* START IMPUTATION ******************* display "Start imputations ""$S_TIME $S_DATE" set matsize 11000 mi set mlong mi register imputed edu* inc* mstatus* smk* alc* exe* weight* height waist* hip* sbp*
dbp* drfat_sum* glu* chol* hdl* tri* ldl* uric* cre* mi register regular sex age* fm_fat fm_dm fm_ht event_composit1 d_composit1 drnsaid* mi impute chained /// (intreg,ll(zedu12_l) ul(zedu12_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk*
xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xedu12 ///
(intreg,ll(zedu13_l) ul(zedu13_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xedu13 ///
(intreg,ll(zedu14_l) ul(zedu14_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xedu14 ///
Win Khaing Appendices / 162
(ologit,iterate(100) augment include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) inc12 ///
(ologit,iterate(100) augment include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) inc13 ///
(ologit,iterate(100) augment include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) inc14 ///
(ologit,iterate(100) augment include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) inc15 ///
(intreg,ll(zmstatus12_l) ul(zmstatus12_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* event_composit1 d_composit1)) xmstatus12 ///
(intreg,ll(zmstatus13_l) ul(zmstatus13_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* event_composit1 d_composit1)) xmstatus13 ///
(intreg,ll(zmstatus14_l) ul(zmstatus14_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* event_composit1 d_composit1)) xmstatus14 ///
(intreg,ll(zmstatus15_l) ul(zmstatus15_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* event_composit1 d_composit1)) xmstatus15 ///
(intreg,ll(zsmk12_l) ul(zsmk12_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xsmk12 ///
(intreg,ll(zsmk13_l) ul(zsmk13_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xsmk13 ///
(intreg,ll(zsmk14_l) ul(zsmk14_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xsmk14 ///
(intreg,ll(zsmk15_l) ul(zsmk15_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xsmk15 ///
(intreg,ll(zalc12_l) ul(zalc12_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xalc12 ///
(intreg,ll(zalc13_l) ul(zalc13_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xalc13 ///
(intreg,ll(zalc14_l) ul(zalc14_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xalc14 ///
(intreg,ll(zalc15_l) ul(zalc15_u) include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) xalc15 ///
(mlogit, iterate(100) augment include (i.sex age* xedu* i.inc* xmstatus* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* event_composit1 d_composit1)) exe* ///
(regress, include (i.sex age* height weight* waist* hip* event_composit1 d_composit1)) height ///
(regress, include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xcre* event_composit1 d_composit1)) weight* ///
(regress, include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xcre* event_composit1 d_composit1)) waist* ///
(regress, include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xcre* event_composit1 d_composit1)) hip* ///
(regress, include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.fm_ht xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) sbp* ///
(regress, include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.fm_ht xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) dbp* ///
(logit, iterate(100) augment include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* i.fm_fat xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) drfat_sum* ///
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 163
(intreg,ll(zglu12_l) ul(zglu12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_dm event_composit1 d_composit1)) xglu12 ///
(intreg,ll(zglu13_l) ul(zglu13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_dm event_composit1 d_composit1)) xglu13 ///
(intreg,ll(zglu14_l) ul(zglu14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_dm event_composit1 d_composit1)) xglu14 ///
(intreg,ll(zglu15_l) ul(zglu15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_dm event_composit1 d_composit1)) xglu15 ///
(intreg,ll(zchol12_l) ul(zchol12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xchol12 ///
(intreg,ll(zchol13_l) ul(zchol13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xchol13 ///
(intreg,ll(zchol14_l) ul(zchol14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xchol14 ///
(intreg,ll(zchol15_l) ul(zchol15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xchol15 ///
