Birth weight ethnic differences and health outcomes in childhood

Differences in infant mortality between ethnic groups and the “low birth weight paradox”.. Main findings i The effect of birth weight on infant mortality has shown the existence of the “

BIRTH WEIGHT: ETHNIC DIFFERENCES AND HEALTH OUTCOMES IN CHILDHOOD

JEANNETTE LEE MBBS, BMed Sci, FRACGP

A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF MEDICINE DEPARTMENT OF COMMUNITY, OCCUPATIONAL AND FAMILY MEDICINE, YONG LOO LIN SCHOOL

OF MEDICINE NATIONAL UNIVERSITY OF SINGAPORE

2005

i

ACKNOWLEDGEMENTS

My deepest appreciation to my supervisor, Professor David Koh who provided

ongoing encouragement, support and advice for this thesis His motivation and

support has not only encouraged me to complete this thesis but to venture into the world of academia

I am also grateful for the support of Associate Professors Chia Kee Seng and Chew Suok Kai and Drs Derrick Heng and Stefan Ma for their collaboration in this research project Many thanks also to the Singapore National Registry of Births and Deaths, the Central Claims Processing System and the Singapore Cancer Registry This thesis would certainly not exist without their kind assistance and permission to use the data Also many thanks to Mr Cheung Kwok Hang, who so very kindly and patiently assisted in the preparation of the data, and Doris for your editing help

My friends and colleagues of COFM, who have provided much support,

encouragement and a wonderful work environment throughout the years

My dear husband Kit Min for his loving, kind and steadfast support, and our children Shu-Wen and Wei-Sheng for being the mostly wonderful children they are

My dear parents, brothers and sister for their never ending love and support

My precious Lucky dog, who has always given me his unconditional love

And finally to my daughter, Lauren Chye Shu-Lin, you are the one who inspired and

