dietary carbohydrates, associated biomarker, and genetic

Dietary Carbohydrates, Associated Biomarker, and Genetic Risk

in Cardiometabolic Diseases

by

Mengna Huang, M.P.H.

A dissertation submitted in partial fulfillment of the

requirements for the degree of Doctor of Philosophy

in the Department of Epidemiology at Brown University

Providence, Rhode Island

May 2018

iii

This dissertation by Mengna Huang is accepted in its present form

by the Department of Epidemiology as satisfying the

dissertation requirement for the degree of Doctor of Philosophy.

Date_______________ ___________________________________

Simin Liu, Advisor

Recommended to the Graduate Council

Date_______________ ___________________________________

Stephen T. McGarvey, Reader

Date_______________ ___________________________________

Xi Luo, Reader

Approved by the Graduate Council

Date_______________ ___________________________________

Andrew G. Campbell, Dean of the Graduate School

iv

Curriculum Vitae

Mengna Huang graduated with a B.S. degree in Biological Sciences from Tsinghua

University, Beijing, China. She then obtained an M.P.H. degree from Brown University,

and later joined the Ph.D. program in Epidemiology at the Brown University School of

Public Health in Fall 2013, with Dr. Simin Liu as her primary advisor. Mengna has been

interested in nutritional and chronic disease epidemiology since her M.P.H. studies, and

her research has contributed to publications in dietary carbohydrates intake in relation to

biomarker levels and disease risk in cardiometabolic health. During her Ph.D., she also

served as teaching assistant for four graduate-level courses in the School of Public health,

including two in Epidemiology and two in Biostatistics. She is also one of the founding

members of the Brown Chapter of Statistics in the Community, a pro-bono statistical

consulting group led by graduate students that provides analytical services to community

partners in Providence.

v

Preface

This dissertation takes the form of three publishable manuscripts (Chapters 2-4). This

document was formatted in accordance with the requirements of the Graduate School at

Brown University.

Chapter 1 is an introduction of the significance, rationale, and specific aims of the

research conducted in the following chapters.

Chapter 2 is a version of the manuscript titled “Relations between Dietary

Carbohydrates Intake and Circulating Sex Hormone-Binding Globulin Levels in

Postmenopausal Women”, that was published in the Journal of Diabetes.

Chapter 3 is a version of the manuscript titled “Comparisons of Multiple Sources

of Starchy Foods, and Their Estimated Substitution Effects on the Risk of Type 2 Diabetes

and Atherosclerotic Cardiovascular Disease in Postmenopausal Women”. This manuscript

is currently in preparation for submission.

Chapter 4 is a version of the manuscript titled “Genetic Variations Related to Type

2 Diabetes and Atherosclerotic Cardiovascular Disease, and Their Interactions with

Dietary Carbohydrates in Influencing Long-Term Disease Risk”. This manuscript is also

currently in preparation for submission.

vi

Acknowledgements

I would like to thank my primary advisor, Dr. Simin Liu, for his guidance, support, and

encouragement during my doctoral training. Dr. Liu has provided me with great resources

and opportunities, and has always advocated on my behalf, encouraged me to be confident,

and pushed me forward. Without these I would not have been able to complete this work

and become an independent researcher.

I am also deeply grateful to my committee members, Dr. Stephen T. McGarvey and

Dr. Xi Luo, for their input, comments, and feedback to my dissertation work, as well as

their encouragement and patience throughout.

I would like to thank all the faculty and staff of the Epidemiology Department, and

all other faculty and staff of the School of Public Health, who have taught or assisted me

during my years here. And to past and current members of my research group, my friends,

peers, and colleagues, I am grateful to have been inspired by you all.

Finally, I would like to give my special thanks to my family, especially my parents

Wei Huang and Jialing Gong, who believe in me unconditionally; and to my best friends

Xiaochen, Yi, and Su, for always being there for me.

vii

Table of Contents

Curriculum Vitae ............................................................................................................. iv

Preface ................................................................................................................................ v

Acknowledgements .......................................................................................................... vi

Table of Contents ............................................................................................................ vii

List of Tables ..................................................................................................................... x

List of Illustrations .......................................................................................................... xii

Chapter 1 Introduction..................................................................................................... 1

1.1 Background and Significance .................................................................................... 1

1.2 Specific Aims ............................................................................................................ 3

Chapter 2 Relations between Dietary Carbohydrates Intake and Circulating Sex

Hormone-Binding Globulin Levels in Postmenopausal Women .................................. 4

2.1 Introduction ............................................................................................................... 4

2.2 Methods ..................................................................................................................... 6

2.2.1 Study Subjects .................................................................................................... 6

2.2.2 Measurement of Serum SHBG Concentrations .................................................. 7

2.2.3 Dietary Measurements ........................................................................................ 7

2.2.4 Statistical Analysis ............................................................................................. 9

2.3 Results ..................................................................................................................... 10

2.4 Discussion ............................................................................................................... 13

2.5 Tables ...................................................................................................................... 18

viii

2.6 Supplementary Tables ............................................................................................. 28

Chapter 3 Comparisons of Multiple Sources of Starchy Foods, and Their Estimated

Substitution Effects on the Risk of Type 2 Diabetes and Atherosclerotic

Cardiovascular Disease in Postmenopausal Women ................................................... 35

3.1 Introduction ............................................................................................................. 35

3.2 Methods ................................................................................................................... 36

3.2.1 Study Population .............................................................................................. 36

3.2.2 Measurement of Outcomes ............................................................................... 37

3.2.3 Measurement of Pasta....................................................................................... 38

3.2.4 Statistical Analysis ........................................................................................... 39

3.3 Results ..................................................................................................................... 42

3.4 Discussion ............................................................................................................... 45

3.5 Tables ...................................................................................................................... 50


Chapter 4 Genetic Variations Related to Type 2 Diabetes and Atherosclerotic

Cardiovascular Disease, and Their Interactions with Dietary Carbohydrates in

Influencing Long-Term Disease Risk ............................................................................ 59

4.1 Introduction ............................................................................................................. 59

4.2 Methods ................................................................................................................... 61

4.2.1 Study Population .............................................................................................. 61

4.2.2 Study Variables ................................................................................................ 63

4.2.2.1 Dietary variables ........................................................................................ 63

4.2.2.2 Genetic data ............................................................................................... 64

ix

4.2.2.3 Type 2 diabetes .......................................................................................... 65

4.2.2.4 Atherosclerotic cardiovascular diseases .................................................... 66

4.2.3 Analytic Approaches ........................................................................................ 67

4.3 Results ..................................................................................................................... 70

4.4 Discussion ............................................................................................................... 72

4.5 Tables ...................................................................................................................... 76


Appendix .......................................................................................................................... 85

Bibliography .................................................................................................................... 87

x

List of Tables

Table 2.5.1 Baseline characteristics by quartiles of untransformed plasma SHBG

concentrations in a subpopulation of the postmenopausal women from the

Women’s Health Initiative (n = 11,159) .........................................................18

Table 2.5.2 Adjusted means of serum SHBG concentrations according to quartiles of

dietary glycemic load, glycemic index, and intakes of fiber, sugar, and total

carbohydrates ................................................................................................. 20


intake of carbohydrates abundant food items ................................................ 22



carbohydrates in controls of the ancillary studies .......................................... 24


intake of carbohydrates abundant food items in controls of the ancillary studies

....................................................................................................................... 26

Table 3.5.1 Baseline characteristics of WHI participants in analytic sample by quartiles of

total pasta intake (n = 84,555) ....................................................................... 50

Table 3.5.2 Estimates of relative risk and 95% confidence intervals (CIs) of diabetes

according to quartiles of pasta intake ............................................................ 52

Table 3.5.3 Estimates of relative risk and 95% confidence intervals (CIs) of CHD


xi

Table 3.5.4 Estimates of relative risk and 95% confidence intervals (CIs) of stroke


Table 3.5.5 Estimates of relative risk and 95% confidence intervals (CIs) of ASCVD


Table 3.5.6 Estimates of relative risk and 95% confidence intervals (CIs) of diseases of

interest according to quartiles of residual spaghetti intake from Model 3*... 56

Table 3.5.7 Estimates of relative risk and 95% confidence intervals (CIs) of diseases of

interest by statistically substituting pasta for other starch-dense foods from

Model 3* ........................................................................................................ 57

Table 4.5.1 Baseline characteristics of WHI-SHARe African American (AA, n=5,811) and

Hispanic American (HA, n=2,718) participants, and WHI-GARNET European

American (EA, n=3,472) participants according to their incident diabetes status

....................................................................................................................... 76

Table 4.5.2 Baseline characteristics of WHI-SHARe African American (AA, n=5,811) and

Hispanic American (HA, n=2,718) participants, and WHI-GARNET European

American (EA, n=3,472) participants according to their incident ASCVD

status .............................................................................................................. 77

Table 4.5.3 Significant interactions between genes of interest and components of dietary

carbohydrate identified in WHI-AA .............................................................. 78


carbohydrate identified in WHI-HA .............................................................. 79


carbohydrate identified in WHI-EA .............................................................. 81

xii

List of Illustrations

None.

1

Chapter 1 Introduction

1.1 Background and Significance

Despite medical advances in prevention, diagnosis, and treatment, cardiometabolic

diseases remain a global epidemic and inflict millions of deaths worldwide. Of these deaths,

an estimated 7.4 million were due to coronary heart disease (CHD) and 6.7 million due to

stroke (leading cause of disabilities).1 The number of people with diabetes has risen from

108 million in 1980 to 422 million in 2014 globally.2, 3 In the United States, cardiovascular

disease (CVD) remains the leading cause of death for both men and women, affecting an

estimated 85.6 million Americans adults.4 It was also estimated that a total of 28.9 million

people aged 20 years or older were affected by diagnosed or undiagnosed diabetes in 2012.5

The quality and quantity of dietary carbohydrates have been extensively studied in

terms of their roles in the development of cardiometabolic conditions, including obesity,

metabolic syndrome, type 2 diabetes, and atherosclerotic cardiovascular diseases.6-11 They

have also been related to biomarkers such as insulin sensitivity, lipid profiles, adipokines

and inflammatory markers,12, 13 while their relations with another important marker of

cardiometabolic health, sex hormone-binding globulin (SHBG),14 have not been

comprehensively assessed.

Traditional dietary patterns such as the Mediterranean diet typically contain

considerable amounts of carbohydrate intake, particularly starchy foods such as pasta

which have received little research attention. Plant-based foods containing high quantities

2

of starch have been argued to play an essential role in the evolution of the human nutritional

ecological niche and biological phenotypes.15 Although different foods high in dietary

carbohydrates are a major source of the energy required for basal metabolism and daily

activities, they have substantially varying properties, especially in terms of their chemical

structure and glycemic index. For instance, given the same amount, the traditional Italian

dish pasta appears to have lower glycemic index (GI) as well as glycemic load (GL),

compared to other major sources of starch, such as white bread or mashed potato.16 Since

the introduction of the concept of glycemic index (GI) and glycemic load (GL) in the

1980s,17 convincing evidence has been accumulated indicating that a higher GI/GL diet

was associated with increased risk of type 2 diabetes and coronary heart disease (CHD),

the most common type of cardiovascular diseases.18 It is then natural to hypothesize that

consumption of different food sources of dietary starch may have different effects on the

long term risk of cardiometabolic diseases, given the same total energy intake and total

carbohydrate consumption.

Different cardiometabolic diseases have been found to share common genetic

basis.19, 20 While genome-wide association studies (GWAS) have successfully identified

multiple single nucleotide polymorphisms (SNPs) that are associated with the risk of

obesity, type 2 diabetes, and cardiovascular diseases, most of them individually confer a

relatively modest amount of elevated risk, and the complex interplay between genetic

variants and our daily diet are still to be elucidated. In the development of cardiometabolic

diseases, the possibility that some people may be more prone to the harmful effect of high

GI diet, or that some others may benefit more from low GI starchy foods, remains to be

explored. Investigations into these potential interactions may provide additional insight and

3

/or generate hypothesis for future mechanistic research in personalized and precision

medicine.

1.2 Specific Aims

This dissertation includes the rationale, analyses, and findings regarding the following

specific aims:

Aim 1: to evaluate the relations between the intakes of various dietary carbohydrates and

circulating sex hormone-binding globulin (SHBG) levels in postmenopausal women of the

Women’s Health Initiative study (Chapter 2).

Aim 2: to evaluate the associations between the intake of a major food sources of dietary

starch, pasta, and the long-term risk of type 2 diabetes and atherosclerotic cardiovascular

diseases (ASCVD, including coronary heart disease and stroke) in the Women’s Health

Initiative study (Chapter 3).

Aim 3: to evaluate the interactions between major food sources of dietary starch and

genetic variants related to type 2 diabetes and ASCVD, and collectively their associations

with the risk of these diseases, in the 1000G imputed genetic data from the European,

Hispanic, and African American participants of the Women’s Health Initiative study

(Chapter 4).

4

Chapter 2 Relations between Dietary Carbohydrates Intake and Circulating Sex

Hormone-Binding Globulin Levels in Postmenopausal Women

2.1 Introduction

Sex hormone-binding globulin (SHBG) is a serum protein synthesized by the liver that

binds to both androgens and estrogens, with higher affinity to androgens.21, 22 SHBG was

originally thought to primarily regulate the amount of sex hormones that are bioavailable

to the cells. However, recent epidemiological studies consistently show that low SHBG

concentrations are strongly associated with the development of insulin resistance, type 2

diabetes, cardiovascular diseases (CVD), hormone-dependent cancers, as well as hip

fractures, either indirectly by modulating the biologic effects of testosterone or exert more

direct effects through its own SHBG receptor.14, 21, 23-29 Mendelian randomization analyses

using single nucleotide polymorphisms within or near the SHBG gene as instrumental

variables for blood SHBG concentrations also provided supporting evidence to the causal

relationship between SHBG and the risk of type 2 diabetes.14, 30

Given the role SHBG may play in the etiologies of type 2 diabetes, CVD, and

hormone-dependent cancers, investigating the determinants of blood SHBG concentrations

is of great importance. In addition to several common variants identified within or near the

SHBG gene,14, 30 lifestyle factors, especially dietary factor, may have a direct effect on

circulating concentrations of free endogenous sex hormones through the regulation of

SHBG concentrations.31 Physical activity, regular coffee consumption, as well as weight

5

loss by exercise and/or caloric restriction has been found to increase SHBG concentrations

in postmenopausal women.32, 33 Emerging evidence also shows that different types of

dietary carbohydrates may have heterogeneous associations with SHBG concentrations.22,

34-37 In a dietary intervention study, lower serum SHBG concentrations were observed

among participants on a conventional high glycemic load diet, while the SHBG

concentrations increased among those on a high-protein low glycemic load diet.34 Fiber

intake was found to be positively correlated with SHBG concentrations in a previous study

in men,35 but another study failed to observe a similar correlation in postmenopausal

women.36 Moreover, although an inverse association between monosaccharides and SHBG

production was reported previously in transgenic mice and hepatic cell models,37 a more

recent study found that sweets intake may positively correlate with SHBG concentrations,

although the result was not significant.22 Increasing attention has been attracted to the

impact of dietary factors on circulating SHBG concentrations. Nevertheless, studies

examining the effect of quality and quantity of dietary carbohydrates on SHBG are still

scarce and inconclusive, and very few have studied foods that are abundant in

carbohydrates. Therefore, we conducted a comprehensive examination of the relations

between various measures of dietary carbohydrates and concentrations of circulating

SHBG among a subsample from the large-scale national Women’s Health Initiative

study.38

6

2.2 Methods

2.2.1 Study Subjects

The Women's Health Initiative (WHI) is a long-term national health study that focused on

strategies for preventing heart diseases, breast and colorectal cancer, and osteoporotic

fractures in postmenopausal women. The original WHI study included 161,808

postmenopausal women enrolled between 1993 and 1998 in two major parts: a partial

factorial randomized Clinical Trial (CT) and an Observational Study (OS); both were

conducted at 40 Clinical Centers nationwide. The CT enrolled 68,132 postmenopausal

women between the ages of 50 to 79 into trials testing three prevention strategies. The OS

examined the relationship between lifestyle, environmental, medical and molecular risk

factors and specific measures of health or disease outcomes. This component involved

tracking the medical history and health habits of 93,676 women not participating in the CT.

