THE MEDIATION ANALYSIS OF EDUCATION AND INCOME ON … · the mediation analysis of education and income on major cardiovascular events . win khaing. a thesis submitted in partial

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


............................................................. 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


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


2.7 Pooled education and income effect on strokes (subgroup analyses) 50

2.8 Pooled education and income effect on cardiovascular deaths


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


among medium versus high income levels 61


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.


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


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


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


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


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",


"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


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


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


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.


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.


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


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


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


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,


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.


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


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)


1.48 (1.24, 1.76)


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


1.28 b (0.67, 2.35)

W

in Khaing

Literature R

eview / 22

Table 2.1 Characteristics of included studies (continued)

Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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)


1.35 b, c (1.10, 1.64)


1.39 c (1.04, 1.86) 14 39.2 53.8 100 12.3 NA 25.6


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


1.40 c (1.10, 1.77)


1.24 c (1.05, 1.46)


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


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)


1.45 c (1.30, 1.63)


Fac. of G


niv.

Ph.D.(C


Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


0.67 c (0.60, 0.75)


0.83 c (0.77, 0.90)


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


0.61 c (0.49, 0.76)


1.81 c (1.39, 2.36)


2.98 c (2.17, 4.10)

MI HR Education (Medium vs High)

0.25 c (0.18, 0.35)


0.64 c (0.45, 0.91)


2.39 c (1.55, 3.67)


4.06 c (2.36, 6.97)

Lammintausta31 2012 Finland, Europe MI RR Income


(High) Income (Low vs High)

2.82 c (2.56, 3.10)


W

in Khaing

Literature R

eview / 24

Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


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


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


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


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


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


1.22 (0.91, 1.62)

Fac. of G


niv.

Ph.D.(C



Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


1.71 b, c (1.41, 2.07)

Huisman102 2008 Netherlands,

Europe (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


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


1.42 (0.83, 2.45)

Netherlands, Europe Education



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)


1.65 (1.38, 1.97)

W

in Khaing

Literature R

eview / 26


Author Year


Outcome Risk

measure



NC Mean Age

(years)

Male (%)

DM

(%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


1.90 (1.30, 2.80)

Lapidus & Bengtsson107 1986

Sweden, Europe (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


0.87 (0.71, 1.07)


1.15 (0.97, 1.36) 25 NA 73.2 9.5 46.3 NA NA


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


1.31 (0.63, 2.74)

Fac. of G


niv.

Ph.D.(C



Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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)


0.96 b (0.95, 0.97) 2 60.7 71.1 9.5 NA NA NA


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


1.24 (0.92, 1.68)


1.30 (0.94, 1.79)


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)


1.05 b (0.98, 1.12)


1.18 b (0.81, 1.70)

W

in Khaing

Literature R

eview / 28


Author Year Country, Setting

(income level) Outcome Risk

measure Study factors (Categories)


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


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


1.13 b, c (1.08, 1.17) 0 52.4 45.3 2.87 63.4 6.25 25.15


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


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


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


Fac. of G


niv.

Ph.D.(C


Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


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


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


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


1.63 (1.14, 2.33)



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


Author Year





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


0.96 b, c (0.86, 1.08)


1.00 b, c (0.99, 1.01)


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


1.82 b, c (1.33, 2.50)


1.29 c (0.88, 1.69) 14 39.2 53.8 100 12.3 NA 25.6


2.09 c (1.62, 2.69)

Li126 2008 Sweden, Europe Stroke RR Income



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


4.79 b (1.48, 15.5)


1.42 b (1.04, 1.96)


1.75 b (0.81, 3.85)


Fac. of G


niv.

Ph.D.(C


Author Year





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


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


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


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


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


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


1.10 b (0.96, 1.28)


W

in Khaing

Literature R

eview / 32





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


2.07 a (1.68, 2.55)


1.29 a (1.16, 1.42) 0 65.5 54.4 27.0 27.6 45.0 NA


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


1.17 b (1.15, 1.22)


1.51 b (1.49, 1.52)


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


0.73 (0.51, 1.04)


0.74 (0.44, 1.26)


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)


Fac. of G


niv.