spurred me to finish this thesis before your arrival into this world

ii

TABLE OF CONTENTS

ACKNOWLEDGEMENTS i

TABLE OF CONTENTS ii

SUMMARY x

LIST OF TABLES xiii

LIST OF FIGURES xv

LIST OF APPENDICES xvi

LIST OF MANUSCRIPTS, PUBLICATION AND PRESENTATIONS xvii

CHAPTER 1 INTRODUCTION AND GENERAL OBJECTIVE 1

CHAPTER 2 GENERAL BACKGROUND 3

2.1 FACTORS THAT AFFECT BIRTH WEIGHT 3

2.2 EPIDEMIOLOGICAL RESEARCH USING RECORD LINKAGE OF NATIONAL REGISTERS 5

2.2.1 Introduction 5

2.2.2 Advantages of register-based data 5

2.2.3 Limitations of register-based data 6

2.2.4 Summary 8

iii

2.3 DATA LINKAGE IN SINGAPORE 8

2.3.1 Introduction 8

2.3.2 Birth certificate/National Registration Identity Card Number (NRIC) 8

2.3.3 National Registers 9

2.3.4 Linkage of national registers 13

2.3.5 Data privacy and confidentiality 13

CHAPTER 3 ETHNIC DIFFERENCES IN BIRTH WEIGHT AND INFANT MORTALITY 14

3.1 BACKGROUND 14

3.1.1 Introduction 14

3.1.2 Asian Indians and low birth weight 14

3.1.3 Differences in infant mortality between ethnic groups and the “low birth weight paradox” 17

3.2 OBJECTIVES 19

3.3 METHODS 20

3.3.1 Data linkage 20

3.3.2 Cohort selection 20

3.3.3 Variables 20

iv

3.3.4 Outcome 21

3.3.5 Data analysis 21

3.4 RESULTS 23

3.4.1 Characteristics of the three ethnic groups 23

3.4.2 Variables associated with moderately low birth weight 26

3.4.3 Infant Mortality 29

3.5 DISCUSSION 33

3.5.1 Summary 33

3.5.2 Birth weight and infant mortality of the ethnic groups 33

3.5.3 “Fetal origins” hypothesis and further research 35

3.5.4 Strengths and limitations 36

CHAPTER 4 EFFECT OF LOW BIRTH WEIGHT AND PREMATURITY ON SUBSEQUENT HOSPITAL ADMISSIONS IN CHILDREN 37

4.1 BACKGROUND 37

4.1.1 Introduction 37

4.1.2 Prospective short-term studies 37

4.1.3 Prospective long-term studies 38

4.2 OBJECTIVES 42

v

4.3 METHODS 43

4.3.1 Data linkage 43

4.3.2 Cohort selection 43

4.3.3 Variables 43

4.3.4 Outcomes 44

4.3.5 Data analysis 45

4.4 RESULTS 46

4.4.1 Hospitalization by preterm status and birth weight groups 46

4.4.2 Factors associated with hospital admission 48

4.4.3 Common admissions 50

4.5 DISCUSSION 53

4.5.1 Summary 53

4.5.2 Comparison of incidence rate of preterm births 53

4.5.3 Hospital admissions 54

4.5.4 Further areas for research: Non-hospital associated morbidity and socio-economic and psychological impact 55

4.5.5 Possible preventive measures 56

4.5.6 Specific strengths and limitations 56

vi

CHAPTER 5 THE EFFECT OF BIRTH, MATERNAL FACTORS AND PRIOR

HOSPITAL ADMISSIONS FOR RESPIRATORY TRACT

INFECTIONS ON INCIDENT ASTHMA ADMISSION 58

5.1 BACKGROUND 58

5.1.1 Introduction 58

5.1.2 Birth weight and gestational age and risk of asthma 58

5.1.3 Infectious diseases and risk of asthma 62

5.1.4 Other risk factors for asthma that can be studied using data from the registries in Singapore 64

5.2 OBJECTIVES 66

5.3 METHODS 67

5.3.1 Data linkage 67

5.3.2 Cohort selection 67

5.3.3 Variables 68

5.3.4 Outcome 68

5.3.5 Data analysis 69

5.4 RESULTS 70

5.4.1 Asthma admissions 70

vii

5.4.2 Univariate analysis of variables 70

5.4.3 Multivariate analysis of variables 71

5.5 DISCUSSION 74

5.5.1 Summary 74

5.5.2 Birth weight was not associated with IAA but gestational age is 74

5.5.3 Other factors associated with IAA 74

5.5.4 Areas for further research 76

5.5.5 Specific strengths and limitations 77

CHAPTER 6 BIRTH WEIGHT AND THE RISK OF EARLY CHILDHOOD CANCER 79

6.1 BACKGROUND 79

6.1.1 Introduction 79

6.1.2 Birth weight and all cancers combined 79

6.1.3 Birth weight and leukaemia 80

6.1.4 Birth weight and other childhood cancers 80

6.1.5 Biological relationship between birth weight and childhood cancer 83

6.1.6 Studies of birth weight and childhood cancer in Asian populations 84

6.2 OBJECTIVE 85

viii

6.3 METHODS 85

6.3.1 Data linkage 85

6.3.2 Cohort selection 85

6.3.3 Variables 85

6.3.4 Outcomes 86

6.3.5 Data analysis 86

6.4 RESULTS 87

6.4.1 Childhood cancer in Chinese, Malays and Asian Indians 87

6.4.2 Types of childhood cancer 90

6.4.3 Factors associated with childhood cancer in Chinese children 91

6.4.4 Birth weight and specific cancers for Chinese children 93

6.5 DISCUSSION 96

6.5.1 Summary 96

6.5.2 Higher birth weight and risk of childhood cancer 96

6.5.3 Higher birth weight and risk of childhood leukaemia 96

6.5.4 Birth weight and the risk of other childhood cancers 98

6.5.5 Further research 98

ix

6.5.6 Specific strengths and limitations 99

CHAPTER 7 CONCLUSION AND RECOMMENDATIONS 101

7.1 RECOMMENDATIONS FROM STUDIES 101

7.2 RECOMMENDATIONS FOR NATIONAL REGISTERS 103

7.2.1 Suggested changes to existing registers 103

7.2.2 Assessment of quality of data 104

REFERENCES 107

APPENDICES 135

x

SUMMARY

Introduction

The effect of birth weight on infant survival and mortality is well established

However its effect on morbidity outcomes, especially those in childhood is less clear The primary aim is to assess the effect of birth weight on infant mortality and

morbidity outcomes in early childhood Secondary aims include assessment of other birth and maternal characteristics on the same outcomes

Cox proportional hazards and logistic regression models were used in the data

analysis

Birth and maternal characteristics included birth weight (500gms categories),

gestational age (weeks), gender (male/female), ethnic group (Chinese, Malay, Asian

xi

Indian or Others), birth order, maternal age and education Birth weight was

subsequently grouped into different categories that were specifically justified for each study

Outcomes assessed included infant mortality, hospital admissions between 1-5 yrs of age, incident asthma hospital admissions (IAA) between 0-5 yrs and cancer between 0-5 yrs In general follow-up was up to the time of the event, death or date of

censoring Person years (py) of follow-up ranged from 324,874-1,156,897 py in the various studies

Main findings

(i) The effect of birth weight on infant mortality has shown the existence of the

“low birth weight” paradox in Asian Indians but not Malay babies in Singapore That

is, compared to Chinese, for term babies with birth weight 1500-2499gms, those of

Asian Indian origin had a decreased risk of infant mortality, whilst Malays had an increased risk However compared to Chinese, for babies with birth weight 2500-