The current analysis included an initial total of 13,955 unique participants from either the

WHI-CT or the WHI-OS whose blood samples from baseline had been measured for serum

SHBG in the following ancillary studies: AS90 (400 hip fracture cases and 400 controls),

AS110 (385 coronary heart disease cases and 385 controls), AS167 (311 breast cancer

cases and 592 controls), AS238 (700 type 2 diabetes cases and 1,400 controls), BA7 (422

venous thromboembolism, 534 stroke, 753 CHD, 204 spine fracture, 830 non-hip-or-spine

fracture cases, and 1,576 controls), BA9 (1,132 fracture cases and 1,132 controls), BA21

(400 colorectal cancer cases and 800 controls), W5 (300 controls), W9 (750 hip fracture

cases and 750 controls), W10 (755 breast cancer cases and 755 controls), and W18 (240

controls). Participants were excluded if they had implausible total energy intake (< 600 or >

5000 kcal/day) as determined by the food frequency questionnaire (n = 711 excluded), self-

7

reported diabetes at baseline (n = 1072 further excluded), or if they had missing information

in important covariates such as age (no missing), race/ethnicity (n = 27), body mass index

(n = 74), smoking status (n = 170), physical activity (n = 768), and hormone therapy use

(n = 6) (since there were overlap, this step excluded 1013 observations). No missing in

dietary measurements were observed after applying the above exclusion criteria.

2.2.2 Measurement of Serum SHBG Concentrations

For each study participant, blood was collected at the baseline visit after at least a 12-hour

fast and then stored at −80 °C to -70 °C. Samples used for the hormone measurements were

taken from these baseline specimens. The serum SHBG concentrations were measured

using an electrochemiluminescence immunoassay (ECL) in AS238, a solid-phase, two-site

chemiluminescent immunoassay (solid-phase, two-site CIA) in AS90, AS110, AS167,

BA7, BA9, BA21, W9, W10, and W18), or an immunoradiometric assay (IRMA) in W5.

The inter-assay coefficients of variation ranged between 3.7% and 17.7%.22

2.2.3 Dietary Measurements

The methods of data collection and validation have been reported previously.38, 39

Participants completed at baseline a 122-item standardized food frequency questionnaire

(FFQ) developed for the WHI to estimate average daily dietary intake over the past 3

months.40 The FFQ was based on instruments used in the WHI feasibility studies and the

original National Cancer Institute/Block FFQ.41-43 The dietary database, linked to the

University of Minnesota Nutrition Coordinating Center Nutrition Data System for

Research (Nutrition Coordinating Center, Minneapolis, MN, USA), is based on the U.S.

8

Department of Agriculture standard reference releases and manufacturer information.44

The detailed description of the methods used to calculate GI and GL values can be found

elsewhere.45 In summary, GI values based on food consumption or expert judgment were

assigned to each food items that contained at least five grams of carbohydrates, and then

for each FFQ line item GL values were calculated by multiplying GI by intake frequencies

and portion sizes. Both total carbohydrates and available carbohydrates (total

carbohydrates minus total fiber) were used to calculate GI and GL values. In addition,

dietary intakes of total carbohydrates, total sugar, and total fiber were included in the

analyses in separate models. As a secondary analysis, the associations of different

carbohydrates abundant food items (daily servings of white bread, dark bread, rice grains

and noodles, potato, cereal, fruits, beans, sugar sweetened beverages, pasta, and whole

grains) with serum SHBG concentrations were also examined. The potato variable included

French fries, potato salad, sweet potatoes and yams, and other potato/cassava/yucca. The

cereal variable included cold and cooked cereals. The beans variable included green or

string beans, English peas, refried beans, all other beans, and bean soup. The sugar

sweetened beverage variable included regular soft drinks (not diet), orange or grapefruit

juice, other fruit juice, and fruit drinks. The pasta variable included macaroni and cheese,

lasagna, or noodles with a cream sauce, spaghetti with meat sauce, and spaghetti with

tomato sauce. All other variables were pre-calculated by the WHI. This FFQ has

demonstrated reasonably good validity as a measurement of dietary intake compared with

24-hour dietary recalls and food records.40

9

2.2.4 Statistical Analysis

Baseline characteristics were summarized according to SHBG quartiles. Continuous

variables were presented as means ± standard deviations, and categorical variables were

presented in percentages. The statistical significances of differences among SHBG

quartiles were tested by ANOVA for continuous variables and by chi-square test for

categorical variables. Individual level data from different ancillary studies were pooled and

analyzed to assess the associations between measures of carbohydrate intakes and natural-

log-transformed SHBG concentrations using linear multivariable models. Dietary

carbohydrate intakes were each analyzed in quartiles. We adjusted for potential

confounding factors including total energy intake, total carbohydrates intake (except when

total carbohydrates intake was exposure of interest), ancillary study indicators, case/control

status in each ancillary study, age (continuous), ethnicity, body mass index (BMI,

continuous), cigarette smoking (never, past, or current), alcohol consumption (continuous),

physical activity (metabolic equivalent of tasks per week, continuous), and hormone

therapy use (never, past, or current user of unopposed estrogen and/or estrogen plus

progesterone). From this model, we calculated the adjusted geometric means of SHBG

concentrations for each quartile of the carbohydrate of interest by exponentiating the

estimated mean log SHBG concentrations evaluated at the mean of each continuous

variable and averaged over the groups of each categorical variable in the model. We also

performed a linear trend analysis for each measure of carbohydrate by assigning the median

of each quartile to each observation and using the resulting continuous variable as the

independent variable in the model. In order to address multiple testing issue, Benjamini

and Hochberg’s procedure for controlling the false discovery rate (FDR) was performed

10

with the results of the trend analysis.46 Measures of carbohydrate intake with q-value below

0.05 were considered statistically significant, which corresponded to less than one false

positive result per 20 comparisons.

Sensitivity analyses were performed by (a) restricting to only controls of each

ancillary study, (b) using linear mixed effects models to pool the estimates from each

ancillary study, (c) additionally adjust for dietary total protein, fruits and vegetables intake,

and baseline history of cardiovascular diseases and cancer, and (d) restricting to studies

that measured SHBG using solid-phase, two-site CIA (nine out of eleven studies).

Furthermore, since SHBG concentrations has been inversely linked to the risk of type 2

diabetes previously,14, 21, 24 we hypothesized that dietary carbohydrates may influence the

risk of type 2 diabetes through affecting serum SHBG concentrations. Thus, we performed

exploratory mediation analyses within the type 2 diabetes case control study in our sample

(AS238, n = 1,586 after applying exclusion criteria), with quartiles of dietary carbohydrate

measures as the exposure (contrasting the highest and the lowest quartile), SHBG

concentrations as the mediator, and case control status as the outcome.47-49 The average

causal mediation effects, average direct effects, the proportion mediated, and their

respective confidence intervals were quantified. All statistical analyses were conducted

using R version 3.3.1 (The R Foundation for Statistical Computing, Vienna, Austria).50

2.3 Results

We included a total of 11,159 postmenopausal women in the current analysis, with a

median serum SHBG concentration of 47.3 nmol/L and interquartile range of 33.0 – 68.8

nmol/L. This group of participants were on average 65.3 years old (SD = 7.5), had an

11

average BMI of 28.6 kg/m2 (SD = 6.1), an average total energy intake of 1,617.5 kcal per

day (SD = 660.6), and an average total carbohydrates intake of 201.4 grams per day (SD =

80.9). Sixty-eight percent of them were white and 8.6% were smokers at study baseline.

When comparing women across SHBG quartiles, those within higher quartiles of SHBG

concentrations tended to be older, less likely to be white and more likely to be black or

Asian, had lower BMI, and were more likely to be current smokers. Women with higher

concentrations of serum SHBG also had lower intake of total energy, total carbohydrates,

protein, and fat, as well as sugar, GL, GI, and fruits and vegetables, while they had similar

intake of fiber compared to women with lower concentrations of serum SHBG (Table

2.5.1).

Since the original continuous SHBG variable was skewed to the right, we

performed natural logarithm transformation and used the log-transformed SHBG variable

as the dependent variable in the subsequent linear regression analyses. After adjusting for

total energy intake, total carbohydrates (except when total carbohydrates was the exposure

of interest), age, race, BMI, smoking status, physical activity, alcohol intake, hormone

therapy use, ancillary study indicator, and case-control status in each ancillary study, higher

dietary GL based on both total carbohydrates and available carbohydrates were

significantly associated with lower concentrations of serum SHBG (P-value for trend =

0.01 and 0.002, q-value = 0.04 and 0.01, respectively). Women within the lowest quartile

of dietary GL based on available carbohydrates had an adjusted average SHBG of 56.7

nmol/L (95% CI: 54.6, 58.9), while women within the highest quartile of dietary GL had

an adjusted average SHBG of 52.6 nmol/L (95% CI: 50.5, 54.7), and results were very

similar for GL based on total carbohydrates. Similar trend was observed for dietary GI

12

based on total and available carbohydrates (P-value for trend = 0.02 and 0.04, q-value =

0.07 and 0.10, respectively) and dietary sugar intake (P-value for trend < 0.001, q-value =

0.01), for which the lowest intake quartile had an average SHBG concentration of 56.2

nmol/L (95% CI: 54.3, 58.1), while the highest intake quartile had an average of 52.1

nmol/L (95% CI: 50.2, 54.1). We also found a positive trend for fiber, where higher intake

of fiber was associated with higher SHBG concentrations (P-value for trend = 0.01, q-value

= 0.04). No significant findings were observed for total carbohydrates intake (Table 2.5.2).

For analyses regarding carbohydrate-abundant food items, we found a significant

inverse relationship between quartiles of sugar sweetened beverages and circulating SHBG

concentrations (P-value for trend < 0.001, q-value < 0.001). The lowest intake quartile

corresponded to an adjusted average SHBG concentration of 56.7 nmol/L (95% CI: 55.0,

58.6), while the highest intake quartile corresponded to an average of 52.7 nmol/L (95%

CI: 51.1, 54.4). Interestingly, we also observed borderline inverse associations for potatoes

intake (P-value for trend = 0.07, q-value = 0.16) and beans intake (P-value for trend = 0.10,

q-value = 0.19). Other food items were not significantly associated with circulating SHBG

concentrations (Table 2.5.3).

When restricting to just controls from ancillary studies, this subgroup included

5,457 participants. Without correction for multiple comparison, results were similar to

those obtained from the primary analyses in the whole group (Table 2.5.4 and 2.5.5).

Dietary GL based on total and available carbohydrates, fiber, sugar, and sugar sweetened

beverages remained significantly associated with SHBG concentrations after the FDR

procedure, and the effect sizes were also similar. The discrepancy was that the P-values for

trend for dietary GI based on total and available carbohydrates were significant before and

13

after the FDR procedure in the controls, while in the whole sample they were only

significant before multiple testing correction. From the second sensitivity analyses we

performed, the linear mixed effects models where ancillary study indicators were treated

as random effects yielded very similar results to the primary analyses (data not shown). In

the third sensitivity analysis, when additionally adjusting for dietary total protein, fruits

and vegetables intake, and baseline history of cardiovascular diseases and cancer, the

results were virtually the same with those from the primary analysis (for details see

Supplementary Table 2.6.1). From the fourth sensitivity analysis where we restricted to

studies that measured SHBG using solid-phase, two-site CIA (nine out of eleven studies),

the estimated adjusted means of SHBG concentrations were overall lower than what we

estimated in the original analytic sample, but the results regarding the trends across

quartiles of dietary intake and their significance were similar to the results of our primary

analysis (for details see Supplementary Table 2.6.2). The exploratory mediation analyses

did not find significant average causal mediation effects or average direct effects, possibly

due to the fact that one single case control study was not powered enough to detect

significant mediation effects (for details see Supplementary Table 2.6.3).

2.4 Discussion

In this cross-sectional analysis of 11,159 non-diabetic postmenopausal women enrolled in

the WHI, positive associations with SHBG concentrations were observed for total dietary

fiber intake. Total dietary sugar intake and dietary GL based on total and available

carbohydrates were observed to be significantly associated with reduced serum

concentrations of SHBG, before and after correction for multiple comparisons. Dietary GI

14

based on total and available carbohydrates were also associated with lower levels of SHBG

before multiple testing correction. In addition, significant association between sugar-

sweetened beverages and decreased concentrations of serum SHBG was demonstrated

based on analyses regarding carbohydrate-abundant food items, corroborating our results

for total sugar intake.

Our finding of the inverse relationship between dietary GL based on total and

available carbohydrates and serum SHBG concentrations was consistent with results from

a previous dietary intervention trial where SHBG was considerably lowered after a 12-

week high GL diet compared to the baseline, and also significantly decreased compared to

those on a low GL diet,34 although another study contrasting low GI and high GI diet did

not find significant difference in SHBG after an 8-week intervention.51 This somewhat

contradicted the fact that in our analysis dietary GI based on total or available

carbohydrates was significantly associated with SHBG concentrations, although only

before correction for multiple comparisons. Mechanistically, it has been suggested that

high GL/GI diet induced greater insulin production, and insulin could act as an inhibitor of

hepatic synthesis of SHBG.37, 52-54 Dietary sugar, which is usually high in glycemic index

and glycemic load, was found to be significantly and inversely associated with serum

SHBG. This result was in line with biological evidence from human-SHBG-transgenic

mice and human hepatic cells, where glucose and fructose reduced human SHBG

production by hepatocytes via the downregulation of hepatocyte nuclear factor-4α, which

was independent of the actions of insulin.37 Sugar also likely contributed to the relation

between dietary GL/GI and SHBG concentrations. The null relationship we observed

between total carbohydrates intake and SHBG was consistent with observations from

15

another previous study.35 Collectively, these results suggested that the quality of dietary

carbohydrates might be of greater importance than quantity in affecting circulating SHBG

levels, given that our analyses with respect to GL and GI were adjusted for the total amount

of dietary carbohydrates.

Dietary fiber, which did not contribute to GL or GI based on available

carbohydrates, was positively associated with serum SHBG in our analysis. Previous

findings from the WHI Dietary Modification trial associated a low fat dietary pattern with

significant reduction in SHBG after 1 year of intervention, which were thought to be

partially contributed by the concurrent increase in fiber intake as well as weight loss.55

Even though an early study found no correlation between dietary fiber and SHBG,36 a more

recent investigation with regression modelling did reveal significant positive relations

between the two, accounting for age, BMI, and other covariates.35 The biomarker of lignans

intake was also found to be positively related to SHBG levels, albeit the null association

between dietary fiber and SHBG in the same study.56 The mechanism by which fiber intake

may be a controlling factor on SHBG is not yet well-understood, but it is possible that it

acts through modulating glucagon-like peptide-1 and insulin secretion.57

We also systematically examined the primary carbohydrate-abundant food items

that might be responsible for the dietary effects on SHBG concentrations in these data. A

significant inverse relationship between sugar sweetened beverages and circulating SHBG

concentrations was discovered, and this mirrored our findings in the relationship between

total sugar intake and SHBG concentrations. The significant inverse associations between

total sugar intakes, sugar sweetened beverages intakes and plasma SHBG concentrations

illustrated the detrimental effect of excessive sugar consumptions, which indicated that

16

cutting down sugar intake may be an important intervention to increase SHBG

concentrations. We also identified 2 categories of foods, potatoes and beans, which were

borderline significantly inversely associated with SHBG concentrations, albeit the

complete null association after correction for multiple comparison or in controls only.