Ph.D.(C






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


1.36 (1.13, 1.64)


1.72 (1.45, 2.05)

Zhou133 2006 China, Asia (High)


(Medium vs High) 1.02 (0.57, 1.83) 9 77.2 54.8 26.5 27.7 NA NA


0.88 (0.46, 1.68)


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)


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


0.74 (0.52, 1.04)


1.05 (0.77, 1.44) 25 NA 73.2 9.5 47.3 NA NA


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


W

in Khaing

Literature R

eview / 34

Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


1.41 (1.31, 1.50)


1.38 (1.27, 1.51)


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


(Medium vs High) 1.36 c (1.18, 1.57)

Pednekar136 2011 India, Asia

(Lower-


(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)


(Low vs High) 2.31 b, c (1.98, 2.68)

Rawshani92 2015 Sweden, Europe


(Medium vs High) 1.54 c (1.41, 1.68) 11 39.3 53.8 100 11.81 NA 25.61


1.27 c (0.94, 1.70)


1.92 c (1.31, 2.81) 14 39.2 53.8 NA 12.25 NA 25.6


3.40 c (2.64, 4.37)


Fac. of G


niv.

Ph.D.(C


Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


1.08 c (0.99, 1.17)

Gallo138 2012 Europe (High)


(Medium vs High) 1.21 b, c (1.20, 1.23) 7 52.0 35.9 NA 27.3 NA NA


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)


1.87 b, c (1.60, 2.19)


(Medium vs High) 1.16 b, c (1.12, 1.19)


1.43 b, c (1.12, 1.83)

Rasmussen44 2006 Demark, Europe


(Medium vs High) 1.10 c (1.02, 1.19) 6 61.0 70.9 3.6 NA NA NA


1.15 c (1.06, 1.25)


1.14 c (1.08, 1.20) 6 61 60.2 3.6 NA NA NA


1.42 c (1.35, 1.50)

Salomaa30 2000 Finland, Europe


(Low vs High) 1.96 c (1.84, 2.08) 2 NA NA NA NA NA NA


2.61 c (2.47, 2.76)


W

in Khaing

Literature R

eview / 36

Author Year





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


2.07 (0.94, 4.57)

Death due to stroke RR Education

(Medium vs High) 1.27 (0.97, 1.66)


1.08 (0.56, 2.08)

Tonne140 2005 US (High)


(Medium vs High) 1.21 (1.05, 1.39) 13 69 58.1 31 NA 63.5 NA


1.32 (1.15, 1.52)


1.25 (1.04, 1.52)


1.38 (1.14, 1.67)

Chen39 2015 China, Asia (High)


(Low vs High) 1.88 (1.05, 3.36) 14 73.4 46.1 10.2 22.6 72.5 NA


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


1.85 (1.43, 2.43)


2.83 (2.16, 3.82)


Fac. of G


niv.

Ph.D.(C






NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2) (d) Cardiovascular deaths

Ito141 2008 Japan, Asia (High)


(Medium vs High) 1.15 c (0.77, 1.70) 14 NA 48.3 NA 28.8 NA NA


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


3.6 (1.99, 6.60)


(Medium vs High) 2.49 (1.40, 4.44)


4.08 (2.31, 7.21)

Chicago, WEPG, US


(Medium vs High) 1.00 (0.63, 1.59) 1 48.6 100 NA 70.2 NA NA


1.62 (1.08, 2.44)


(Medium vs High) 0.97 (0.62, 1.42)


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)


W

in Khaing

Literature R

eview / 38





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


(Medium vs High) 1.60 (0.90, 2.84)

Bosma104 1995 Lithuania, Europe




1.08 (0.62, 1.88)


(Medium vs High) 1.12 (0.67, 1.86)


1.16 (0.72, 1.88)

Netherlands, Europe




1.06 (0.46, 2.43)


(Medium vs High) 1.56 (0.88, 2.77)


1.40 (0.76, 2.58)




(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


1.33 (0.93, 1.92)


Fac. of G


niv.

Ph.D.(C


Author Year





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


Rawshani144 2016 Sweden, Europe


(Medium vs High) 1.08 b (1.04, 1.10) 16 58.3 60.4 100 17.2 NA 30.3


1.19 b (1.10, 1.28)


1.54 (1.42, 1.68)


1.87 (1.76, 1.99)

Lammintausta31 2012 Finland, Europe

Death due to MI RR Income



4.78 c (4.13, 5.54)

Li126 2008 Sweden, Europe

Death due to stroke RR Income



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


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)


W

in Khaing

Literature R

eview / 40





NC Mean Age

(years)

Male (%)

DM (%)

SMK (%)

HT (%)

Mean BMI

(kg/m2)


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


0.86 (0.56, 1.32)


1.64 (1.23, 2.17) 0 65.5 54.3

5 27.0

3 27.6 44.95 NA


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


niv.

Ph.D.(C


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



of cohort



Outcome of interest




up


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


niv.

Ph.D.(C



Authors Year



of cohort



Outcome of interest




up


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


Authors Year



of cohort



Outcome of interest




up


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


Fac. of G


niv.