3499gms and >=3500gms, those of Asian Indian origin appeared to have an increased risk of infant mortality, whilst for Malays the infant mortality remained elevated for

these birth weight groups

(ii) Both early gestational age and lower birth weights were associated with

increased rates of hospitalization in children between 1-5 yrs of age A preterm baby with birth weight 500-999gms had a 1.77 times (adjusted for ethnicity) increased risk

of being ever admitted compared to a preterm baby of birth weight >=2500gms (iii) Birth weight was not associated with incident asthma admissions for children less than 2 years old or children 2-5 yrs of age

xii

(iv) High birth weight (>3499gms) was associated with a 1.5 times increased risk

of childhood cancer before 5 yrs of age in Chinese children

The main secondary findings were an increased risk of incident asthma admissions for children with recurrent respiratory tract infections, among Malay and Asian Indian ethnic group and males

Conclusion and recommendations

Birth weight has been shown to have ethnic differences Low birth weight in

premature babies is associated with hospitalization whilst high birth weight is

associated with increased risk of childhood cancer in this cohort of Singaporean children However further epidemiological and biological studies are needed to

confirm these findings because of either limitations in data available or small number

of outcomes as is the case for childhood cancer Given the evidence provided so far in this and other studies, the control of birth weight is not advocated as a preventive measure for either asthma or cancer and pregnant mothers should be encouraged to have a healthy balanced diet and adequate exercise

27 Table 3.6 Prevalence of MLBW within birth order and ethnic groups in term

babies 27 Table 3.7 Prevalence of MLBW within maternal age and ethnic groups in term

babies 28 Table 3.8 Prevalence of MLBW by maternal educational and ethnic groups in

term babies .28 Table 3.9 Causes of death in the first year for Chinese, Malay and Asian Indian

and All ethnic groups combined 30 Table 3.10 Infant mortality by ethnic groups, birth and maternal characteristics in

term babies .31 Table 3.11 Infant mortality by birth weight categories for Chinese, Malays and

Asian Indians in term babies 32 Table 4.1 Long-term prospective studies of children born premature or with low

birth weight .40 Table 4.2 Summary indices of admissions and days spent in hospital by birth

weight groups for children ever admitted between 1-5 years of age 47

xiv

Table 4.3 Children who had three or more admissions or whose total hospital stay

was equal to or greater than 7 days between 1-5 years by birth weight groups .47 Table 4.4 Admission rate by each year of age between 1-5 years of age .48Table 4.5 Incidence of ever being admitted between 1-5 years of age for birth and

maternal characteristics .49 Table 4.6 Adjusted risk for ever being admitted by term status for different birth

weight groups between 1-5 years of age .50 Table 4.7 Number and rate of admissions for the most common main ICD-9

groups for admissions between 1-5 years of age 51 Table 4.8 Incidence and risk of ever being admitted for respiratory tract diseases

between ages 1-5 years for preterm children by birth weight groups .52 Table 5.1 Studies that excluded children who were very low birth weight,

extremely premature, or had congenital diseases and that assessed the effect of birth weight on childhood asthma 60 Table 5.2 Rates of incident asthma admissions for children born between 1st

January 1992-31st December 1997 .70 Table 5.3 Variables that were significant in univariate analysis and adjusted in a

multivariate model for incident asthma admission between 0-<2 years 72 Table 5.4 Variables that were significant in univariate analysis and adjusted in

multivariate model for incident asthma admission between 2-5 years of age .73 Table 6.1 Population-based studies of different childhood cancers 81 Table 6.2 Cancer incidence by ethnic and gender groups for children born

between 1st January 1992-31st December 1998 88 Table 6.3 Number of cancers in children born between 1st January1992-31st

December 1998 and followed-up till 5 years of age, death or censor at

31st December 1999 89 Table 6.4 Types of cancer found in Chinese, Malays and Asian Indians 90 Table 6.5 Association of all childhood cancer with birth and maternal

characteristics for Chinese children 92 Table 6.6 Risk of leukaemia and lymphoma by birth weight categories for

Chinese children 94 Table 6.7 Risk of other cancers by birth weight categories for Chinese children 95

xv

LIST OF FIGURES

Figure 3.1 ICD-9 category codes used in the Singapore National Registry of

Births and Deaths .29 Figure 6.1 Distribution of cancers between 0-5 years of age for children born

between 1st January 1992-31st December 1998 for Chinese, Malays and Asian Indians 89

xvi

LIST OF APPENDICES

Appendix 1 Variables available in the Singapore National Registry of Births and

Deaths 135 Appendix 2 Variables available in the Singapore Cancer Register 137

xvii

LIST OF MANUSCRIPTS, PUBLICATION AND

PRESENTATIONS

Papers from this study

Lee J, ChiaKS, Chia SE, Hughes H, Cheung KH, Koh D Ethnic differences in birth weight and infant mortality amongst Chinese, Malay and Asian Indian term singletons

in Singapore Paed Perinat Epidemiol (manuscript submitted)