Physiological studies show that most potatoes are of high GI regardless of cooking method,

which over the long term may increase the risk of obesity and chronic diseases such as type

2 diabetes and cardiovascular diseases.58

The cross-sectional nature of the current investigation raises concern over the

temporality of the associations that we observed. However, the WHI food frequency

questionnaire inquired dietary intakes during the period of 3 months prior to study baseline,

while the blood samples from which SHBG was measured were taken at baseline. Thus,

the temporality between dietary carbohydrates intake and serum SHBG concentrations can

be established to a certain extent. Another limitation of this study is that measurements of

SHBG from different ancillary studies of the WHI were included in order to boost the

power in detecting the associations, especially with the moderately large number of testings

in the current analysis. Heterogeneity among these studies in measuring serum SHBG, as

well as different criteria for choosing study participants may introduce bias into our results.

We attempted to address this issue by including both the indicators of ancillary studies and

indicators of case or control status in each ancillary study in our statistical modelling. To

evaluate the extent of bias, we also performed sensitivity analyses in controls only, as well

as using linear mixed effects models, and our results were largely robust to the different

methods used. Adiposity could also potentially influence both dietary intake and blood

SHBG concentrations.22 Although we controlled for BMI in our analyses, we could not

17

rule out the possibility of residual confounding as it is not a perfect measure of adiposity.

Finally, while a large national sample of postmenopausal women participated in the WHI

studies, which was broader and more representative than those in studies based on samples

of convenience, the findings presented here can only be generalized to postmenopausal

women, which is another limitation of this investigation.

In conclusion, our study found that dietary fiber intake, sugar intake, and GL/GI

based on total and available carbohydrates have significant associations with serum SHBG

concentrations, thus supporting a role of diet in influencing blood levels of SHBG, which

is in turn an important protective factor probably associated with insulin resistance, type 2

diabetes, cardiovascular disease, and hormone-dependent cancers. Further studies are

needed to better elucidate the biological mechanisms underlying the associations between

dietary carbohydrates and circulating SHBG concentrations, and mediation analyses with

sufficient power are also needed to evaluate whether these possible effects of dietary

carbohydrates on SHBG extend to the ultimate cardiometabolic disease risk, which will

contribute to a better understanding of the mechanisms of action underlying the effect of

diets, particularly of high GL/GL diets rich in refined carbohydrates.

18

2.5 Tables

Table 2.5.1 Baseline characteristics by quartiles of untransformed plasma SHBG concentrations in a subpopulation of the

postmenopausal women from the Women’s Health Initiative (n = 11,159)

SHBG

P-value Q1 Q2 Q3 Q4

Number of participants 2801 2794 2776 2788

Median (nmol/L) 25.7 40.0 56.3 92.0

(Interquartile range) (21.0, 29.7) (36.5, 43.5) (51.7, 62.0) (78.0, 125.0)

Age (years) [mean (SD)] 63.7 (7.1) 65.6 (7.3) 66.5 (7.5) 65.6 (7.8) <.001

BMI (kg/m2) [mean (SD)] 31.6 (5.9) 29.5 (6.0) 27.3 (5.5) 26.0 (5.5) <.001

Race/ethnicity [n (%)]

<.001

White 1930 (68.9) 1997 (71.5) 1972 (71.0) 1734 (62.2)

Black/African American 502 (17.9) 456 (16.3) 443 (16.0) 592 (21.2)

Hispanic/Latino 198 (7.1) 190 (6.8) 183 (6.6) 227 (8.1)

Asian/Pacific Islander 124 (4.4) 102 (3.7) 137 (4.9) 196 (7.0)

Other 47 (1.7) 49 (1.8) 41 (1.5) 39 (1.4)

Smoking status [n (%)]

0.02 Never 1444 (51.6) 1495 (53.5) 1457 (52.5) 1471 (52.8)

Former 1144 (40.8) 1067 (38.2) 1076 (38.8) 1040 (37.3)

Current 213 (7.6) 232 (8.3) 243 (8.8) 277 (9.9)

19

Total energy (kcal/d) [mean (SD)] 1715.1 (694.6) 1625.0 (665.1) 1573.7 (627.5) 1555.4 (641.5) <.001

Alcohol intake (g/d) [mean (SD)] 4.5 (11.0) 5.1 (11.6) 5.1 (10.5) 4.4 (11.2) <.001

GL (total CHO) [mean (SD)] 110.4 (46.8) 104.9 (43.5) 104.0 (43.7) 103.2 (43.3) <.001

GL (available CHO) [mean (SD)] 102.9 (44.3) 97.5 (41.0) 96.5 (41.1) 95.8 (40.8) <.001

GI (total CHO) [mean (SD)] 52.6 (3.7) 52.1 (3.9) 52.0 (3.9) 52.1 (3.8) <.001

GI (available CHO) [mean (SD)] 53.0 (3.7) 52.5 (3.9) 52.4 (3.9) 52.5 (3.7) <.001

Total fiber (g/d) [mean (SD)] 15.4 (6.9) 15.4 (6.7) 15.7 (7.0) 15.7 (6.9) 0.13

Total sugar (g/d) [mean (SD)] 103.2 (51.0) 98.7 (47.1) 97.9 (45.6) 96.9 (46.2) <.001

Total carbohydrates (g/d) [mean (SD)] 208.9 (84.6) 200.4 (79.7) 199.0 (79.4) 197.4 (79.4) <.001

Total protein (g/d) [mean (SD)] 70.7 (30.8) 67.1 (30.2) 64.6 (27.6) 62.9 (282.2) <.001

Total fat (g/d) [mean (SD)] 65.9 (35.6) 60.9 (34.0) 57.1 (31.8) 57.1 (32.2) <.001

Fruits & vegetables (servings/d) [mean (SD)] 3.9 (2.2) 3.9 (2.1) 4.1 (2.2) 4.1 (2.2) <.001

Abbreviations: SHBG: sex hormone-binding globulin; Q: quartile; SD: standard deviation; BMI: body mass index; GL: glycemic load; CHO:

carbohydrates; GI: glycemic index.

20



carbohydrates

Adjusted

mean* 95% CI Ptrend q-value

GL (total CHO)

0.01 0.04

Q1 56.4 (54.3, 58.6)

Q2 55.6 (53.8, 57.4)

Q3 53.0 (51.3, 54.7)

Q4 52.9 (50.8, 55.1)

GL (available CHO)

0.002 0.01

Q1 56.7 (54.6, 58.9)

Q2 55.5 (53.7, 57.3)

Q3 53.1 (51.4, 54.8)

Q4 52.6 (50.5, 54.7)

GI (total CHO)

0.02 0.07

Q1 55.5 (53.7, 57.3)

Q2 54.8 (53.1, 56.5)

Q3 53.9 (52.3, 55.7)

Q4 53.9 (52.2, 55.6)

GI (available CHO)

0.04 0.10

Q1 55.3 (53.5, 57.1)

Q2 54.9 (53.2, 56.6)

Q3 54.1 (52.4, 55.8)

Q4 53.9 (52.2, 55.6)

Total fiber (g/d)

0.01 0.04

Q1 53.6 (51.9, 55.4)

Q2 54.1 (52.5, 55.9)

Q3 54.0 (52.3, 55.8)

Q4 56.4 (54.5, 58.5)

Total sugar (g/d)

<.001 0.01

Q1 56.2 (54.3, 58.1)

Q2 55.1 (53.3, 56.8)

Q3 54.1 (52.4, 55.9)

Q4 52.1 (50.2, 54.1)

21

Total carbohydrates (g/d)

Q1

(53.0, 56.8)

0.37 0.42

54.9

Q2 55.2 (53.4, 57.0)

Q3 53.4 (51.7, 55.1)

Q4 54.3 (52.3, 56.3)

Abbreviations: SHBG: sex hormone-binding globulin; CI: confidence interval; Ptrend: P-value for

trend; Q: quartile; GL: glycemic load; CHO: carbohydrates; GI: glycemic index.

* Adjusted means were computed by exponentiating the least squares means of estimated log-

transformed SHBG concentrations from model including the exposure of interest, total

carbohydrates (except when total carbohydrates was the exposure of interest), total energy intake,

age, race, BMI, smoking status, physical activity, alcohol intake, hormone therapy use, ancillary

study indicator, and case-control status in each ancillary study. Linear mixed model with ancillary

study indicator as random effect yielded very similar results.

22


intake of carbohydrates abundant food items

Adjusted mean* 95% CI Ptrend q-value

White bread (servings/d)

0.13 0.20

Q1 54.5 (52.8, 56.2)

Q2 55.2 (53.5, 56.9)

Q3 54.5 (52.8, 56.2)

Q4 53.7 (52.0, 55.5)

Dark bread (servings/d)

0.64 0.68

Q1 54.0 (52.3, 55.6)

Q2 54.8 (53.1, 56.6)

Q3 54.8 (53.1, 56.6)

Q4 54.7 (52.9, 56.5)

Rice, grains and plain noodles

(servings/d)

0.12 0.20

Q1 54.0 (52.3, 55.6)

Q2 54.5 (52.7, 56.4)

Q3 53.9 (52.1, 55.7)

Q4 55.2 (53.6, 57.0)

Potato (servings/d)

0.07 0.16

Q1 55.4 (53.7, 57.1)

Q2 54.6 (52.9, 56.3)

Q3 53.7 (52.1, 55.5)

Q4 53.9 (52.2, 55.7)

Cereal (servings/d)

0.92 0.92

Q1 54.5 (52.9, 56.2)

Q2 54.8 (53.1, 56.6)

Q3 54.0 (52.3, 55.7)

Q4 54.8 (53.0, 56.7)

Fruits (servings/d)

0.21 0.27

Q1 54.3 (52.6, 56.0)

Q2 54.1 (52.4, 55.9)

Q3 54.4 (52.7, 56.2)

Q4 55.3 (53.5, 57.1)

23

Beans (servings/d)

0.10 0.19

Q1 54.8 (53.1, 56.5)

Q2 54.8 (53.1, 56.5)

Q3 54.9 (53.2, 56.7)

Q4 53.6 (51.9, 55.3)

Sugar sweetened beverages

(servings/d)

<.001 <.001

Q1 56.7 (55.0, 58.6)

Q2 55.2 (53.4, 56.9)

Q3 54.0 (52.3, 55.7)

Q4 52.7 (51.1, 54.4)

Pasta (servings/d)

0.20 0.27

Q1 55.0 (53.3, 56.8)

Q2 53.7 (52.1, 55.4)

Q3 53.8 (52.1, 55.5)

Q4 55.5 (53.7, 57.4)

Whole grains (servings/d)

0.22 0.27

Q1 54.2 (52.5, 56.0)

Q2 54.4 (52.7, 56.1)

Q3 54.2 (52.5, 55.9)

Q4 55.3 (53.5, 57.1)


trend; Q: quartile.



carbohydrates, total energy intake, age, race, BMI, smoking status, physical activity, alcohol intake,

hormone therapy use, ancillary study indicator, and case-control status in each ancillary study.

Linear mixed model with ancillary study indicator as random effect yielded very similar results.

24



carbohydrates in controls of the ancillary studies


GL (total CHO)

0.03 0.06

Q1 55.9 (53.0, 58.9)

Q2 55.8 (53.3, 58.4)

Q3 52.8 (50.5, 55.2)

Q4 51.7 (48.8, 54.7)

GL (available CHO)

0.01 0.05

Q1 56.3 (53.4, 59.3)

Q2 55.5 (53.0, 58.1)

Q3 52.8 (50.5, 55.2)

Q4 51.5 (48.6, 54.5)

GI (total CHO)

0.001 0.01

Q1 55.7 (53.3, 58.2)

Q2 54.7 (52.4, 57.1)

Q3 53.3 (51.0, 55.7)

Q4 52.5 (50.2, 54.8)

GI (available CHO)

0.01 0.02

Q1 55.5 (53.1, 58.0)

Q2 54.5 (52.2, 56.9)

Q3 53.5 (51.2, 55.9)

Q4 52.6 (50.4, 55.0)

Total fiber (g/d)

<.001 0.004

Q1 52.0 (49.7, 54.5)

Q2 53.5 (51.1, 55.9)

Q3 53.6 (51.3, 56.0)

Q4 57.5 (54.8, 60.4)

Total sugar (g/d)

0.02 0.05

Q1 55.5 (52.9, 58.2)

Q2 54.8 (52.4, 57.3)

Q3 53.9 (51.5, 56.3)

Q4 51.5 (48.9, 54.3)

25


Q1

0.05 0.11

55.0 (52.4, 57.7)

Q2 55.7 (53.2, 58.2)

Q3 53.2 (50.9, 55.6)

Q4 52.2 (49.6, 55.1)






age, race, BMI, smoking status, physical activity, alcohol intake, hormone therapy use, and

ancillary study indicator. Linear mixed model with ancillary study indicator as random effect

yielded very similar results.

26


intake of carbohydrates abundant food items in controls of the ancillary studies



0.91 0.92

Q1 54.3 (52.0, 56.7)

Q2 53.5 (51.2, 55.9)

Q3 54.3 (52.0, 56.7)

Q4 53.8 (51.4, 56.4)


0.68 0.77

Q1 53.7 (51.5, 56.1)

Q2 54.0 (51.7, 56.5)

Q3 54.2 (51.8, 56.6)

Q4 54.3 (51.8, 56.9)


(servings/d)

0.10 0.20

Q1 53.0 (50.7, 55.4)

Q2 54.0 (51.5, 56.7)

Q3 53.8 (51.5, 56.1)

Q4 55.3 (52.7, 58.1)

Potato (servings/d)

0.13 0.20

Q1 55.5 (53.2, 58.0)

Q2 52.9 (50.7, 55.3)

Q3 54.1 (51.7, 56.5)

Q4 53.1 (50.7, 55.6)

Cereal (servings/d)

0.53 0.75

Q1 53.7 (51.4, 56.0)

Q2 54.4 (52.1, 56.8)

Q3 52.9 (50.6, 55.4)

Q4 55.0 (52.5, 57.6)

Fruits (servings/d)

0.13 0.20

Q1 53.7 (51.4, 56.2)

Q2 52.6 (50.4, 55.0)

Q3 55.1 (52.7, 57.6)

Q4 54.9 (52.5, 57.5)

27

Beans (servings/d)

0.68 0.77

Q1 53.3 (51.0, 55.7)

Q2 54.1 (51.8, 56.5)

Q3 54.8 (52.4, 57.3)

Q4 53.9 (51.6, 56.4)

Sugar sweetened beverages

(servings/d)

<.001 0.003

Q1 56.6 (54.1, 59.1)

Q2 54.6 (52.3, 57.1)

Q3 53.3 (51.0, 55.7)

Q4 52.2 (49.9, 54.6)

Pasta (servings/d)

0.66 0.77

Q1 54.6 (52.3, 57.1)

Q2 53.6 (51.4, 56.0)

Q3 53.1 (50.8, 55.5)

Q4 54.8 (52.4, 57.4)


0.92 0.92

Q1 54.5 (52.1, 56.9)

Q2 53.2 (51.0, 55.6)

Q3 54.1 (51.8, 56.5)

Q4 54.2 (51.8, 56.7)


trend; Q: quartile.



carbohydrates, total energy intake, age, race, BMI, smoking status, physical activity, alcohol intake,

hormone therapy use, and ancillary study indicator. Linear mixed model with ancillary study

indicator as random effect yielded very similar results.