Ph.D.(C


Authors Year



of cohort



Outcome of interest




up


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


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


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)


Education Income


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;


Table 2.6 Pooled education and income effect on cardiovascular events (subgroup analyses)

Education Income


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



Table 2.7 Pooled education and income effect on strokes (subgroup analyses)

Education Income


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



Table 2.8 Pooled education and income effect on cardiovascular deaths (subgroup analyses)

Education Income


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



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


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


Figure 2.7 Pooling effects of income on cardiovascular outcomes

Fac. of G


niv.

Ph.D.(C


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


niv.

Ph.D.(C


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


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.


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


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,


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.


• 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


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


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,


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


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


• 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


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,


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


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)


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


� = 𝑎𝑎012+ 𝑎𝑎12𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝜀𝜀𝑚𝑚 (5)

For path a21



For path a22



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


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)


� = 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


(c) Medium versus high education and low income


(d) Medium versus high education and medium income


The total effects (TE) were estimated as follows:


(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


(c) Total effect of medium education on MCVE through low

income


(d) Total effect of medium education on MCVE through

medium income


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


× 100

(c) Percent total effect of medium education mediated through

low income


× 100

(d) Percent total effect of medium education mediated through

medium income


× 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


95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑒𝑒𝑒𝑒𝑒𝑒1𝑖𝑖𝑖𝑖𝑖𝑖1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎11𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1)�

𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11𝑏𝑏1) = 𝑎𝑎112 𝑣𝑣𝑎𝑎𝑣𝑣(𝑏𝑏1) + 𝑏𝑏12𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎11)

(b) Low versus high education and medium income




(c) Medium versus high education and low income




(d) Medium versus high education and medium income




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


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


� = 𝑎𝑎012+ 𝑎𝑎12𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝜀𝜀𝑚𝑚 (11)

For path a21



For path a22



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).



� = 𝑏𝑏01 + 𝑏𝑏1𝑒𝑒𝑒𝑒𝑒𝑒1 + 𝑐𝑐′1𝑖𝑖𝑖𝑖𝑐𝑐1 + 𝜀𝜀𝑦𝑦 (14)


� = 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


(c) Medium versus high income and low education


(d) Medium versus high income and medium education


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


(c) Total effect of medium income on MCVE through low

education


(d) Total effect of medium income on MCVE through medium

education


The percent total effects mediated was then estimated as

follows:


(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


× 100

(c) Percent total effect of medium income mediated through low

education


× 100

(d) Percent total effect of medium income mediated through

medium education


× 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)�


(b) Low versus high income and medium education




(c) Medium versus high income and low education


95%𝐶𝐶𝐶𝐶 𝑂𝑂𝑂𝑂𝑀𝑀𝑀𝑀𝑖𝑖𝑖𝑖𝑖𝑖2𝑒𝑒𝑒𝑒𝑒𝑒1 = 𝑒𝑒𝑒𝑒𝑒𝑒�𝑎𝑎21𝑏𝑏1±𝑧𝑧𝛼𝛼 2⁄ �𝑣𝑣𝑎𝑎𝑣𝑣(𝑎𝑎21𝑏𝑏1)�


(d) Medium versus high income and medium education





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.


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.


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


Participant A Male Male Male Male Male

Participant B Female Female Female Female Loss

For Height


Participant A 160 160 16 (outlier) 160 160

Participant B 145 115 (outlier) 145 Death Death

Participant C 150 150 - (missing) 151 Loss


Participant A 160 160 160 160 160

Participant B 145 145 145 Death Death

Participant C 150 150 150 151 Loss


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


Participant A Single Married Divorce Single Single

Participant B Married Single Married - (missing) Divorce

Participant C Separate Married Single Divorce Single


Participant A Single Married Divorce Divorce Divorce

Participant B Married Married Married - (missing) Divorce

Participant C Separate Married Married Divorce Divorce


Table 3.2 Example of 3 subjects who had inconsistent data for education, marital status,

smoking status and alcohol consumption (Continued)

For smoking status


Participant A None Current Ex- None None

Participant B Current None Current - (missing) Ex-

Participant C Ex- Current None Ex- None


Participant A None Current Ex- Ex- Ex-

Participant B Current Current Current - (missing) Ex-

Participant C Ex- Current Current Ex- Ex-

For alcohol consumption


Participant A None Current Ex- None None

Participant B Current None Current - (missing) Ex-

Participant C Ex- Current None Ex- None


Participant A None Current Ex- Ex- Ex-

Participant B Current Current Current - (missing) Ex-

Participant C Ex- Current Current Ex- Ex-

Fac. of G


niv.