Lee J; Ma S; Chia KS; Cheung KH; Heng D, Chew SK, Koh D Effect of low birth weight and prematurity on subsequent hospital admissions in children between 1-5 years of age in Singapore Eur J Pub Health (manuscript submitted)

Lee J, Heng D, MaS, ChewSK, Chia KS, Koh D The effect of respiratory tract infections, birth weight and gestational age on incident hospital admissions for asthma between birth and 5 years of age Paed Perinat Epidemiol (manuscript submitted)

Lee J, Chia KS, Cheung KH, Chia SE, Lee HP (2004) Birthweight and the risk of early childhood cancer among Chinese in Singapore Int J Cancer 110, 465-467

1

CHAPTER 1 INTRODUCTION AND GENERAL OBJECTIVE

The observation of association between birth weight and disease was perhaps the most publicized by

the “thrifty hypothesis” of DJ Barker for ischaemic heart disease in adults (Barker et al., 1989) This

hypothesis stems from the belief that in-utero effects may play a role in the development of diseases

in later life Prior to this the focus of birth weight was predominantly on survival very early in life Recently more research has also been conducted on mortality and morbidity outcomes ranging from respiratory, to cardiovascular and cancer outcomes; in childhood and adulthood This thesis focuses

on outcomes in infancy and early childhood

Register-based data linkage research can be a useful tool to determine the effects of factors on

outcomes collected in population registry data bases In particular, the use of population registers for prospective follow-up is ideal for cohort studies as it greatly minimizes the biases associated with smaller epidemiological studies However it is dependent on the content and quality of the databases, and this may vary between registries and for different outcomes All relevant information may not be available and the standardization may differ within and between some registers Nevertheless such research can be useful indicator of possible associations that may demand further attention

Singapore is a city state with a population of approximately 3.5 million residents and is comprised of three main ethnic groups Chinese, Malays, Asian Indians and other ethnic groups represent 76.7%, 13.8% and 8.3% and 1.7% of the population respectively (Health Information Management Branch,

2004) The social and political stability has allowed the establishment of nationwide registries that

collect socio-economic and health data Health services are also well-developed with affordable and equitable access for the majority of the population Thus Singapore a good location to conduct population-wide register based research

2 Using the population-based national registers available in Singapore, the general objective of this study is to determine the effect of birth weight on mortality and morbidity outcomes in childhood

3

CHAPTER 2 GENERAL BACKGROUND

2.1 FACTORS THAT AFFECT BIRTH WEIGHT

There are many known factors that can influence birth weight These have been previously

extensively studied in different populations worldwide A detailed assessment and meta-analysis of

43 determinants of low birth weight by Kramer states “factors with well-established direct causal impacts on intrauterine growth include infant sex, racial/ethnic origin, maternal height, pre-

pregnancy weight, paternal weight and height, maternal birth weight, parity, history of prior low birth weight infants, gestational weight gain and caloric intake, general morbidity and episodic

illness, malaria, cigarette smoking, alcohol consumption and tobacco chewing” (Kramer, 1987) The

effects of these factors are summarized in Table 2.1 Although these factors are known to affect birth weight, they may not necessarily also affect the outcomes of interest for the following studies and are thus not all considered as confounders

Table 2.1 Factors that have an independent effect on birth weight

Factors Effect

Infant sex Males tend to have higher BW than females

Racial/ethnic origin Blacks, Indians and Pakistanis have lower BW than Europeans

Maternal Height Positive association with BW

Pre-pregnancy weight Positive association with BW

Paternal weight and height Possible positive association but less effect than maternal size

Maternal BW Positive association with BW

Parity Positive association with BW but may not be so for multi-paras History of prior low BW infants Positive association with BW

Gestational weight gain/caloric intake Positive association with BW

General morbidity and episodic illness Possible positive association with BW

Malaria Possible inverse association with BW

Cigarette smoking Inverse association with BW

Alcohol consumption Inverse association with BW

Tobacco chewing Inverse association with BW

Summary of factors assessed by Kramer et al., 1987, Birth weight (BW)

4

Birth weight and gestational age are two common measures that are used to assess maturity at birth

In a normally developing and healthy fetus, gestational age correlates in a positive and linear manner with birth weight (Robertson, 1999) Opinions differ as to which measure should be used Although

it makes biological sense to consider gestational age with birth weight as a consequence of this, many researchers have used birth weight as the measure of maturity primarily because of availability and reliability of data Also, it is well recognized that gestational age that is determined by the

woman’s last menstrual period may not be particularly precise The majority of collaborative

networks thus report data in terms of birth weight (Ward et al., 2003)