28

2.6 Supplementary Tables

Supplementary Table 2.6.1 Results of sensitivity analysis with additional adjustment

for total protein, total fruit and vegetable consumption, and baseline history of

cardiovascular diseases and cancer: adjusted means of serum SHBG concentrations

according to quartiles of dietary glycemic load, glycemic index, and intakes of fiber,

sugar, total carbohydrates, and intake of carbohydrates abundant food items

Adjusted


GL (total CHO)

0.005 0.01

Q1 56.4 (54.3, 58.6)

Q2 55.6 (53.8, 57.4)

Q3 52.9 (51.3, 54.7)

Q4 52.6 (50.5, 54.9)

GL (available CHO)

0.001 0.01

Q1 56.8 (54.6, 59.0)

Q2 55.5 (53.7, 57.4)

Q3 53.0 (51.3, 54.8)

Q4 52.3 (50.2, 54.5)

GI (total CHO)

0.005 0.01

Q1 55.7 (53.9, 57.5)

Q2 54.8 (53.1, 56.6)

Q3 53.7 (52.0, 55.5)

Q4 53.4 (51.6, 55.2)

GI (available CHO)

0.01 0.02

Q1 55.5 (53.7, 57.3)

Q2 54.8 (53.1, 56.6)

Q3 53.9 (52.2, 55.7)

Q4 53.4 (51.7, 55.2)

Total fiber (g/d)

0.003 0.01

Q1 53.1 (51.3, 55.0)

Q2 53.8 (52.1, 55.6)

Q3 54.0 (52.3, 55.7)

Q4 56.9 (54.8, 59.0)

29

Total sugar (g/d)

<.001 0.004

Q1 56.3 (54.5, 58.3)

Q2 55.1 (53.3, 56.9)

Q3 54.0 (52.3, 55.8)

Q4 51.9 (50.0, 54.0)


Q1

(53.5, 57.5)

0.05 0.10

55.5

Q2 55.5 (53.7, 57.3)

Q3 53.2 (51.5, 55.0)

Q4 53.4 (51.4, 55.5)


0.16 0.20

Q1 54.4 (52.7, 56.1)

Q2 55.1 (53.3, 56.8)

Q3 54.5 (52.8, 56.3)

Q4 53.7 (51.9, 55.5)


0.45 0.48

Q1 53.9 (52.2, 55.6)

Q2 54.7 (53.0, 56.4)

Q3 54.7 (53.0, 56.5)

Q4 54.8 (53.0, 56.6)


(servings/d)

0.08 0.14

Q1 53.8 (52.2, 55.6)

Q2 54.3 (52.5, 56.3)

Q3 53.7 (52.0, 55.5)

Q4 55.3 (53.6, 57.0)

Potato (servings/d)

0.08 0.14

Q1 55.3 (53.5, 57.0)

Q2 54.5 (52.8, 56.3)

Q3 53.9 (52.2, 55.6)

Q4 53.8 (52.0, 55.6)

Cereal (servings/d)

0.77 0.77

Q1 54.4 (52.8, 56.1)

Q2 54.7 (53.0, 56.5)

Q3 53.8 (52.2, 55.6)

Q4 54.9 (53.1, 56.8)

30

Fruits (servings/d)

0.36 0.40

Q1 54.2 (52.3, 56.1)

Q2 54.0 (52.3, 55.8)

Q3 54.3 (52.6, 56.1)

Q4 55.2 (53.2, 57.3)

Beans (servings/d)

0.11 0.17

Q1 54.7 (52.9, 56.4)

Q2 54.8 (53.1, 56.6)

Q3 54.9 (53.2, 56.7)

Q4 53.5 (51.8, 55.3)

Sugar sweetened beverages (servings/d)

<.001 <.001

Q1 56.9 (55.1, 58.8)

Q2 55.2 (53.5, 57.0)

Q3 53.8 (52.1, 55.6)

Q4 52.4 (50.7, 54.1)

Pasta (servings/d)

0.14 0.20

Q1 55.0 (53.3, 56.7)

Q2 53.6 (51.9, 55.3)

Q3 53.7 (52.0, 55.5)

Q4 55.6 (53.8, 57.5)


0.17 0.20

Q1 54.2 (52.4, 56.0)

Q2 54.2 (52.5, 55.9)

Q3 54.1 (52.4, 55.8)

Q4 55.3 (53.5, 57.2)







study indicator, case-control status in each ancillary study, and total protein, total fruit and

vegetable consumption, and baseline history of cardiovascular diseases and cancer.

31

Supplementary Table 2.6.2 Results of sensitivity analysis restricting to nine studies

using the chemiluminescent immunoassay method (n = 9,328): adjusted means of

serum SHBG concentrations according to quartiles of dietary glycemic load, glycemic

index, and intakes of fiber, sugar, total carbohydrates, and intake of carbohydrates

abundant food items

Adjusted


GL (total CHO)

0.02 0.05

Q1 49.1 (47.1, 51.2)

Q2 48.2 (46.5, 50.0)

Q3 46.3 (44.7, 48.0)

Q4 46.1 (44.1, 48.2)

GL (available CHO)

0.004 0.02

Q1 49.4 (47.4, 51.5)

Q2 48.3 (46.6, 50.1)

Q3 46.3 (44.7, 48.0)

Q4 45.7 (43.7, 47.8)

GI (total CHO)

0.03 0.06

Q1 48.4 (46.7, 50.2)

Q2 47.6 (46.0, 49.3)

Q3 47.0 (45.4, 48.7)

Q4 47.0 (45.4, 48.7)

GI (available CHO)

0.07 0.10

Q1 48.2 (46.5, 49.9)

Q2 47.8 (46.1, 49.5)

Q3 47.0 (45.4, 48.7)

Q4 47.1 (45.5, 48.8)

Total fiber (g/d)

<.001 0.002

Q1 46.3 (44.6, 48.0)

Q2 47.0 (45.4, 48.7)

Q3 47.2 (45.5, 48.9)

Q4 50.0 (48.1, 52.0)

32

Total sugar (g/d)

<.001 0.002

Q1 49.2 (47.3, 51.0)

Q2 48.1 (46.4, 49.8)

Q3 47.1 (45.4, 48.8)

Q4 45.1 (43.3, 47.1)


Q1

(46.0, 49.6)

0.69 0.69

47.8

Q2 47.6 (46.0, 49.4)

Q3 46.8 (45.2, 48.5)

Q4 47.6 (45.7, 49.5)


0.04 0.08

Q1 47.7 (46.1, 49.4)

Q2 48.3 (46.6, 50.1)

Q3 47.2 (45.5, 48.9)

Q4 46.7 (45.0, 48.4)


0.64 0.68

Q1 47.3 (45.7, 48.9)

Q2 47.4 (45.7, 49.2)

Q3 47.9 (46.2, 49.6)

Q4 47.6 (45.9, 49.4)


(servings/d)

0.01 0.05

Q1 46.8 (45.2, 48.4)

Q2 47.0 (45.2, 48.8)

Q3 47.0 (45.4, 48.8)

Q4 48.5 (46.8, 50.2)

Potato (servings/d)

0.20 0.29

Q1 48.2 (46.6, 49.9)

Q2 47.2 (45.6, 48.9)

Q3 47.1 (45.4, 48.8)

Q4 47.1 (45.4, 48.9)

Cereal (servings/d)

0.64 0.68

Q1 47.6 (46.0, 49.3)

Q2 47.7 (46.1, 49.5)

Q3 46.9 (45.2, 48.6)

Q4 47.5 (45.8, 49.3)

33

Fruits (servings/d)

0.35 0.43

Q1 47.2 (45.6, 48.9)

Q2 47.3 (45.6, 49.0)

Q3 47.7 (46.1, 49.5)

Q4 47.9 (46.1, 49.7)

Beans (servings/d)

0.33 0.43

Q1 47.5 (45.8, 49.2)

Q2 47.7 (46.1, 49.5)

Q3 48.0 (46.3, 49.7)

Q4 46.9 (45.3, 48.6)

Sugar sweetened beverages (servings/d)

<.001 <.001

Q1 49.5 (47.8, 51.3)

Q2 48.0 (46.3, 49.8)

Q3 47.2 (45.6, 48.9)

Q4 46.0 (44.5, 47.7)

Pasta (servings/d)

0.03 0.06

Q1 47.6 (46.0, 49.2)

Q2 46.9 (45.2, 48.6)

Q3 47.0 (45.3, 48.7)

Q4 48.8 (47.0, 50.6)


0.06 0.10

Q1 47.4 (45.7, 49.1)

Q2 47.1 (45.4, 48.8)

Q3 46.8 (45.2, 48.5)

Q4 48.8 (47.0, 50.6)







study indicator, and case-control status in each ancillary study.

34

Supplementary Table 2.6.3 Results of exploratory mediation analysis in AS238

diabetes case-control study contrasting the highest to lowest quartile of dietary

glycemic load, glycemic index, and intakes of fiber, sugar, total carbohydrates, and

intake of carbohydrates abundant food items (all data presented in log odds ratio with

point estimate (95% confidence interval)

ACME ADE TE

GL (total CHO) 0.01 (-0.02, 0.05) -0.05 (-0.18, 0.06) -0.04 (-0.17, 0.09)

GL (available CHO) 0.02 (-0.008. 0.06) -0.02 (-0.15, 0.10) 0.002 (-0.13, 0.12)

GI (total CHO) 0.01 (-0.004, 0.03) -0.01 (-0.07, 0.05) -0.002 (-0.06, 0.06)

GI (available CHO) 0.01 (-0.002, 0.03) -0.005 (-0.06, 0.05) 0.007 (-0.05, 0.07)

Total fiber (g/d) -0.005 (-0.03, 0.02) 0.02 (-0.07, 0.11) 0.02 (-0.07. 0.11)

Total sugar (g/d) 0.008 (-0.02. 0.03) -0.004 (-0.11, 0.10) 0.004 (-0.10, 0.11)

Total carbohydrates (g/d) 0.03 (-0.01, 0.06) 0.002 (-0.12, 0.14) -0.03 (-0.10, 0.18)

White bread (servings/d) -0.007 (-0.03, 0.01) -0.01 (-0.07, 0.06) -0.02 (-0.07, 0.06)

Dark bread (servings/d) -0.003 (-0.02, 0.02) -0.009 (-0.07, 0.07) -0.01 (-0.07, 0.07)

Rice, grains and plain

noodles (servings/d) 0.01 (-0.009, 0.03) 0.03 (-0.04, 0.10) 0.04 (-0.03, 0.12)

Potato (servings/d) 0.01 (-0.004, 0.03) 0.05 (-0.02, 0.11) 0.06 (-0.009, 0.12)

Cereal (servings/d) -0.001 (-0.02, 0.02) -0.03 (-0.09, 0.04) -0.02 (-0.09, 0.05)

Fruits (servings/d) -0.007 (-0.02, 0.01) 0.01 (-0.06, 0.08) 0.003 (-0.07, 0.07)

Beans (servings/d) 0.01 (-0.007, 0.03) -0.02 (-0.08, 0.04) -0.01 (-0.07, 0.05)

Sugar sweetened

beverages (servings/d) 0.01 (-0.006, 0.03) 0.04 (-0.03. 0.11) 0.05 (-0.03, 0.11)

Pasta (servings/d) 0.01 (-0.007, 0.03) 0.02 (-0.04, 0.08) 0.03 (-0.04, 0.09)

Whole grains (servings/d) 0.004 (-0.01, 0.03) 0.02 (-0.04, 0.08) 0.02 (-0.04, 0.09)

Abbreviations: ACME: average causal mediation effects; ADE: average direct effects; TE: total

effects; GL: glycemic load; CHO: carbohydrates; GI: glycemic index.

35

Chapter 3 Comparisons of Multiple Sources of Starchy Foods, and Their Estimated

Substitution Effects on the Risk of Type 2 Diabetes and Atherosclerotic

Cardiovascular Disease in Postmenopausal Women

3.1 Introduction

Among major sources of carbohydrates in our diet, pasta has long been of interest due to

its different glycemic effects on the human body compared to other carbohydrate-dense

foods. Traditionally an Italian dish, pasta is now widely recognized as a staple food by

many cultures around the world. It has been demonstrated since the 1980s that in diabetic

patients, blood glucose response was remarkably reduced after ingesting spaghetti

compared to white bread,59, 60 potato,60, 61, or rice.61 Many characteristics of pasta have been

studied in relation to its glycemic response. Notably, the structure (i.e. viscosity, particle

size, and shape) of pasta appears to be more important in determining its glycemic response

than the types of cereal used in its production.62-64 It has also been found that consumption

of pasta produced a lower postprandial insulin response than consumption of white bread

in healthy subjects.62 Given the same amount, pasta appears to have lower glycemic index

(GI) as well as glycemic load (GL) compared to other major sources of carbohydrates.

Dietary glycemic index and glycemic load have been positively linked with the risk

of a number of cardiometabolic diseases, including metabolic syndrome,65 type 2

diabetes,66-71 coronary heart disease (CHD),7, 71-75 stroke76, 77 and stroke mortality,78 as well

as risk factors including triglycerides,12, 79 lowered high-density-lipoprotein cholesterol,12,

36

79, 80 low-density-lipoprotein cholesterol,79, 81 and high-sensitivity C-reactive protein.13 The

International Carbohydrate Quality Consortium reached consensus in 2015 that there was

convincing evidence that low GI/GL diets reduce the risk of type 2 diabetes and CHD, and

GI represents another characteristic of carbohydrate-foods apart from fiber and whole grain

content.18 Since pasta has been shown to produce lower glycemic response, it is then

natural to hypothesize that consumption of pasta may have beneficial effects on the long

term risk of diabetes and atherosclerotic cardiovascular diseases (ASCVD, including CHD

and stroke), given the same total carbohydrate consumption. The primary objective of the

current analysis was to evaluate the prospective relationship between dietary intake of pasta,

as measured by the baseline food frequency questionnaire (FFQ), and the risk of

developing diabetes, CHD, stroke, and ASCVD in post-menopausal women from the

national Women’s Health Initiative.38 To our knowledge, no other large scale, long term

prospective cohort studies have specifically evaluated these relationships.

3.2 Methods

3.2.1 Study Population

The Women’s Health Initiative (WHI) recruited a total of 161,808 postmenopausal women

aged 50 to 79 years at 40 clinical centers across the United States (US) between 1993 and

1998, including a cohort of 93,676 women in a prospective Observational Study (OS) and

68,133 women in one or more of the following 3 clinical trials (CTs): the hormone therapy

(HT) trial, the calcium and vitamin D (CaD) trial, and the dietary modification (DM) trial.

We analyzed baseline data from participants of the OS, and the HT and CaD trials of the

WHI, for whom valid information was obtained from a validated 122-item food frequency

37

questionnaire (FFQ).40 Participants of the DM trial were excluded due to potential major

alterations in dietary behavior after baseline resulting from their participation in the trial.