Ph.D.(C


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


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


Subjects in EGAT1/2n = 2,960

Baseline data

n = 2,960Lab datan = 2,728


Baseline data

n = 2,324Lab datan = 2,331


Baseline data

n = 1,958Lab datan = 1,885


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


Figure 3.3 A causal effect pathway of education → income → MCVE

Income (M) b

Cardiovascular Events (Y)

Education (X) c

A

B

a


e2

e3

Education (X) c'


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


Figure 3.5 A causal effect pathway of income → education → MCVE

Education (M) b


Income (X) c

A

B

a


e2

e3

Income (X) c'


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).


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


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.


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.


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.


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


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


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


Table 4.2 Report of number of missing data (continued)

Variables EGAT

period Observed

%


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


Table 4.2 Report of number of missing data (continued)

Variables EGAT

period Observed

%


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


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


Table 4.3 Comparison of characteristics of subjects between original dataset and

imputed dataset (continued)


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


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


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;


Table 4.5 Mediation analysis of education and income (Multivariate Analysis) E

quat

ions


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;


Table 4.6 Mediation analysis of MCVE (Univariate Analysis)


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;


Table 4.7 Mediation analysis of MCVE (Multivariate Analysis)


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;


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;


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;


Table 4.10 Causal effects of low education on MCVE through medium income (bias-



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 direct effect 7.17 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduL, Low education; IncM, Medium

income;


Table 4.11 Causal effects of medium education on MCVE through low income (bias-



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 direct effect 16.53 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduM, Medium education; IncL,

Low income;


Table 4.12 Causal effects of medium education on MCVE through medium income

(bias-corrected bootstrapped)


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 direct effect 39.36 b, coefficient; SE, standard error; t, t-test; CI, confidence interval; EduM, Medium education; IncM,

Medium income;


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

;


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.


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


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.


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.


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


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


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.


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


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


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


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


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.


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.


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.


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.


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.

http://www.who.int/social_determinants/sdh_definition/en/


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

http://fxtop.com/en/currency-converter-past.php

https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups

https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups


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.


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.


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


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.


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


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


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.


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.


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


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.


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.


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.


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.


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.


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


((((((((((((((((("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 ) ) )


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


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


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


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))


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]


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


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)


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 ///


(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 ///




(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(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(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 ///




(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* ///


(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(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(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(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(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(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(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 ///




,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


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 à11' = el(e(b_mi),1,14) scalar à12' = el(e(b_mi),1,22) scalar à21' = el(e(b_mi),1,15) scalar à22' = 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 = à11'*`b1' return scalar edu1inc1_d = `c_prime1' return scalar edu1inc1_total= (à11'*`b1') + `c_prime1' return scalar edu1inc1_ind2total = (à11'*`b1')/((à11'*`b1') + `c_prime1') return scalar edu1inc1_ind2dir = (à11'*`b1')/`c_prime1' return scalar edu1inc1_dir2total = `c_prime1'/((à11'*`b1') + `c_prime1') return scalar edu1inc2_ind = à12'*`b2' return scalar edu1inc2_d = `c_prime1' return scalar edu1inc2_total= (à12'*`b2') + `c_prime1' return scalar edu1inc2_ind2total = (à12'*`b2')/((à12'*`b2') + `c_prime1') return scalar edu1inc2_ind2dir = (à12'*`b2')/`c_prime1' return scalar edu1inc2_dir2total = `c_prime1'/((à12'*`b2') + `c_prime1') return scalar edu2inc1_ind = à21'*`b1' return scalar edu2inc1_d = `c_prime2' return scalar edu2inc1_total= (à21'*`b1') + `c_prime2' return scalar edu2inc1_ind2total = (à21'*`b1')/((à21'*`b1') + `c_prime2') return scalar edu2inc1_ind2dir = (à21'*`b1')/`c_prime2' return scalar edu2inc1_dir2total = `c_prime2'/((à21'*`b1') + `c_prime2') return scalar edu2inc2_ind = à22'*`b2' return scalar edu2inc2_d = `c_prime2' return scalar edu2inc2_total= (à22'*`b2') + `c_prime2' return scalar edu2inc2_ind2total = (à22'*`b2')/((à22'*`b2') + `c_prime2') return scalar edu2inc2_ind2dir = (à22'*`b2')/`c_prime2' return scalar edu2inc2_dir2total = `c_prime2'/((à22'*`b2') + `c_prime2') return scalar or_edu1inc1_ind = exp(à11'*`b1') return scalar or_edu1inc2_ind = exp(à12'*`b2') return scalar or_edu2inc1_ind = exp(à21'*`b1') return scalar or_edu2inc2_ind = exp(à22'*`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) ///


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


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


Appendix E

Diagnostics plot between missing and observed values

(Height variable)


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]