However some investigators believe that the use of birth weight is inappropriate because it can lead

to the misclassification of babies that although small, are more mature and that these differences in maturity and growth results in a misleading protective effect of intrauterine growth retardation (Arnold et al., 1991) Thus restricting or stratifying by gestational age allows for better assessment of the proportion of babies who are small but not necessarily premature Thus it has been generally recommended to assess the effect of birth weight restricted to babies born at term as the effect of gestational age at this stage is minor (Wilcox, 2001) This is particularly important for studies that assess the affect of fetal growth and retardation on clinical outcomes and survival However from a public health perspective birth weight can be used to determine resource requirements for the

different birth weight groups Thus the literature is mixed in the use of either birth weight or

gestational age

Population-based epidemiological studies utilizing record linkage have been used for several

decades in countries that have nationwide registers that collect information on general demographic factors as well as health indices These include the Nordic countries, USA and UK This data is often termed as “secondary” in comparison to primary data collected by the use of traditional

epidemiological methods

Advantages of register-based data

The strengths of using registry-based data can be seen especially in the comparison of methods used

in cohort studies Gissler et al commented that, “follow-up studies have generally been based on costly ad hoc cohort studies with detailed information collected specifically for research purposes on

a certain group of people” whilst “the increasing collection of routine health data provides an

alternative method of gathering follow-up data” (Gissler et al., 1998)

The use of register-based data can also enable long-term follow-up for cohort studies For example, Gissler et al discuss the importance of long-term follow-up in monitoring the health of children, whereby “among young children, future problems may be more important than immediate health status In addition, long term follow-up is especially needed for studies on risk factors for diseases and special social, health or educational needs.” Furthermore “most longitudinal data has been on mortality and data on diseases have been derived from cross-sectional or ad hoc follow-up studies” and “health monitoring is hampered by incomplete or discontinuous data collection”(Gissler et al., 1998) Thus it is possible using secondary data sources to cover the entire population and sample

Limitations of register-based data

The benefits of utlizing registers are based on certain assumptions; that health care data are of high quality; that information on individuals should be linkable across data sets; that individuals in the database should be traceable through time to provide longitudinal follow-up (Roos et al., 1987)

The type and quality of data collected may also vary Unfortunately some but not all desired

information may be available, unlike in primary studies whereby the investigator may choose to collect comprehensive information For example birth certificates may state birth weight but not gestational age Also, birth weight may not be entered as an exact number but within predetermined categories Other information such as antenatal care and maternal diseases may not be recorded

The data that is available may also not be accurately recorded This is more likely to occur for fine distinctions between diagnoses of diseases rather than obvious major disease groups, or death The registration of diseases is dependent upon patient presentation and contact with medical care Also for some registries, no uniform set of tests or investigations are done for the diagnosis and the

7

absence of recording of a particular disease in an individual’s history is no guarantee that the disease does not exist The data entered may also be subject to preferences of individual physicians or even the trend of diagnosis for that period of time Gissler et al used several health registers in Finland to study the feasibility of the data collection method and to evaluate the possibilities of using the

information for further research (Gissler et al., 1998) Validation of the data collection system was done by obtaining information from different registers and comparing them with each other, as well

as with earlier published surveys The cumulative incidence of all diseases and of four specific diseases/conditions was chosen as examples (asthma, diabetes, epilepsy, special delay in

development and intellectual disabilities) The conclusion was that in general the databases in

Finland are a good information source for follow-up studies However there are potential problems with variation in the content of databases and in data quality of different registries

Bias in reporting of different diseases may also exist For example, Romano et al compared 2 clinical

databases and 2 administrative databases of patients who had coronary artery bypass surgery and found that chronic or asymptomatic co-morbidities such as cardiomegaly, previous myocardial

infarction, tobacco use and hyperlipidaemia are less prevalent in the administrative databases

(Romano et al., 1994) However the prevalence of serious diseases such as diabetes, unstable angina and congestive cardiac failure were reflected in a similar manner Also, certain conditions such as myocardial infarction or cancers tend to be recorded whereas those that may not require

hospitalization will not Most registries also rely on the ICD-9 classification and this has the

limitation of unclear coding definitions Diagnostic errors may thus bias estimates of risk in either direction, depending on whether the misclassification of risk status is random or differential

Random misclassification always biases risk estimates towards the null Differential

misclassification may result in bias in either direction

infections or asthma The establishment of population-based disease registers with defined criteria for diseases would overcome some of these limitations Nevertheless, population-based registers are useful in providing preliminary answers to research questions These would need to be further

assessed using traditional epidemiological studies, or a combination of epidemiological baseline data and high quality (complete and accurate) population registers to follow-up outcomes of interest

2.3 DATA LINKAGE IN SINGAPORE

Introduction

This series of prospective population-based studies was conducted using the linkage of pre-existing national databases in Singapore Like many other developed countries, several types of nation-wide population-based registers exist in Singapore The registers used for this research project are

described

Birth certificate/National Registration Identity Card Number (NRIC)

All Singaporeans are issued with a compulsory birth certificate number at birth that subsequently becomes the NRIC number at the age of 15 years This number is used for all governmental and administrative purposes in Singapore Because all registries utilize the birth certificate number/NRIC