Additional exclusion criteria included: implausible total energy intake (< 600 or > 5000

kcal/day); prevalent diseases including diabetes, cardiovascular diseases, and cancer that

may alter dietary behaviors; measurements not available for outcomes of interest (incident

diabetes, CHD, or stroke); measurements not available for important covariates such as

race and body mass index (BMI); being underweight (BMI < 18.5 kg/m2) which may reflect

underlying medical conditions.

3.2.2 Measurement of Outcomes

Incident diabetes was assessed via questionnaires at enrollment and each annual visit.

Participants were asked if “a doctor prescribed for the first time any of the following pills

or treatments: pills for diabetes or insulin shots for diabetes” since their last medical update.

Those who responded “yes” were considered having been diagnosed with diabetes. Since

these were all post-menopausal women, newly diagnosed diabetes cases were most likely

type 2 diabetes cases. Women who self-reported diabetes at baseline were excluded from

the current analysis.

Incident coronary heart disease (CHD) was defined as the first occurrence of

clinical MI, definite silent MI or a death due to definite or possible CHD. Clinical MI and

death were adjudicated for CT and OS participants during the core WHI study (until 2005)

and Extension Study I (Ext1, until 2010). Definite Silent MI was adjudicated for CT

participants during the core WHI study.

38

Incident stroke was defined as the first occurrence of stroke or a death due to

cerebrovascular event. Stroke was adjudicated for CT and OS participants through Ext1.

We further examined the concept of incident atherosclerotic cardiovascular diseases

(ASCVD) as an outcome, which encompassed incident cases of CHD and stroke as defined

above, according to the 2013 ACC/AHA Guideline on the Assessment of Cardiovascular

Risk.82, 83

3.2.3 Measurement of Pasta

Participants were asked on the baseline FFQ how often they consumed each of the

following forms of pasta during the past 3 months: macaroni and cheese/lasagna/noodles

with a cream sauce, spaghetti or other noodles with meat sauce, and spaghetti or other

noodles with tomato sauce (and no meat), in frequency of medium servings. Nine

frequency options were given, including: “never or less than 1 per month”, “1 per month”,

“2-3 per month”, “1 per week”, “2 per week”, “3-4 per week”, “5-6 per week”, “1 per day”

and “2+ per day”. Participants also had the choices of “small”, “medium”, and “large” for

portion size per serving. The midpoint of the nine categories were used to compute the

semi-continuous variables for these three forms of pasta, in number of medium servings

per day. The sum of the three were used as a measure of total pasta intake. The residual

method was then used to partially control for measurement error in the FFQ, where total

pasta intake was linearly regressed on total energy intake, and the residuals added with

mean total pasta intake were taken as a measure of pasta intake uncorrelated with total

energy intake.

39

In addition to residual total pasta intake, we also constructed two standardized

measures of pasta intake. The ratio of pasta to glycemic load (GL) was computed by

dividing total pasta intake with total dietary glycemic load (GL) and then multiplied by

100. The ratio of pasta to total energy intake was computed similarly, by dividing total

pasta intake with dietary total energy intake and then multiplied by 1000. The respective

scaling was done in order to obtain more interpretable measurements. These two ratio

measures were analyzed in parallel with the original measure of residual total pasta intake.

3.2.4 Statistical Analysis

Baseline characteristics of participants included in the current analysis were described

according to quartiles of raw total pasta intake instead of residual total pasta intake for

easier interpretation. Means and standard deviations (SDs) were generated as descriptive

statistics for each continuous variable, while frequencies and percentages were generated

for each categorical variable. The statistical significance of differences across pasta intake

quartiles were tested by analysis of variance (ANOVA) for continuous variables and by

chi-square test for categorical variables.

Cox proportional hazards models were used to evaluate the association between

residual total pasta intake and the risk of diabetes, CHD, stroke and ASCVD in terms of

hazard ratios (HRs) and associated 95% confidence intervals (CIs), with study baseline as

the origin of analysis and time-to-event or time-to-censoring as defined hereafter. For

diabetes, follow-up durations were calculated as the interval between baseline and the

earliest of any of the following: (1) date of annual medical history update when new

diabetes was reported, (2) date of last data collection from the main study if the participant

40

did not enter the Extension Study, (3) date of last data collection from the Extension Study,

or (4) date of reported death. For CHD, follow-up durations were calculated as the interval

between baseline and the earliest of any of the following: (1) date of first confirmed CHD

diagnosis, (2) date of last data collection from the core WHI study if participant did not

enter the Extension Study, (3) date of last data collection from the Extension Study, or (4)

date of reported death. For stroke, follow-up durations were calculated as the interval

between baseline and the earliest of any of the following: (1) date of first confirmed stroke

diagnosis, (2) date of last data collection from the core WHI study if participant did not

enter the Extension Study, (3) date of last data collection from the Extension Study, or (4)

date of reported death. For ASCVD, follow-up durations were calculated as the interval

between baseline and the earliest of any of the following: (1) date of first confirmed CHD

or stroke diagnosis, whichever came first, (2) date of last data collection from the core WHI

study if participant did not enter the Extension Study, (3) date of last data collection from

the Extension Study, or (4) date of reported death. Residual pasta intake was analyzed both

in quartiles and as continuous variables. We also tested for linear trend after assigning the

median of each quartile to the participants, and entering this variable as continuous into the

models. The proportional hazards assumption was tested statistically.84 We adjusted for the

following potential confounding factors in Model 1: study group indicator (OS/HT/CaD),

age (continuous), race/ethnicity (Caucasian, African American, Hispanic, Asian/Pacific

Islander, or other), and region (Northeast, South, Midwest, or West of the US). In Model

2, we additionally adjusted for BMI (continuous), total energy intake, and percent energy

intake from carbohydrates (each gram of carbohydrate is considered to have 4 calories). In

Model 3 and the final model, we further adjusted for cigarette smoking (never, past, or

41

current), alcohol consumption (continuous), physical activity (in metabolic equivalent

hours (METs) per week, continuous), and healthy eating index (HEI 2005,85 continuous),

and the respective family history of each outcome (diabetes, CHD, stroke, or ASCVD).

Raw total pasta intake, pasta to GL ratio, and pasta to total energy ratio were analyzed in

similar procedures as the residual total pasta intake, in both continuous form and quartiles,

and then tested for linear trend.

As a sensitivity analysis, we included only pasta meals with spaghetti as the main

carbohydrates source in the exposure variable, as macaroni and cheese had been observed

to have higher glycemic index than spaghetti.63 The residual method was again used to

generate a residual spaghetti intake variable, which was analyzed similarly as residual total

pasta intake. We also statistically tested for the substitutional effects of replacing pasta for

the same amount of white bread, potato, or rice (in terms of medium servings per day),

which were all measured with the same FFQ. To model such effects, firstly the residual

variables for white bread, potato, and rice intake were created using the residual method;

then the sum of residual pasta and residual white bread, potato, or rice were computed

respectively; finally, residual total pasta intake and the sum variable were entered into the

Model 3 at the same time, so that the effect estimates for the residual pasta variable

represented the estimated log(HR) for replacing one medium serving of white bread, potato,

or rice by pasta, respectively. All statistical analyses were conducted using R version 3.2.3

(The R Foundation for Statistical Computing, Vienna, Austria).52

42

3.3 Results

A total of 84,555 participants of the WHI OS, HT, and CaD were included in the final

analytical sample. Among these postmenopausal women, the median raw intake of pasta

was 0.15 servings per day, or equivalently 1.04 servings per week. The interquartile range

was 0.08 to 0.26 servings per day, or equivalently 0.54 to 1.84 servings per week. This

group of participants were on average 63.3 years old (SD = 7.3), had an average BMI of

27.3 kg/m2 (SD = 5.6), an average total energy intake of 1,576.2 kcal per day (SD = 598.8),

and an average total carbohydrates intake of 203.7 grams per day (SD = 78.0), which

translated into an average of 52.4% energy from carbohydrates (SD = 9.6). Eighty-five

percent of them were white and 6.8% were smokers at study baseline. Thirty percent had a

family history of diabetes, while 51.8% had a family history of CHD, 36.1% had a family

of history of stroke, and 65.2% had a family history of ASCVD.

Those in the higher quartiles of raw total pasta intake were on average younger but

with higher BMI, more likely to be white, less likely to be never-smokers, and more likely

to have family history of diabetes and CHD, but not stroke or ASCVD. In terms of dietary

intakes, women in the highest quartiles of pasta intake had on average higher total energy

intake, higher total carbohydrates intake, higher dietary glycemic index and glycemic load,

but interestingly somewhat lower percent energy from carbohydrates compared to the

lower quartiles. Those in higher quartiles of raw total pasta intake also had relatively higher

whole grains intake, higher alcohol intake, but lower dietary quality as measured by HEI

2005 (Table 3.5.1).

Results from the Cox proportional hazards models were summarized as follows by

outcomes of interest. For the diabetes outcome, residual total pasta intake was generally

43

not associated with increased or decreased risk across quartiles for postmenopausal women,

after adjusting for age, race, region, study indicators, BMI, total energy intake, percent

energy from carbohydrates, smoking status, alcohol consumption, physical activity, HEI

2005, and family history of diabetes (Model 3), and Models 1-3 had similar results (Table

3.5.2). Specifically from Model 3, compared to those in the lowest quartile of residual pasta

intake, women in the second, third, and highest intake quartiles had essentially no change

in risk (HR = 0.97, 95% CI: 0.92, 1.03, HR = 1.00, 95% CI: 0.94, 1.05, and HR = 1.02, 95%

CI: 0.96, 1.07, respectively), and the linear trend was also not significant (P-value for trend

= 0.328). Results were largely similar when examining quartiles of pasta-to-GL ratio or

pasta-to-energy ratio.

For the CHD outcome, overall increased intake of pasta appeared to be associated

with a somewhat decreased risk of developing CHD, especially when comparing women

in the highest and lowest quartile of residual total pasta intake (Table 3.5.3). Specifically

from Model 3, compared to women in the lowest quartile of residual pasta intake, those in

the second and third quartile had no change in the risk of developing CHD (HR = 0.99, 95%

CI: 0.90, 1.09, and HR = 1.02, 95% CI: 0.93, 1.12, respectively), and women in the highest

intake quartile had an estimated 9% reduction in risk (HR = 0.91, 95% CI: 0.83, 1.00),

while holding constant age, race, region, study indicators, BMI, total energy intake, percent

energy from carbohydrates, smoking status, alcohol consumption, physical activity, HEI

2005, and family history of diabetes, and the linear trend was borderline significant (P-

value for trend = 0.058). Results were again largely similar across Models 1-3 as well as

when examining quartiles of pasta-to-GL ratio or pasta-to-energy ratio.

44

Women within the highest intake quartile of pasta appeared to have a significantly

reduced risk of developing stroke compared to those in the lowest intake quartile of pasta

(Table 3.5.4). Results from Model 3 suggested that women in the highest quartile of

residual total pasta intake had a 16% reduction in the risk of developing stroke compared

to women in the lowest quartile (HR = 0.84, 95% CI: 0.75, 0.93), while those in the second

and third quartile had virtually no change in risk (HR = 0.97, 95% CI: 0.88, 1.08, and HR

= 1.00, 95% CI: 0.91, 1.11, respectively), adjusting for age, race, region, study indicators,

BMI, total energy intake, percent energy from carbohydrates, smoking status, alcohol

consumption, physical activity, HEI 2005, and family history of diabetes. Testing for linear

trend showed a significant inverse association (P-value for trend = 0.001), and results were

highly consistent across Models 1-3 as well as when examining quartiles of pasta-to-GL

ratio or pasta-to-energy ratio.

With ASCVD being a composite outcome of CHD and stroke, participants within

the highest intake quartile of residual total pasta had an estimated 11% decreased risk of

developing ASCVD (HR = 0.89, 95% CI: 0.83, 0.96) compared to those in the lowest

intake quartile in Model 3, while those in the second and third quartile had no change in

risk (HR = 0.99, 95% CI: 0.92, 1.06, and HR = 1.03, 95% CI: 0.96, 1.11, respectively),

with age, race, region, study indicators, BMI, total energy intake, percent energy from

carbohydrates, smoking status, alcohol consumption, physical activity, HEI 2005, and

family history of diabetes being constant (Table 3.5.5). Significant inverse trends were also

observed, with P-value for trend = 0.002. Results were again highly consistent across

Models 1-3 as well as when examining quartiles of pasta-to-GL ratio or pasta-to-energy

ratio.

45

We also estimated the effects of one medium serving/day increase in pasta intake

variables on each disease of interest by entering the continuous variables of pasta intake

into the models as exposure instead of quartiles. These analyses had similar results to the

counterparts with quartiles as exposure, but the effect sizes seemed generally larger in

magnitude (Supplementary Table 3.6.1).

Since results were largely similar across Models 1-3, we presented here results from

Model 3 for the sensitivity analyses. When restricting to only spaghetti meals, the results

were consistent with what we observed from the primary analyses for each of the outcomes,

in terms of effect estimates, confidence intervals, as well as P-value from trend analysis

(Table 3.5.6). When statistically modeling the substitution effects, replacing potato with

pasta was associated with a significant decreased risk of stroke (HR = 0.66, 95% CI: 0.52,

0.84) and ASCVD (HR = 0.78, 95% CI: 0.66, 0.92), but it was not associated with the risk

of diabetes, while there was a suggestive reduction in the risk of CHD (HR = 0.83, 95%

CI: 0.67, 1.02). Statistically substituting pasta for white bread was also associated with a

significant reduction in the risk of stroke (HR = 0.73, 95% CI: 0.59, 0.92), and a borderline

significant reduction in the risk of ASCVD (HR = 0.88, 95% CI: 0.76, 1.01), while the risk

was unchanged for diabetes or CHD. Statistically substituting pasta for rice was not

associated with altered risk of developing diabetes, CHD, stroke, or ASCVD (Table 3.5.7).

3.4 Discussion

In this prospective analysis of 84,555 postmenopausal women enrolled in the WHI, we

observed a significant association between higher intakes of pasta and long-term risk of

developing stroke and ASCVD, and a suggestive association between higher intakes of

46

pasta and long-term risk of developing CHD, while no significant relation was observed

between pasta intake and risk of developing diabetes. When we statistically estimated the

substitutional effects of replacing other types of common starchy foods with pasta, we also

found that substituting pasta for potato or white bread could potentially be associated with

lower risk of stroke and ASCVD, and again a suggestive association was observed between

substituting pasta for potato and lowered risk of CHD.

To our knowledge, our finding of the inverse relations between pasta intake and

risk of stroke and ASCVD was the first time that such associations were reported.

Measurements of body weight and adiposity are recognized as important risk factors for

cardiometabolic diseases, and we considered our findings in the context of previous

evidence relating pasta intake to these measurements. A cross-sectional analysis in two

Italian cohorts, the Moli-sani study and the Italian Nutrition and Health Survey, which

included over 20,000 participants, demonstrated that higher pasta intake was associated

with better adherence to Mediterranean diet,86 a dietary pattern which has a demonstrated

cardiovascular benefit.87, 88 The authors also found that higher pasta intake was associated

with lower BMI, waist circumference, waist-to-hip ratio, and lower prevalence of being

overweight and obese, which was independent of adherence to Mediterranean diet and total

energy intake. Similar cross-sectional associations between higher pasta intake and lower

BMI were observed in US adults as part of the International Study of Macro-

/Micronutrients and Blood Pressure (INTERMAP) study.89 From another analysis of the

same study where 17 population samples in four countries (China, Japan, UK, US) were

included, it was also found that individuals with low risk of developing cardiovascular

disease (CVD), as classified by favorable profile of CVD risk factors, reported higher

47

intake of pasta, among other food items such as fruits, vegetables, and fish. A recent

systematic review and meta-analysis of the effect of pasta in the context of low-GI dietary

patterns on body weight and markers of adiposity concluded that pasta intake did not

adversely affect adiposity and reduced body weight and BMI compared with higher-GI

dietary patterns, while cautioning the interpretation of their results due to unexplained

inconsistency among the trials included in the meta-analysis.90 Another previous meta-

analysis confirmed that pasta meals had significantly lower postprandial glucose response,

in other words lower GI, than white bread or potato meals,91 which was consistent with our

findings that statistically substituting pasta for white bread or potato were associated with

reduced risk of stroke and ASCVD, while keeping constant total carbohydrates intake

among other potential confounding factors, although we did not obverse potential benefits

associated with diabetes risk. Collectively, available observational evidence so far

suggested that higher intake of pasta may have a beneficial influence on cardiometabolic

risk profile or even ASCVD risk, in the context of unchanged total carbohydrate intake or

a healthy dietary pattern, which warrants further investigation into these relations in other

prospective cohort studies.