(intreg,ll(zhdl12_l) ul(zhdl12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xhdl12 ///
(intreg,ll(zhdl13_l) ul(zhdl13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xhdl13 ///
(intreg,ll(zhdl14_l) ul(zhdl14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xhdl14 ///
(intreg,ll(zhdl15_l) ul(zhdl15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xhdl15 ///
(intreg,ll(ztri12_l) ul(ztri12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xtri12 ///
(intreg,ll(ztri13_l) ul(ztri13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xtri13 ///
(intreg,ll(ztri14_l) ul(ztri14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xtri14 ///
(intreg,ll(ztri15_l) ul(ztri15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xchol* xhdl* xtri* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xtri15 ///
(intreg,ll(zldl12_l) ul(zldl12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xldl* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xldl12 ///
(intreg,ll(zldl13_l) ul(zldl13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xldl* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xldl13 ///
(intreg,ll(zldl14_l) ul(zldl14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xldl* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xldl14 ///
(intreg,ll(zldl15_l) ul(zldl15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drfat_sum* xglu* xldl* xuric* xcre* i.fm_fat event_composit1 d_composit1)) xldl15 ///
(intreg,ll(zuric12_l) ul(zuric12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xuric12 ///
(intreg,ll(zuric13_l) ul(zuric13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xuric13 ///
(intreg,ll(zuric14_l) ul(zuric14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xuric14 ///
(intreg,ll(zuric15_l) ul(zuric15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xuric15 ///
Win Khaing Appendices / 164
(intreg,ll(zcre12_l) ul(zcre12_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xcre12 ///
(intreg,ll(zcre13_l) ul(zcre13_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xcre13 ///
(intreg,ll(zcre14_l) ul(zcre14_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xcre14 ///
(intreg,ll(zcre15_l) ul(zcre15_u) include (i.sex age* xedu* i.inc* xsmk* xalc* i.exe* height weight* waist* hip* sbp* i.drnsaid* xglu* xchol* xhdl* xtri* xuric* xcre* event_composit1 d_composit1)) xcre15 ///
,add(70) noimpute force rseed(1234) display "End imputations ""$S_TIME $S_DATE"
b) Diagnostic plots: *** comparing distribution of the imputed values with the observed values midiagplots height, sample(all) combine ksmirnov
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 165
Appendix D
Commands used for mediation analysis
program define eduincmed, rclass tempname a11 tempname a12 tempname a21 tempname a22 tempname b1 tempname b2 tempname c_prime1 tempname c_prime2 mi estimate, cmdok: gsem (outcomecve <- ib(3).iinc ib(3).iedugr3 i.ht i.dyslipid1
i.dm, logit) (ib(3).iinc <- ib(3).iedugr3 age i.imstatus, mlogit), vce(cluster empnm)
scalar `a11' = el(e(b_mi),1,14) scalar `a12' = el(e(b_mi),1,22) scalar `a21' = el(e(b_mi),1,15) scalar `a22' = el(e(b_mi),1,23) scalar `b1' = el(e(b_mi),1,1) scalar `b2' = el(e(b_mi),1,2) scalar `c_prime1' = el(e(b_mi),1,4) scalar `c_prime2' = el(e(b_mi),1,5) return scalar edu1inc1_ind = `a11'*`b1' return scalar edu1inc1_d = `c_prime1' return scalar edu1inc1_total= (`a11'*`b1') + `c_prime1' return scalar edu1inc1_ind2total = (`a11'*`b1')/((`a11'*`b1') + `c_prime1') return scalar edu1inc1_ind2dir = (`a11'*`b1')/`c_prime1' return scalar edu1inc1_dir2total = `c_prime1'/((`a11'*`b1') + `c_prime1') return scalar edu1inc2_ind = `a12'*`b2' return scalar edu1inc2_d = `c_prime1' return scalar edu1inc2_total= (`a12'*`b2') + `c_prime1' return scalar edu1inc2_ind2total = (`a12'*`b2')/((`a12'*`b2') + `c_prime1') return scalar edu1inc2_ind2dir = (`a12'*`b2')/`c_prime1' return scalar edu1inc2_dir2total = `c_prime1'/((`a12'*`b2') + `c_prime1') return scalar edu2inc1_ind = `a21'*`b1' return scalar edu2inc1_d = `c_prime2' return scalar edu2inc1_total= (`a21'*`b1') + `c_prime2' return scalar edu2inc1_ind2total = (`a21'*`b1')/((`a21'*`b1') + `c_prime2') return scalar edu2inc1_ind2dir = (`a21'*`b1')/`c_prime2' return scalar edu2inc1_dir2total = `c_prime2'/((`a21'*`b1') + `c_prime2') return scalar edu2inc2_ind = `a22'*`b2' return scalar edu2inc2_d = `c_prime2' return scalar edu2inc2_total= (`a22'*`b2') + `c_prime2' return scalar edu2inc2_ind2total = (`a22'*`b2')/((`a22'*`b2') + `c_prime2') return scalar edu2inc2_ind2dir = (`a22'*`b2')/`c_prime2' return scalar edu2inc2_dir2total = `c_prime2'/((`a22'*`b2') + `c_prime2') return scalar or_edu1inc1_ind = exp(`a11'*`b1') return scalar or_edu1inc2_ind = exp(`a12'*`b2') return scalar or_edu2inc1_ind = exp(`a21'*`b1') return scalar or_edu2inc2_ind = exp(`a22'*`b2') return scalar or_edu1cve_d = exp(`c_prime1') return scalar or_edu2cve_d = exp(`c_prime2') end eduincmed set seed 12345 bootstrap e1i1_ind=r(edu1inc1_ind) e1i1_d=r(edu1inc1_d) e1i1_total=r(edu1inc1_total)