9

2.3.3

as a form of identification, it is possible to carry out data-linkage studies in Singapore with a

reasonable degree of accuracy and ease

National Registers

2.3.3.1 Singapore National Registry of Births and Deaths (SNRBD)

By law all births and deaths that occur in Singapore must be registered and population coverage is ensured (http://app.ica.gov.sg/serv_citizen/birth_death_reg/birth_)

However the information available in the birth and death certificates is limited The register is only available in electronic format starting from the birth cohort of 1992 Information available on the birth certificate includes ethnic group, gender, date and hour of birth, birth weight, gestational age (weeks), type of birth (singleton, twin, triplet etc.), total number of liveborn children (including the current pregnancy), parents’ names and NRIC (Appendix 1) Birth weight is recorded in 500gram

categories ranging from less than 500gms to greater than 5000gms (eleven categories in total)

Information available on the death certificate include place, date and time of death, as well as

primary and secondary cause of death coded using the International Classification of Diseases –9thEdition (ICD-9) (Appendix 1) The death certificate and in particular the cause of death is only completed by qualified medical doctors Unfortunately no published studies have assessed the

quality of the data in this register

The main strength of this register was the accessibility to a national cohort and thus large sample size The data collected in the SNRBD is also prior to the onset of disease and is thus not subject to recall bias In particular, birth weight is measured at birth by trained healthcare professionals and recorded soon after during the registration of the birth Information bias that may occur such as

proportion of mothers was incomplete The data is complete for almost all of the variables, with the exception of maternal education where up to 31.3%, 27.3% and 17.3% of Chinese, Malay and Asian Indians respectively did not specify their highest attained educational level on the birth certificate This is likely to be because although the specific birth details are completed by medical doctors or trained midwives, parental education is often left for the parents to complete The proportion of gestational age dated by ultrasound is also thought to be low in Singapore (personal communication

PC Wong, Head of Obstetrics & Gynaecology, National University Hospital, Singapore) and these are indistinguishable from pregnancies dated using last menstrual cycle

Also, data on other maternal factors that are known to influence birth weight were not available Examples include cigarette smoking and high intake of alcohol Nevertheless in Singapore, the prevalence of these habits are low and thus likely to play a smaller role than for other populations In

the National Health Surveys conducted in 1992 and 1998, age standardized smoking rates for

females were 2.9% and 3.2% for 1992 and 1998 respectively (National Health Survey 1998) In

1998 smoking rates for females aged 20-24 was 6.7% and 2.6 % for those aged 25-44 years For alcohol intake, age standardized prevalence rates were similar for 1992 and 1998, being about 0.8% for women Finally, although the majority of deaths that occur in this cohort are known, because there is no active follow-up, children not registered in the death register were assumed to be alive

11

2.3.3.2 Central Claims Processing System (CCPS)

All discharges from public and private hospitals are captured in this register which thus provides information on admissions for the whole nation (personal communication, Dr Derrick Heng, Deputy Director, Epidemiology and Disease Control Division, Ministry of Health, Singapore) The

discharges are coded as ICD-9 diagnoses and also as primary and secondary causes of admission Date and duration of hospital stay is also recorded The CCPS has been used primarily as submission

of claim forms for Medisave, a compulsory savings scheme for medical care for all Singaporeans Thus all discharges from both private and public hospitals are submitted to be entered in this register (Heng et al 2000) However this register is dependent on the individual’s admission to hospital for the particular illness and thus cannot serve as a comprehensive coverage of disease prevalence within the community Thus coverage is poor for diseases that usually do not require hospitalization Nevertheless more severe illnesses for the particular disease may require admission and would be captured in this register

The strengths of using this register were that it captures all hospital discharges and thus admissions

in Singapore Thus selection bias for hospital admission is minimal Admissions for the outcome of interest as well as admissions for other diseases could be obtained for each individual Also because

of the large number of outcomes captured, subgroup analysis was possible

However the decision to admit a child (such as for social circumstances) cannot be ruled out It is also possible that doctors’ awareness of the child’s prior birth and medical history may influence the admission threshold thus causing differential bias that could artificially increase the risk Although all diagnoses were made by medical doctors and coded in an internationally accepted classification system (ICD-9), the main limitation of using registered admissions was that no standardized criteria are provided for the diagnoses However at the time of this study, the majority of doctors working in