Our study was the first to report potential inverse associations between pasta intake

and long-term incident cardiometabolic disease risk in a prospective cohort study with

large sample size. However, several limitations should be considered when interpreting the

results. Total intake of pasta in this study were measured by summing the semi-quantitative

intake frequencies of pasta meals which had other commonly used ingredients such as

cheese or tomato. We attempted to account for other aspects of participants’ diet by

adjusting for an overall measurement of dietary quality, due to the fact that adjusting for

48

individual components may cause over-adjustment and result in non-parsimonious models.

Another limitation was that we were not able to obtain information on the types of pasta

that was consumed by the participants, in terms of regular or whole grain or legume-based,

which prevented us from conducting sensitivity analyses for these subtypes which may

have different GI or GL. We attempted a sensitivity analysis where macaroni and cheese

intake was excluded from total pasta intake due to its potential higher GI, and the results

were similar to those of the primary analysis. Measurement error associated with FFQ is

always a potential source of bias, but the WHI FFQ was validated against 24-hour dietary

recalls and food records40, and we employed the residual method in primary and sensitivity

analyses which could partially correct for this issue. While ASCVD outcomes were

adjudicated based on medical records review in the WHI, diabetes was self-reported and

under-reporting or undiagnosed diabetes may have partially contributed to the null findings

with respect to this outcome. Finally, the findings presented here stemmed from a cohort

of postmenopausal women and the overall consumption of pasta was relatively low

(average of 1 serving /week), and further analysis within cohorts of men or both sex with

higher levels of consumption should be conducted to evaluate the prospective associations

between pasta intake and risk of cardiometabolic disease.

In conclusion, our study demonstrated that consumption of pasta did not have

adverse effects on long-term diabetes, CHD, stroke, or ASCVD risk, and may even be

associated with reduced risk of developing stroke and ASCVD in postmenopausal women,

in the context of unchanged total carbohydrates and total energy intake. In addition,

substituting pasta for other commonly consumed starchy foods such as potato white bread

may also present beneficial effects on future stroke and ASCVD risk. Pasta may represent

49

a feasible and easy-to-implement method of dietary modification, and further analyses

should be conducted in existing long-term prospective cohort studies with cardiometabolic

disease endpoints, for which long-term randomized controlled clinical are often costly and

not feasible.

50

3.5 Tables

Table 3.5.1 Baseline characteristics of WHI participants in analytic sample by quartiles of total pasta intake (n = 84,555)

Raw Total Pasta Intake Quartiles

P-value Q1 Q2 Q3 Q4

Number of participants (n) 23,232 19,124 21,868 20,331

Range (servings/d) (0.00, 0.08) (0.08, 0.15) (0.15, 0.26) (0.26, 4.09)

Mean (SD) (servings/d) 0.04 (0.03) 0.12 (0.02) 0.20 (0.03) 0.49 (0.26)

Median (servings/d) 0.04 0.11 0.20 0.43

IQR (servings/d) (0.02, 0.07) (0.10, 0.14) (0.16, 0.23) (0.32, 0.57)

OS or HT or CaD participant, n*

OS 18,990 15,655 17,731 16,621

HT 4,242 3,469 4,137 3,710

CaD 2,371 2,049 2,425 2,192

Age at baseline, years, mean (SD) 64.5 (7.3) 63.7 (7.2) 62.9 (7.2) 61.8 (7.2) <.001

BMI, kg/m2, mean (SD) 26.8 (5.4) 27.1 (5.5) 27.3 (5.6) 27.9 (6.0) <.001

Self-reported ethnicity, n(%) <.001

White, non-Hispanic 18,299 (78.8) 16,400 (85.8) 19,292 (88.2) 18,084 (88.9)

African American 2,162 (9.3) 1,293 (6.8) 1,250 (5.7) 1,093 (5.4)

Hispanic/Latino 1,271 (5.5) 689 (3.6) 633 (2.9) 617 (3.0)

Asian/Pacific Islander 1,098 (4.7) 507 (2.7) 397 (1.8) 253 (1.2)

Other 402 (1.7) 235 (1.2) 296 (1.4) 284 (1.4)

Cigarette smoking status, n(%) <.001

51

Never smoker 12,196 (53.1) 9,814 (51.9) 10,892 (50.4) 9,742 (48.5)

Past smoker 9,207 (40.1) 7,832 (41.4) 9,240 (42.7) 8,986 (44.8)

Current smoker 1,555 (6.8) 1,266 (6.7) 1,498 (6.9) 1,346 (6.7)

Alcohol consumption, g/day, mean (SD) 5.3 (11.3) 5.8 (11.2) 6.2 (11.5) 6.2 (11.8) <.001

Physical activity, METs/wk, mean (SD) 14.2 (14.9) 13.4 (13.8) 13.5 (13.9) 13.8 (14.7) <.001

Total energy intake, kcal/day, mean (SD) 1314.7 (486.7) 1472.0 (507.0) 1633.2 (545.7) 1911.7 (673.7) <.001

Total carb intake, g/day, mean (SD) 174.2 (68.1) 190.5 (68.5) 207.5 (71.0) 245.6 (85.2) <.001

Percent energy from carb, mean (SD) 53.6 (10.5) 52.3 (9.4) 51.4 (9.1) 52.2 (9.2) <.001

Glycemic load, mean (SD) 89.7 (36.9) 98.8 (37.1) 108.2 (38.6) 128.4 (46.4) <.001

Glycemic index, mean (SD) 51.3 (4.4) 51.8 (3.8) 52.0 (3.5) 52.2 (3.3) <.001

Whole grains, ounce equivalent/day, mean (SD) 1.1 (1.0) 1.2 (1.0) 1.3 (1.0) 1.4 (1.1) <.001

AHEI 2005, mean (SD) 69.8 (10.8) 69.4 (10.6) 68.5 (10.5) 68.1 (10.4) <.001

Family history of diabetes, yes, n(%) 6,964 (30.1) 5,606 (29.4) 6,379 (29.3) 6,375 (31.5) <.001

Family history of CHD, yes, n(%) 11,097 (50.6) 9,392 (51.7) 10,855 (52.1) 10,263 (53.0) <.001

Family history of stroke, yes, n(%) 8,432 (36.5) 6,929 (36.4) 7,846 (36.0) 7,179 (35.5) 0.101

Family history of ASCVD, yes, n(%) 14,717 (64.8) 12,245 (65.3) 14,062 (65.5) 13,044 (65.3) 0.135

Incident diabetes, yes, n(%) 2,744 (11.8) 2,286 (12.0) 2,792 (12.8) 2,770 (13.6) <.001

Incident CHD, yes, n(%) 1,189 (5.1) 993 (5.2) 986 (4.5) 864 (4.2) <.001

Incident stroke, yes, n(%) 983 (4.2) 755 (3.9) 811 (3.7) 624 (3.1) <.001

Incident ASCVD, yes, n(%) 2,046 (8.8) 1,649 (8.6) 1,709 (7.8) 1,420 (7.0) <.001

Abbreviations: Q, quartile; IQR, inter quartile range; SD, standard deviation.

* The WHI-HT and CaD trials had overlapping participants, while those in the WHI-OS were not in these trials by design, so the numbers may not

add up to the total in each quartile.

52

Table 3.5.2 Estimates of relative risk and 95% confidence intervals (CIs) of diabetes according to quartiles of pasta intake

Residual Total Pasta Intake Quartiles P-value

for trend Q1 Q2 Q3 Q4

Number of cases 2,689 2,524 2,618 2,761

Person-years 290374.8 296741.9 302417.7 305846.2

HR (95% CI)

Model 1a 1.00 0.92 (0.87, 0.97) 0.95 (0.90, 1.00) 1.01 (0.96, 1.07) 0.167

Model 2b 1.00 0.97 (0.92, 1.03) 1.00 (0.95, 1.06) 1.03 (0.97, 1.09) 0.148

Model 3c 1.00 0.97 (0.92, 1.03) 1.00 (0.94, 1.05) 1.02 (0.96, 1.07) 0.328

Pasta/100GL Ratio Quartiles P-value


HR (95% CI)

Model 1a 1.00 0.98 (0.93, 1.04) 1.03 (0.98, 1.09) 1.08 (0.98, 1.09) 0.001

Model 2b 1.00 0.95 (0.90, 1.01) 0.97 (0.92, 1.03) 1.00 (0.95, 1.06) 0.539

Model 3c 1.00 0.95 (0.90, 1.01) 0.97 (0.92, 1.03) 1.00 (0.94, 1.06) 0.590

Pasta/1000kcal Energy Ratio Quartiles P-value


HR (95% CI)

Model 1a 1.00 0.95 (0.90, 1.01) 1.00 (0.95, 1.05) 1.04 (0.98, 1.10) 0.032

Model 2b 1.00 0.94 (0.89, 0.99) 0.98 (0.93, 1.03) 1.01 (0.96, 1.07) 0.173

Model 3c 1.00 0.94 (0.89, 0.99) 0.97 (0.92, 1.03) 1.00 (0.95, 1.06) 0.368 a Model 1 adjusted for age, race, region, and study indicators

b Model 2 adjusted for age, race, region, study indicators, BMI, total energy intake, and percent energy from carbohydrates

c Model 3 adjusted for age, race, region, study indicators, BMI, total energy intake, percent energy from carbohydrates, smoking status, alcohol

consumption, physical activity, healthy eating index (HEI) 2005, and family history of diabetes.

53

Table 3.5.3 Estimates of relative risk and 95% confidence intervals (CIs) of CHD according to quartiles of pasta intake



Number of cases 1,063 1,061 1,047 861

Person-years 303901.7 309179.7 315659.0 320927.2

HR (95% CI)

Model 1a 1.00 1.01 (0.93, 1.10) 1.03 (0.95, 1.12) 0.95 (0.86, 1.03) 0.207

Model 2b 1.00 1.02 (0.93, 1.11) 1.04 (0.95, 1.14) 0.94 (0.86, 1.03) 0.152

Model 3c 1.00 0.99 (0.90, 1.09) 1.02 (0.93, 1.12) 0.91 (0.83, 1.00) 0.058



HR (95% CI)

Model 1a 1.00 1.02 (0.94, 1.12) 1.00 (0.92, 1.09) 1.00 (0.92, 1.10) 0.868

Model 2b 1.00 1.01 (0.92, 1.10) 0.96 (0.88, 1.05) 0.94 (0.86, 1.03) 0.132

Model 3c 1.00 0.99 (0.90, 1.08) 0.97 (0.89, 1.07) 0.94 (0.85, 1.03) 0.165



HR (95% CI)

Model 1a 1.00 1.03 (0.95, 1.12) 0.98 (0.90, 1.07) 0.95 (0.87, 1.04) 0.170

Model 2b 1.00 1.02 (0.94, 1.11) 0.97 (0.89, 1.06) 0.93 (0.89, 1.06) 0.070




consumption, physical activity, healthy eating index (HEI) 2005, and family history of CHD.

54

Table 3.5.4 Estimates of relative risk and 95% confidence intervals (CIs) of stroke according to quartiles of pasta intake



Number of cases 889 842 822 620

Person-years 304807.1 310167.8 316120.4 322181.7

HR (95% CI)

Model 1a 1.00 0.96 (0.87, 1.06) 0.98 (0.89, 1.08) 0.83 (0.75, 0.92) 0.001

Model 2b 1.00 0.98 (0.89, 1.08) 1.00 (0.90, 1.10) 0.84 (0.75, 0.93) 0.001

Model 3c 1.00 0.97 (0.88, 1.08) 1.00 (0.91, 1.11) 0.84 (0.75, 0.93) 0.001



HR (95% CI)

Model 1a 1.00 0.98 (0.89, 1.08) 0.99 (0.90, 1.09) 0.87 (0.79, 0.97) 0.009

Model 2b 1.00 0.96 (0.88, 1.06) 0.96 (0.87, 1.06) 0.84 (0.76, 0.93) 0.001

Model 3c 1.00 0.97 (0.88, 1.07) 0.97 (0.88, 1.08) 0.85 (0.77, 0.95) 0.003



HR (95% CI)

Model 1a 1.00 0.91 (0.83, 1.01) 1.00 (0.91, 1.09) 0.84 (0.76, 0.93) 0.003

Model 2b 1.00 0.91 (0.82, 1.00) 0.99 (0.90, 1.09) 0.83 (0.75, 0.92) 0.003

Model 3c 1.00 0.91 (0.82, 1.00) 1.00 (0.91. 1.10) 0.84 (0.75, 0.93) 0.005 a Model 1 adjusted for age, race, region, and study indicators



consumption, physical activity, healthy eating index (HEI) 2005, and family history of stroke.

55

Table 3.5.5 Estimates of relative risk and 95% confidence intervals (CIs) of ASCVD according to quartiles of pasta intake

Pasta Intake Quartiles P-value


Number of cases 1,842 1,790 1,775 1,417

Person-years 300015.7 305683.9 311598.6 318080.3

HR (95% CI)

Model 1a 1.00 0.98 (0.92, 1.05) 1.02 (0.95, 1.08) 0.90 (0.84, 0.97) 0.005

Model 2b 1.00 1.00 (0.93, 1.07) 1.03 (0.96, 1.10) 0.90 (0.84, 0.97) 0.003

Model 3c 1.00 0.99 (0.92, 1.06) 1.03 (0.96, 1.11) 0.89 (0.83, 0.96) 0.002



HR (95% CI)

Model 1a 1.00 1.01 (0.94, 1.08) 1.01 (0.95, 1.08) 0.95 (0.89, 1.02) 0.111

Model 2b 1.00 0.99 (0.93, 1.06) 0.98 (0.92, 1.05) 0.90 (0.84, 0.97) 0.003

Model 3c 1.00 1.00 (0.93, 1.07) 0.99 (0.93, 1.07) 0.91 (0.85, 0.98) 0.008



HR (95% CI)

Model 1a 1.00 0.98 (0.92, 1.04) 1.00 (0.94, 1.07) 0.91 (0.85, 0.98) 0.010

Model 2b 1.00 0.97 (0.91, 1.03) 0.99 (0.93, 1.06) 0.90 (0.84, 0.96) 0.004




consumption, physical activity, healthy eating index (HEI) 2005, and family history of CHD and stroke.