e1i1_ind2total=r(edu1inc1_ind2total) e1i1_ind2dir=r(edu1inc1_ind2dir) e1i1_dir2total=r(edu1inc1_dir2total) ///
Win Khaing Appendices / 166
e1i2_ind=r(edu1inc2_ind) e1i2_d=r(edu1inc2_d) e1i2_total=r(edu1inc2_total) e1i2_ind2total=r(edu1inc2_ind2total) e1i2_ind2dir=r(edu1inc2_ind2dir) e1i2_dir2total=r(edu1inc2_dir2total) /// e2i1_ind=r(edu2inc1_ind) e2i1_d=r(edu2inc1_d) e2i1_total=r(edu2inc1_total) e2i1_ind2total=r(edu2inc1_ind2total) e2i1_ind2dir=r(edu2inc1_ind2dir) e2i1_dir2total=r(edu2inc1_dir2total) /// e2i2_ind=r(edu2inc2_ind) e2i2_d=r(edu2inc2_d) e2i2_total=r(edu2inc2_total) e2i2_ind2total=r(edu2inc2_ind2total) e2i2_ind2dir=r(edu2inc2_ind2dir) e2i2_dir2total=r(edu2inc2_dir2total) /// or_e1i1_ind=r(or_edu1inc1_ind) or_e1i2_ind=r(or_edu1inc2_ind) or_e2i1_ind=r(or_edu2inc1_ind) or_e212_ind=r(or_edu2inc2_ind) or_e1cve_d=r(or_edu1cve_d) or_e2cve_d=r(or_edu2cve_d) /// , reps(1000) cluster(empnm) : eduincmed
estat bootstrap, bc p norm
program define coefmed, rclass tempname 11 tempname 12 tempname 14 tempname 15 tempname 18 tempname 110 tempname 112 tempname 114 tempname 115 tempname 117 tempname 119 tempname 120 tempname 122 tempname 123 tempname 125 tempname 127 tempname 128 mi estimate, cmdok: gsem (outcomecve <- ib(3).iinc ib(3).iedugr3 i.ht i.dyslipid1
i.dm, logit) (ib(3).iinc <- ib(3).iedugr3 age i.imstatus, mlogit), vce(cluster empnm)
scalar `11' = el(e(b_mi),1,1) scalar `12' = el(e(b_mi),1,2) scalar `14' = el(e(b_mi),1,4) scalar `15' = el(e(b_mi),1,5) scalar `18' = el(e(b_mi),1,8) scalar `110' = el(e(b_mi),1,10) scalar `112' = el(e(b_mi),1,12) scalar `114' = el(e(b_mi),1,14) scalar `115' = el(e(b_mi),1,15) scalar `117' = el(e(b_mi),1,17) scalar `119' = el(e(b_mi),1,19) scalar `120' = el(e(b_mi),1,20) scalar `122' = el(e(b_mi),1,22) scalar `123' = el(e(b_mi),1,23) scalar `125' = el(e(b_mi),1,25) scalar `127' = el(e(b_mi),1,27) scalar `128' = el(e(b_mi),1,28) return scalar inc1cve = `11' return scalar inc2cve = `12' return scalar edu1cve = `14' return scalar edu2cve = `15' return scalar htcve = `18' return scalar dyscve = `110' return scalar dmcve = `112' return scalar edu1inc1 = `114' return scalar edu2inc1 = `115' return scalar ageinc1 = `117' return scalar mstat2inc1 = `119' return scalar mstat3inc1 = `120' return scalar edu1inc2 = `122' return scalar edu2inc2 = `123' return scalar ageinc2 = `125' return scalar mstat2inc2 = `127' return scalar mstat3inc2 = `128' end
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 167
coefmed set seed 12345 bootstrap inc1cve=r(inc1cve) inc2cve=r(inc2cve) edu1cve=r(edu1cve) edu2cve=r(edu2cve)
htcve=r(htcve) dyscve=r(dyscve) dmcve=r(dmcve) edu1inc1=r(edu1inc1) edu2inc1=r(edu2inc1) ageinc1=r(ageinc1) /// mstat2inc1=r(mstat2inc1) mstat3inc1=r(mstat3inc1) edu1inc2=r(edu1inc2) edu2inc2=r(edu2inc2) ageinc2=r(ageinc2) mstat2inc2=r(mstat2inc2) mstat3inc2=r(mstat3inc2) , reps(1000) cluster(empnm) : coefmed
estat bootstrap, bc p norm
Win Khaing Appendices / 168
Appendix E
Diagnostics plot between missing and observed values
(Height variable)
Fac. of Grad. Studies, Mahidol Univ. Ph.D.(Clinical Epidemiology) / 169
Appendix F
Ethical Approval
Win Khaing Biography / 170
BIOGRAPHY
NAME Win Khaing
DATE OF BIRTH 22 June 1977
PLACE OF BIRTH Mandalay, Myanmar
INSTITUTIONS ATTENDED University of Medicine, Mandalay,
(1995 – 2002),
Bachelor of Medicine and Bachelor of Surgery
University of Medicine, Mandalay,
(2007 – 2008),
Master of Medical Science (Public Health)
SCHOLARSHIP RECEIVED Norwegian Scholarship Capacity Building for
Institutes in Myanmar
RESEARCH GRANTS Norwegian Scholarship Capacity Building for
Institutes in Myanmar
HOME ADDRESS No. (187), 77th Street, Between 32nd and 33rd
Streets, Chan-Aye-Thar-San Township
Mandalay, MYANMAR
EMPLOYMENT ADDRESS Department of Preventive and Social Medicine,
Mandalay, University of Medicine, Mandalay, 30th
Street, Between 73rd and 74th Streets, Mandalay,
MYANMAR
PHONE Mobile: (959) 2026240
E-MAIL [email protected]