12

hospitals in Singapore received their training in one medical school They are more likely to have similar diagnostic criteria and acumen Hospital admission diagnoses were also not subject to

information bias that may occur in self-reported data

2.3.3.3 Singapore Cancer Registry (SCR)

The Singapore Cancer Registry was set up in 1967 with the assistance of the National University of Singapore, the Ministry of Health and the International Agency for Research on Cancer, World Health Organization Data available from this registry is shown in appendix 2 The completeness of registration is ensured by routine review of pathology records, hospital discharge records and death certificates Notification is also made by medical practitioners A cancer is registered if there is a pathological diagnosis of cancer, a clinical diagnosis of cancer supported by surgical, radiological and laboratory findings, or mention of cancer in the death certificate For cancers in adults the

primary sites are coded according to ICD-9 and histological types according to the Manual of Tumor Nomenclature and Coding (MOTNAC) However cancer cases in children are classified according to the International Childhood Cancer Classification (ICCC) of the International Agency for Cancer

Research (Parkin et al., 1998) Quality assurance checks show that the proportion of “death

certificate only” (DCO) notifications was 4.2% for the period 1968-1977 and 1.0% for the period

1993-1997, thus indicating data of good quality (Chia et al., 1996)

The detection of children with cancer through the SCR removes potential selection bias The only cases that were not captured were for those children who have migrated prior to cancer detection However childhood cancer is a rare disease and the number of such cases, if any, are thought to be very small Unlike the CCPS, cases are well defined through standardized methodology including the use of pathological and hospital reports

13

2.3.4

2.3.5

Linkage of national registers

The linkage of SNRBD and CCPS was done by the Singapore Ministry of Health Record linkage was done using the NRIC number Further cross-checking was performed using gender, race and date of birth For reasons of confidentiality the NRIC number of all individuals in the dataset were replaced with a serial number after linkage, with the key held by the Singapore Ministry of Health (which does not have access to the linked data-set)

The linkage of the SNRBD and SCR was done by the National Disease Registries Office NRIC was used as the main personal identifier with date of birth used to cross-check Similarly all personal identifiers were removed after data-linkage

Data privacy and confidentiality

Strict measures were taken to ensure privacy and confidentiality of data All personal identifiers were removed from the dataset with linkage between SNRBD (for births and deaths) and CCPS and between SNRBD and SCR by the Ministry of Health and National Disease Registries Office

respectively Thus the dataset that was used for final data analysis did not have any personal

identifiers There was also no personal access to identified data

to exist between Asian Indians and Caucasians in Norway for perinatal mortality It would be

interesting to determine if both currently exist in Singapore

Asian Indians and low birth weight

Compared to other ethnic groups, Asians are known to have the highest percentage of low birth weight babies (Kramer 1987) In particular, the lower birth weights of Asian Indians compared to other Asian and Non Asian ethnic groups have been well documented in previous studies conducted locally in Singapore (Hughes et al., 1984; Hughes et al., 1986; Viegas et al., 1989) and abroad in migrants (Dawson et al., 1982; McFadyen et al., 1984; Terry et al., 1987; Wilcox et al., 1993; Le et al., 1996; Vangen et al., 2002) These studies are shown in Table 3.1 The majority of studies

included babies of all gestational ages (Dawson et al., 1982; Hughes et al., 1984; Hughes et al., 1986; Terry et al., 1987; Wilcox et al., 1993; Le et al., 1996; Vangen et al., 2002) As previously discussed this may be considered inappropriate and thus further analysis of ethnic differences in only term babies or stratifying by gestational age should be done Some studies have also not excluded

respectively and this may be attributed to better socio-economic and health care status of the

country It would thus be interesting to see if ethnic differences still remain in the current setting of a

developed nation

16

Table 3.1 Studies of ethnic differences in birth weight between Asian Indians and other ethnic

groups Author Year Country Study population Study size Inclusion Findings

Hughes 1984 Singapo

re

Random sample Singapore Birth Registry 1967-1974

23 591 (76.9%

Chinese, 14.6 Malay, 6.4%

Asian Indian)

Singletons of all gestational ages

Compared to Chinese and Malays, Asian Indians have highest proportion of LBW(11.5%) babies Hughes 1986 Singa-

pore

All births from Singapore Birth Registry 1981-1983

125,189 (74%

Chinese, 17.8% Malay, 7.5% Asian Indian)

Singletons of all gestational ages

Compared to Chinese and Malays, Asian Indians have highest proportion of LBW(6.1%) and

1800 (600 Chinese, Malays and Asian Indians)

Singletons, 37-42 wks

Mean birth weight of Asian Indians 100gms less than Chinese, and overall highest proportion of LBW Dawson 1982 UK Obstetric hospital

in West London1967-1975

6000 Asian Indian, 18000 Whites

Singletons, exclude congenital diseases

Mean birth weight of Asian Indians lower than UK white babies by approx 235 gms

Terry 1987 UK Birmingham

hospital 1979-1982

4185 Indian,

2193 Pakistani,

1859 West Indian 6514 European

Singletons, exclude congenital diseases

West Indians have a higher proportion of VLBW babies, followed by Indian, Pakistani and lastly