56

Table 3.5.6 Estimates of relative risk and 95% confidence intervals (CIs) of diseases of interest according to quartiles of residual

spaghetti intake from Model 3*

Spaghetti Intake Quartiles P-value


Diabetes 1.00 0.98 (0.93, 1.04) 1.01 (0.95, 1.07) 1.02 (0.97, 1.08) 0.223

CHD 1.00 0.99 (0.90, 1.08) 0.98 (0.89, 1.08) 0.92 (0.84, 1.02) 0.084

Stroke 1.00 0.98 (0.89, 1.08) 1.01 (0.91, 1.12) 0.84 (0.76, 0.94) 0.001

ASCVD 1.00 0.98 (0.92, 1.06) 1.01 (0.94, 1.08) 0.90 (0.84, 0.97) 0.005

* Model adjusted for age, race, region, study indicators, BMI, total energy intake, percent energy from carbohydrates, smoking status, alcohol

consumption, physical activity, healthy eating index (HEI) 2005, and family history of the respective outcome.

57

Table 3.5.7 Estimates of relative risk and 95% confidence intervals (CIs) of diseases of interest by statistically substituting pasta

for other starch-dense foods from Model 3*

HR (95% CI) Pasta replacing potato Pasta replacing white bread Pasta replacing rice

Diabetes 1.03 (0.92, 1.15) 1.03 (0.93, 1.14) 1.03 (0.91, 1.16)

CHD 0.83 (0.67, 1.02) 0.97 (0.80, 1.17) 1.00 (0.79, 1.27)

Stroke 0.66 (0.52, 0.84) 0.73 (0.59, 0.92) 0.83 (0.63, 1.08)

ASCVD 0.78 (0.66, 0.92) 0.88 (0.76, 1.01) 0.91 (0.76, 1.10)

* Model adjusted for age, race, region, study indicators, BMI, total energy intake, percent energy from carbohydrates, smoking status, alcohol

consumption, physical activity, healthy eating index (HEI) 2005, and family history of the respective outcome.

58


Supplementary Table 3.6.1 Estimates of relative risk and 95% confidence intervals (CIs) of diseases of interest for one medium

serving/day increase in pasta intake variables

HR (95% CI) Residual Total Pasta

Intake Pasta/100GL Ratio

Pasta/1000kcal Energy

Ratio

Residual Total

Spaghetti Intake

Diabetes

Model 1a 1.11 (1.01, 1.22) 1.28 (1.15, 1.41) 1.22 (1.04, 1.43) 0.98 (0.88, 1.10)

Model 2b 1.08 (0.98, 1.18) 1.10 (0.99, 1.22) 1.16 (0.99, 1.37) 1.02 (0.92, 1.14)

Model 3c 1.06 (0.97, 1.17) 1.09 (0.98, 1.21) 1.12 (0.95, 1.32) 1.04 (0.94, 1.16)

CHD

Model 1a 0.95 (0.80, 1.13) 1.01 (0.84, 1.21) 0.84 (0.63, 1.12) 0.83 (0.68, 1.01)

Model 2b 0.94 (0.80, 1.12) 0.89 (0.74, 1.07) 0.82 (0.61, 1.09) 0.86 (0.70, 1.05)

Model 3c 0.93 (0.77, 1.11) 0.90 (0.74, 1.09) 0.79 (0.59, 1.07) 0.88 (0.72, 1.09)

Stroke

Model 1a 0.69 (0.56, 0.85) 0.73 (0.58, 0.90) 0.54 (0.39, 0.76) 0.59 (0.46, 0.75)

Model 2b 0.70 (0.57, 0.86) 0.67 (0.53, 0.83) 0.54 (0.38, 0.76) 0.61 (0.48, 0.78)

Model 3c 0.71 (0.58, 0.88) 0.69 (0.55, 0.86) 0.56 (0.31, 0.78) 0.64 (0.50, 0.82)

ASCVD

Model 1a 0.85 (0.74, 0.97) 0.89 (0.77, 1.02) 0.72 (0.58, 0.90) 0.74 (0.63, 0.87)

Model 2b 0.84 (0.74, 0.97) 0.80 (0.69, 0.93) 0.71 (0.57, 0.88) 0.76 (0.65, 0.89)

Model 3c 0.85 (0.74, 0.98) 0.82 (0.70, 0.95) 0.72 (0.57, 0.90) 0.80 (0.68, 0.94) a Model 1 adjusted for age, race, region, and study indicators



consumption, physical activity, healthy eating index (HEI) 2005, and family history of CHD and stroke.

59

Chapter 4 Genetic Variations Related to Type 2 Diabetes and Atherosclerotic

Cardiovascular Disease, and Their Interactions with Dietary Carbohydrates in

Influencing Long-Term Disease Risk

4.1 Introduction

Type 2 diabetes (T2D) and cardiovascular disease (CVD) are two of the most prevalent

chronic diseases in the United States (US) currently. It was estimated that a total of 28.9

million people aged 20 years or older in the US were affected by diagnosed or undiagnosed

diabetes in 2012, with 90 – 95 % of the cases being T2D.5 The number of US adults with

diabetes was projected to reach 36.0 million in 2030.92 CVD, being the leading cause of

death for both men and women, affected an estimated 85.6 million Americans adults.4 T2D

and CVD share many common risk factors and pathophysiological intermediates, including

obesity, unhealthy diet, cigarette smoking, lack of physical activity, insulin resistance,

dyslipidemia, chronic inflammation, etc., and they have also been found to share common

genetic basis.19, 20

Dietary carbohydrates have been extensively studied regarding their roles in

influencing the risk of developing cardiometabolic conditions. A large body of evidence

from both experimental and observational studies have demonstrated that higher

consumption of dietary fiber, whole grain, and low-GI carbohydrate-containing foods were

associated with beneficial effects on cardiometabolic health. For instance, greater whole

grain consumption has been associated with lower risk of T2D, lower risk of CVD, as well

60

as less weight gain in a fairly recent meta-analysis of cohort studies.93 The same meta-

analysis also summarized evidence from randomized controlled trials (RCTs), where

whole grain intervention groups were found to have significantly lowered fasting glucose,

total and low-density-lipoprotein (LDL) cholesterol compared to control groups.93 Since

the concept of glycemic index (GI) and glycemic load (GL) was introduced in the 1980s,17

to date convincing evidence has been accumulated and indicated that higher GI/GL diet

was associated with increased risk of T2D and coronary heart disease (CHD), the most

common type of CVD.18 As a typical example of foods with substantial amount of added

sugar, it is believed that now there is sufficient evidence that higher consumption of sugar-

sweetened beverages increases the risk of obesity and obesity related diseases, including

T2D and CVD.94 More relevantly, the genetic risk of obesity, measured by an overall

genetic risk score derived from single nucleotide polymorphisms (SNPs) identified from

GWAS, appeared to be more pronounced with greater consumption of sugar-sweetened

beverages,95 indicating possible interaction between these two factors in causing obesity.

While genome-wide association studies (GWAS) have successfully identified

multiple SNPs that are associated with the risk of T2D and CVD, most of them confer a

relatively modest amount of elevated risk, and the complex interplay between genetic risk

factors and the above-mentioned traditional risk factors are still to be fully elucidated. The

majority of other previous studies have used candidate SNP/gene approach where only one

or a few loci of interest was investigated.96-98 Some other studies used a summary genetic

risk score which incorporates all identified risk alleles to date.99-101 The current study aimed

at investigating potential interactions between various components of dietary

61

carbohydrates and genetic risk of diabetes and CVD based on genes containing SNPs that

have been associated with these diseases in previous GWAS.

4.2 Methods

4.2.1 Study Population

Existing 1000G imputed data from the national Women’s Health Initiative (WHI) SNP

Health Association Resource (WHI-SHARe) and the Genomics and Randomized Trials

Network (WHI-GARNET) were used in the discovery phase. The WHI was a long-term

study of post-menopausal women recruited from 40 clinical centers across the country

between 1993 and 1998. Participating women aged 50 – 79 years at recruitment. It consists

of three partially overlapping randomized controlled clinical trials with a total of 68,133

participants, and a prospective observational study with another 93,676 participants. The

three trials are a hormone therapy (HT) trial, a dietary modification (DM) trial, and a

calcium and vitamin D (CaD) trial. Detailed demographics, dietary, medical history and

outcome, behavioral, social, and psychosocial information was collected from recruitment

through the end of the core study in 2005. The majority of study participants also attended

the WHI extension study 2005 – 2010 (Ext1), during which medical outcome information

was continuously collected. Part of the study participants continued into the WHI extension

study 2010 – 2020 (Ext2) which is still ongoing.

The WHI-SHARe included a subsample of the African American (AA) and

Hispanic American (HA) participants of the WHI. Initially 121,151 self-identified AA and

5,469 self-identified HA participants had consented to genetic research and were eligible

for the WHI-SHARe. Due to budget constraints, a subsample of 12,157 of these women

62

(8,515 AA and 3,642 HA women) were randomly selected. DNA was extracted by the

Specimen Processing Laboratory at the Fred Hutchinson Cancer Research Center from

specimens that were collected at the time of enrollment.20 The WHI-GARNET included a

subsample of largely Caucasian (or European) American (EA) women enrolled in the WHI

HT trial, who met eligibility requirements for this study and eligibility for submission to

the database of Genotypes and Phenotypes (dbGaP), and provided DNA samples. Of the

27,347 women who participated in the HT trial, incident diabetes, CHD, stroke, and venous

thrombosis cases and matched controls free of prevalent or incident diabetes, CHD, stroke,

and venous thrombosis were included in the WHI-GARNET. The matching criteria

included age (± 5 years), race/ethnicity, hysterectomy status, enrollment date (± 1.5 years),

and length of follow-up (± 48 months). Controls were also prioritized on the basis of

availability of plasma biomarker availability (glucose, high density lipoprotein cholesterol,

low density lipoprotein cholesterol, total cholesterol, insulin, triglycerides, C-reactive

protein, and fibrinogen).20 The final study sample consists of the 2431 matched pairs and

the remaining 7 unmatched cases, of whom 2,208 matched pairs were of European ancestry.

Participants were excluded based on the following criteria: implausible total energy intake

(< 600 or > 5000 kcal/day); prevalent diseases including diabetes, cardiovascular diseases,

and cancer that may alter dietary behaviors; and measurements not available for outcomes

of interest (incident diabetes, CHD, or stroke).

The Jackson Heart Study (JHS) is a community-based observational study that

included 5,301 African American (AA) women and men aged 35 – 84 years, who were

followed from the baseline examination in 2000-2004 and at 12-month intervals after the

baseline. Major components of each follow-up exam included medical history update,

63

physical examination, blood/urine analytes and interviews. The JHS was used to replicate

the findings in the WHI-AA.

The Framingham Heart Study (FHS) is a large prospective, community-based,

three-generation family-based cohort study which included primarily European American

(EA) participants. The Original Cohort consists of 5,209 participants. The Offspring

Cohort consists of 5,124 participants and includes biological descendants of the Original

Cohort as well as spouses and adopted offspring. Generation 3 Cohort consists of 4,095

participants who are the biological descendants of the Offspring Cohort and adopted

offspring. Since only Generation 3 cohort of the FHS had dietary measurement from the

first exam, we restricted our replication sample to Generation 3 to reduce bias (for other

generations dietary questionnaires were administered in the middle of their follow-up). The

FHS Generation 3 cohort was used to replicate the findings in the WHI-EA.

4.2.2 Study Variables

4.2.2.1 Dietary variables

Dietary measurements in the WHI were obtained using a validated 122-item food

frequency questionnaire (FFQ) inquiring participants about their intake of various foods

over the past 3 months.40 The FFQ was based on instruments used in the WHI feasibility

studies and the original National Cancer Institute/Block FFQ.41, 42, 102 The dietary database,

linked to the University of Minnesota Nutrition Coordinating Center Nutrition Data System

for Research (Nutrition Coordinating Center, Minneapolis, MN, USA), is based on the U.S.

Department of Agriculture standard reference releases and manufacturer information.103

Total amount of various nutrients were computed accordingly. The detailed description of

64

the methods used to calculate GI and GL values can be found elsewhere.104 Data from the

FFQ were available in the forms of nutrients and food items/groups. Dietary total

carbohydrates, percent energy from carbohydrates, sugar, fiber, starch, added sugar, GI,

and GL can be extracted from the nutrient database, while dietary intake of white bread,

dark bread, potato, rice, pasta, cereal, beans, sugar-sweetened beverage (SSB), and whole

grains can be extracted from the food items and food groups database.

4.2.2.2 Genetic data

Genome-wide genotyping of the WHI-SHARe participants was performed using the

Affymetrix 6.0 array (Affymetrix, Inc, Santa Clara, CA). Genotyping for WHI-GARNET

participants was performed using the Illumina HumanOmni1-Quad SNP platform

(Illumina, Inc, San Diego, CA).20 As the gene chips used for WHI-SHARe and WHI-

GARNET are designed to capture common genetic variants, genetic variants with minor

allele frequency (MAF) ≥ 0.05 were genotyped. Reference panels from the 1000 Genomes

(1000G) Project Consortium (Version 3, March 2012 release), which provide near

complete coverage of common and low-frequency genetic variation with MAF ≥ 0.005,

were used for genotype imputation. Dosage R-squared was used to evaluate the imputation

quality, and an average R-squared greater than 0.8 was achieved.105 For the JHS,

genotyping was performed using the Affymetrix Genome-Wide Human SNP Array 6.0

platform, which interrogates >900,000 SNPs and has >900,000 probes for copy number

variation. These data have been imputed to around 37 million SNPs using the 1000

Genomes project reference panel (version 3, March 2012 release). The FHS cohort was

genotyped using the Affymetrix GeneChip Human Mapping 500K Array and the 50K

65

Human Gene Focused Panel. With the use of extant genotypes, imputation has been

completed for all participants to 40 million SNPs using the 1000G Imputation.

To identify the genes containing SNPs that have been associated with these diseases

in previous GWAS, we consulted the National Human Genome Research Institute –

European Bioinformatics Institute (NHGRI-EBI) Catalog of published genome-wide

association studies106 and downloaded spreadsheet containing all published GWAS

results.107 We then identified all genes that contained SNPs that were associated with any

traits that primarily concerned with type 2 diabetes or cardiovascular diseases and extracted

all SNPs within these genes that had imputation quality R-squared > 0.3, MAF >= 0.01,

and squared pair-wise Pearson correlation < 0.8 from 1000G imputed genetic data of the

WHI. SNPs from a list of 406 genes were available for analysis in WHI-AA and HA, and

SNPs from an additional three genes were available for analysis in WHI-EA after the

above-mentioned quality control procedures.

4.2.2.3 Type 2 diabetes

In WHI incident diabetes was assessed via questionnaire at enrollment and each annual

follow-up visit. Participants were asked if “a doctor prescribed for the first time any of the

following pills or treatments: pills for diabetes or insulin shots for diabetes” since their last

medical update. Those who responded “yes” were considered having been diagnosed with

diabetes. Since these were all post-menopausal women, newly diagnosed diabetes cases

were most likely type 2 diabetes (T2D) cases. In FHS, T2D was diagnosed either by a panel

of three physicians reviewed each cardiovascular event according to pre-established

criteria. In JHS, T2D was ascertained according to the American Diabetes Association

66

definitions: fasting glucose ≥126 mg/dL, hemoglobin A1C ≥ 6.5%, or current use of insulin

or oral hypoglycemic agents.