European Wilcox 1993 UK Nottingham

University

1986-1991

37336 European,

1008 African,

1547 Indian/Pakista

ni

Singletons, exclude congenital disease, or no U/S scan before 24wks

Indian/Pakistani have lowest mean birth weight

Le 1996 USA 1992 US Natality

file

848,993 Whites, Asian American and Pacific

Islander subgroups

All singletons Asian Indians highest

proportion of MLBW and VLBW babies

17

Vangen 2002 Norway Birth registry

1980-1995

808658 Norwegian,

6854 Pakistani,

3283 Vietnamese,

1461 Nth African

3.1.3 Differences in infant mortality between ethnic groups and the “low birth weight paradox”

Linked birth and infant death certificates allow for measurement of birth weight-specific infant mortality (Buehler et al., 2000) Although a definite inverse association between birth weight and perinatal, neonatal or infant mortality has been shown in several large population-based studies (Leon et al., 1998; Samuelson et al., 1998; Power et al., 2000; Friedlander et al., 2003), differences

exist among ethnic groups (Le et al., 1996; Vangen et al., 2002)

Two previous studies have examined migrant/ethnic differences with regard to birth weight and perinatal mortality between Asian Indians and other ethnic groups in the United Kingdom (Dawson

et al., 1982) and Norway (Vangen et al., 2002) Dawson et al showed that although Asian Indian infants were smaller than UK infants, the perinatal mortality of infants <2500gms was lower in the Asian Indian group (Dawson et al., 1982) A more recent study by Vangen et al also showed similar results (Vangen et al., 2002) Interestingly both show the “low birth weight paradox” (Wilcox 2001) whereby in populations that have a higher proportion of low birth weight babies, the babies in the low birth weight category appear to have decreased mortality compared to the population with the lower proportion of low birth weight babies Also, in the former group, babies with higher birth

weights appear to have increased mortality compared to the latter The paradox can be resolved by

converting birth weight values to the relative scale to show that regardless of ethnicity, birth weight

18

and mortality follow a similar trend In the case of Vangen et al after correction to the relative scale,

it appears that mortality in the migrant group is higher at all levels of birth weight when compared to the Norwegians However it also indicates that babies of different ethnic groups have different scales

of growth

Wilcox has also provided several examples of the “low birth weight paradox” with regards to the

smaller babies of mothers who smoked, as well as babies born at high altitude (Wilcox 2001) In the

case of maternal smoking, an increased risk of perinatal death was found for all birth weights after correction to a relative scale In the case of babies born at high altitudes, the apparent protective effect of a decreased risk of perinatal death for the smaller babies disappeared to show that they have

the same risk of perinatal death for babies with similar size in the general US population Thus the

birth weights have shifted to lower weights at higher altitude together with the mortality curve As Asian ethnic groups are known to have the smallest babies (Kramer 1987), it would be interesting to see if the “low birth weight paradox” exists with regard to birth weight and infant mortality in a multiethnic Asian population such as Singapore

consists of 325375 (70.9%) individuals

Variables

The variables from the SNRBD were categorized to include ethnicity (Chinese, Malay and Asian Indian) of the child and both parents, year of birth, gender (male and female), birth weight groups (available in the original form as 500gms categories and further categorized as very low birth

weight, <1500gms (VLBW), moderately low birth weight, 1500-2499gms (MLBW), 2500-3499gms and >=3500gms), gestational age in weeks, birth order of baby (1st, 2nd and >=3) and mother’s age

in years Maternal age is also categorized into three approximately proportionate groups of <27 years, 27-32 years and >=32 years The highest attained maternal educational level was categorized

as none or up to 6 years (no formal/primary), 6-12 years (O/N/A level), more than 12 years

category (Not Specified) in subsequent analyses The variables of gestational and maternal age were also used as continuous variables for adjustment in data analysis where appropriate Also of note, maternal age and birth order were found to be highly correlated with each other and thus not used in the same multivariate model

Outcome

Infant mortality was the outcome of interest

Data analysis

SPSS (version XIII ) (SPSS for Windows, Rel 13.0.1 2004 Chicago: SPSS Inc.) and CIA

(Confidence Interval Estimation, Version 1.0, copyright 1989:Gardner MJ and British Medical Journal) software were used for data analysis of all the following studies

Numbers, proportions and rates for categorical variables and outcomes were calculated

Cox’s proportional hazards model (CPH) (Cox 1972) was used to calculate prevalence rate ratios

(PRR) with 95% confidence intervals (95% CI) of a baby being a particular ethnic group within birth weight categories (Lee et al., 1993) These were both unadjusted and adjusted for other birth and

maternal characteristics where appropriate Chinese babies were the referent group Follow-up time

was set at 1 for this particular analysis

Infant Mortality Rates (IMR) with 95% CI were calculated for each ethnic group overall and within the birth weight groups Confidence intervals for mortality rates were calculated using the normal

distribution when events were large and a near approximate normal method for rare events