4.2.2.4 Atherosclerotic cardiovascular diseases

Incident atherosclerotic cardiovascular diseases (ASCVD) encompassed cases of

incident CHD and incident stroke, according to the 2013 ACC/AHA Guideline on the

Assessment of Cardiovascular Risk.82, 83. Incident coronary heart disease (CHD) was

defined as the first occurrence of clinical MI, definite silent MI or a death due to definite

or possible CHD. Incident stroke was defined as the first occurrence of stroke or a death

due to cerebrovascular event. CHD and stroke occurrences are among the physician

adjudicated health outcomes of the WHI. In FHS, coronary heart diseases (CHD) included

myocardial infarction and sudden and non-sudden CHD death; this outcome is often

referred to as “hard” CHD. CHD death included deaths attributable to myocardial

infarction and sudden and non-sudden CHD deaths. In JHS, hospitalized MI was defined

using ICD-9 codes 402, 410–414, 427, 428, and 518.4. The criteria for classifying death

from coronary heart disease (CHD) are based on any combination of 1) chest pain; 2)

history of MI, CHD, or angina; 3) the absence of evidence of other probable cause of death;

and/or 4) the use of ICD-9 codes (i.e., 250, 401, 402, 410–414, 427–429, 440, 518.4, 798,

799) or ICD-10 codes (E10–14, I10–11, I21–25, I46–51, I70, I97, J81, J96, R96, R98–99)

to identify deaths from CHD.108

67

4.2.3 Analytic Approaches

Baseline characteristics of WHI participants included in the current analysis were described

according to their diabetes or ASCVD status within each race/ethnicity (AA, HA, or EA).

Means and standard deviations (SDs) were generated as descriptive statistics for each

continuous variable, while frequencies and percentages were generated for each categorical

variable. The statistical significance of differences across pasta intake quartiles were tested

by analysis of variance (ANOVA) for continuous variables and by chi-square test for

categorical variables.

In the discovery phase, to investigate the interaction between dietary carbohydrates

factors and genetic risk on the additive scale within each race/ethnicity (AA, HA, or EA),

we constructed dichotomous variables for each dietary carbohydrate component of interest,

and a genetic risk score (GRS) of diabetes or ASCVD for each gene of interest. For each

combination of one dietary carbohydrate component and one gene, their dichotomized

variables were then combined into one categorical variable with four levels: low dietary

intake and low GRS, low dietary intake and high GRS, high dietary intake and low GRS,

and high dietary intake and high GRS. The relative excessive risk of interaction (RERI)

and its 95% CI and associated P-value was then calculated using this four-level categorical

variable.109 Adjustment for multiple testings across all combinations of dietary

carbohydrate components and genes were performed using the false discovery rate

(FDR).110 The analysis procedures were detailed as follows:

In the first step, we used the residual method to partially control for measurement

error in the FFQ, where each variable for a specific carbohydrate component was linearly

regressed on total energy intake, and the residuals added with the mean of the variable were

68

taken as a measure uncorrelated with total energy intake. Then these residual variables

were dichotomized at their respective mean values.

In the second step, we constructed overall summary scores of diabetes or ASCVD

risk for each gene. First, we recoded each SNP so that each one-unit increase represented

one additional copy of the minor allele for that SNP, and estimated the effect of each single

SNP on the risk of incident diabetes or ASCVD using an additive logistic regression model:

𝑙𝑜𝑔𝑖𝑡(𝑃(𝑌 = 1|𝐺, 𝑳)) = 𝛽0 + 𝛽𝑔𝐺 + 𝜷𝒍𝑳

where Y represented the outcome (incident diabetes or ASCVD), G represented the SNP,

and L represent the potential confounding factors included in the model. For WHI-AA and

WHI-HA, L included the first four principal components to account for population

stratification, age, age squared, and region. For WHI-EA, L additionally included

hysterectomy status and indicators of HT trial arm to account for the matched design of the

WHI-GARNET study.

After examining the results from the single-SNP analyses, SNPs with an estimated

standard error of greater than ten were excluded due to their potential of being outliers and

having too much weight in calculating the GRS. We used four methods to calculate the

GRS which have been used in previous studies: 1) for Method 1 the GRS for each

gene/outcome combination was the sum of number of minor alleles weighted by their

estimated 𝛽𝑔 from single-SNP models; 2) for Method 2 the GRS for each gene/outcome

combination was the sum of number of minor alleles weighted by the signs of their

estimated 𝛽𝑔 from single-SNP models; 3) for Method 3 the GRS for each gene/outcome

combination was the sum of number of minor alleles weighted by their estimated z-values

from single-SNP model; 4) for Method 4 we recoded the SNPs so that each one-unit

69

increase represented one additional copy of the risk allele for that SNP (having a positive

estimated 𝛽𝑔) the GRS for each gene/outcome combination was the sum of number of risk

alleles weighted by the absolute value of their estimated 𝛽𝑔 from single-SNP model. Each

GRS went through rank based inverse normalization so that their scale was not affected by

the number of SNPs in each gene, and was then dichotomized at their respective mean

value.

In the third step, as illustrated above, the following logistic regression analysis

including the four-category variable (X) combining the dichotomous carbohydrate and

GRS variables was conducted for each combination of dietary carbohydrate components

and genes on risk of diabetes and ASCVD:

𝑙𝑜𝑔𝑖𝑡(𝑃(𝑌 = 1|𝑋, 𝐿)) = 𝛽0 + 𝛽𝑥𝑋 + 𝜷𝒍𝑳

where Y represented the outcome (incident diabetes or ASCVD), and the same confounding

factors L as the single-SNP analysis were included for WHI-AA, HA, EA, respectively,

while additional covariates for body mass index (BMI), smoking status, alcohol intake and

physical activity were also included as potential common cause between the variable for a

specific carbohydrate component and the outcome.

Results from the above modeling were utilized to compute the RERI effect

estimates and associated P-values and 95% CIs. Let 𝑅00 = 1 represent the effect estimate

for the low dietary intake and low GRS category (reference), 𝑅01 represent the effect

estimate for the low dietary intake and high GRS category, 𝑅10 represent the effect

estimate for the high dietary intake and low GRS category, and 𝑅11 represent the effect

estimate for the high dietary intake and high GRS category (all effect estimates were odds

ratios). The point estimate of RERI was then computed using the following formula:

70

𝑅𝐸𝑅𝐼 = 𝑅11 − 𝑅01 − 𝑅10 + 1

Unadjusted point estimates, P-values, and 95% CIs were calculated using a

modified version of the epi.interaction function within the epiR package in R. We then

used the qvalue function within the qvalue package in R to compute the FDR-adjusted

qvalues based on the nominal P-values and a cut-off of <0.1 was used to evaluate the

significance of the RERI estimates. The multiple testing adjustment did not account for the

four methods of computing GRS since we would like to identify significant RERI estimate

based on any of the GRS method used.

In the fourth and final step, we attempted to replicate the significant results for

either diabetes or ASCVD within WHI-AA from the third step in the JHS cohort, and the

significant results within WHI-EA from the third step in the FHS cohort, wherever relevant

variables were available for the specific carbohydrate component of interest. Identical

analysis procedures and covariate adjustments were used in the replication analyses, while

additionally accounting for sex as well as the family structure in the FHS using their

pedigrees. All statistical analyses were conducted using R version 3.2.3 (The R Foundation

for Statistical Computing, Vienna, Austria).52

4.3 Results

Information regarding ethnicity, genotyping, and imputation for each study population are

provided in Supplementary Table 4.6.1. Baseline characteristics of WHI-AA, HA, EA

participants included in the current analysis were summarized in Table 4.5.1 according to

diabetes status and in Table 4.5.2 according to ASCVD status. In the discovery phase,

5,811 African American (AA), 2,718 Hispanic American (HA), and 3,472 European

71

American (EA) participants were included in the race/ethnicity-specific analysis after

applying the exclusion criteria. As expected, there were generally significant differences

in baseline body mass index (BMI), waist circumference (WC), waist-to-hip ratio (WHR),

and blood pressure (BP) measurements comparing those with cardiometabolic diseases to

those without.

Results with respect to the relative excessive risk of interaction (RERI) estimates

and their associated 95% CIs and nominal P-values for those with FDR q-values < 0.1 were

summarized in Table 4.5.3 for WHI-AA, in Table 4.5.4 for WHI-HA, and in Table 4.5.5

for WHI-EA. Note that GRS calculated with Method 1 and Method 4 yielded the same

results. To illustrate the interpretation of the RERI, we take an example with the gene

PCNXL2 and intake of SSB as exposures of interest, and incident T2D as outcome of

interest, which had the lowest q-value among all interactions examined in WHI-EA. An

estimated RERI of 0.95 (95% CI: 0.62, 1.28, P-value = 1.76e-8, q-value=0.0001),

indicating the existence of positive additive interaction, meant that there was a significant

synergistic effect between the risk of T2D associated with polymorphisms in gene PCNXL2

and SSB intake so that the effect of having both high genetic risk from PCNXL2 and high

SSB intake on T2D risk may be more substantial than the sum of the effect of having high

genetic risk from PCNXL2 alone and the effect of having high SSB intake alone, compared

to those who have both low genetic risk from PCNXL2 and low SSB intake.

For the replication analysis in the JHS, unfortunately the only carbohydrate variable

available was sugar intake, so we attempted to replicate the analysis with respect to the

gene SLC22A3, sugar, and risk of ASCVD, which had a significant RERI estimate in WHI-

AA using GRS Method 2. After applying the same exclusion criteria, a total of 2,410

72

participants were included in the analysis, with 108 cases of incident ASCVD. Using the

same method for GRS calculation, the estimated RERI of ASCVD in JHS for SLC22A3

and sugar was 0.80 (95% CI, -0.38, 1.84, P-value = 0.197).

For the replication analysis in the FHS Generation 3 cohort, a total of 3,415

participants of the Generation 3 cohort were included after applying the exclusion criteria.

Relevant information was available for us to attempt replication of all combinations of

genes and carbohydrate components which had a significant RERI estimate in WHI-EA.

Since GRS Method 1 and 4 yielded identical results in WHI-EA, we omitted Method 4 in

the replication analysis, and the results from FHS were presented in Supplementary Table

4.6.2. We did not observe nominal P-values < 0.05.

4.4 Discussion

In this ethnicity-specific analysis of potential interactions between various components of

dietary carbohydrates and GRS of T2D and ASCVD based on genes that have been

previously associated with these diseases, we identified multiple possible interactions

within each ethnicity using a variety of methods in computing GRS, including five unique

combinations in African Americans for ASCVD, eleven unique combinations for T2D and

ten unique combinations for ASCVD in Hispanic Americans, and nine unique

combinations for T2D and two unique combinations for ASCVD in European Americans.

Unfortunately, our attempts to replicate these results wherever data was available in

external cohorts did not yield significant results.

To our knowledge, the current analysis was the first one to utilize the RERI

estimates in evaluating the potential interaction effects between dietary factors and generic

73

risk scores of cardiometabolic diseases or risk factors in influencing disease risk. In

previous studies, a summary genetic risk score which incorporates all identified risk alleles

to date across different genes was usually used.95, 99, 100 In 6,934 women from the Nurses'

Health Study and 4,423 men from the Health Professionals Follow-up Study, the genetic-

predisposition score was calculated on the basis of 32 alleles associated with BMI, and the

authors demonstrated that the genetic association with adiposity appeared to be more

pronounced with greater intake of sugar-sweetened beverages.95 In the Malmo Diet and

Cancer Study cohort, investigators did not find convincing evidence of an interaction

between genetic susceptibility for dyslipidemia, measured as genetic risk scores of

dyslipidemia-associated variants, and the consumption of carbohydrate-rich foods

(including potatoes, whole grains, refined grains, sugar and sweets, and sugar-sweetened

beverages) on incident CVD risk was observed.99 In 6,414 non-Hispanic whites, 3,073 non-

Hispanic blacks, and 3,633 Mexican American participants from the National Health and

Nutrition Examination Surveys (NHANES), interactions between the GRS based on 15

T2D-associated variants from GWAS and carbohydrate or fiber intake failed to reach

significance in all the racial/ethnic groups.100 While constructing the GRS based on risk

alleles can incorporate potential risk conveyed from different genes, it is difficult to

ascertain which allele or gene contributed the most to the observed

interactions/associations. Our construction of GRS within each gene help focus on

potential target genes and our findings represent novel discoveries of possible interactions

between dietary carbohydrates and genes in influencing prospective T2D and ASCVD risk

in ethnic-specific fashion, and could facilitate hypothesis generation in future mechanistic

studies in either animals or humans.

74

While not all the genes in Tables 4.5.3 – 4.5.5 have apparent functions in

carbohydrate metabolism or glucose homeostasis, some of the interactions that we

identified exhibited biological plausibility. In EA we observed that the GIP gene may

interact with potato intake in influencing ASCVD risk. This gene encodes an incretin

hormone called gastric inhibitory polypeptide, or glucose-dependent insulinotropic

polypeptide, that belongs to the glucagon superfamily. The encoded protein is important in

maintaining glucose homeostasis as a potent stimulator of insulin secretion from pancreatic

beta-cells following food ingestion and nutrient absorption.111 In HA we observed an

possible interaction between the MNX1 gene and sugar intake also in affecting ASCVD

risk. The MNX1 gene encodes a nuclear protein, which contains a homeobox domain and

is a putative transcription factor involved in pancreas development and function,112, 113 and

the pancreas is responsible for producing hormones such as insulin and glucagon to

regulate blood sugar levels. For T2D, the interaction between HHIPL1 and sugar was also

biologically plausible since the HHIPL1 gene encodes for the hedgehog interacting protein-

like 1 protein that is thought to belong to the glucose/sorbosone dehydrogenase family,

which may be involved in alternative metabolic pathways of glucose other than glycolysis.

To our knowledge, this Several limitations of this study should be noted in

interpreting the results. First, since logistic models were used throughout the analysis, the

effect estimates used to calculate the RERI were all odds ratios, which tend to overestimate

the relative risk unless the disease of interest is rare. The incidence of T2D and ASCVD in

WHI-AA, HA, and EA was all quite high (> 5%), so the estimates of RERI may be inflated

due to the use of odds ratios. Second, we did not succeed in our replication attempts in the

JHS and FHS, and there may be several contributing reasons. The disease incident in both

75

JHS and FHS were fairly low (108 cases of ASCVD out of 2,410 participants in the JHS

and 44 cases of T2D and 47 cases of ASCVD out of 3,415 participants in the FHS),

especially for a young cohort such as FHS Generation 3 with an average age of 40 years at

their first examination in 2002 – 2005. Due to quality control procedures applied to the

imputed genetic data, for genes involved in the replication phase, we were not able to use

all SNPs from the discovery phase to construct the GRS in the JHS or FHS for a

corresponding gene. Identification of other suitable cohorts and consortium efforts would

be required for further investigation. Third, we adjusted for the same set of covariates that

may confound the diet-disease associations based on previous knowledge. While the

approach was acceptable since all dietary factors concerned carbohydrates, we recognize

the fact that different dietary carbohydrate components may be influenced by different

causes and we cannot exclude the possibility of residual confounding. Finally,

measurement errors associated with the FFQ could be a source of information bias, and we

applied the residual method to all dietary carbohydrate measurements to partially correct

for this issue.

In conclusion, utilizing the 1000 Genome Project imputed data from three

ethnicities within a well-established prospective cohort, we identified several potential

diet-gene interactions that may be associated with altered risk of T2D and/or ASCVD

across over four hundred genes of interest. The results needed to be further validated in

other external cohorts before drawing firm conclusions, but they could serve as potential

hypotheses for future research in biological mechanisms and personalized prevention

strategies for cardiometabolic diseases.

76

4.5 Tables

Table 4.5.1 Baseline characteristics of WHI-SHARe African American (AA, n=5,811) and Hispanic American (HA, n=2,718)

participants, and WHI-GARNET European American (EA, n=3,472) participants according to their incident diabetes status

Mean (SD)

WHI-AA WHI-HA WHI-EA

Diabetes

(No)

Diabetes

(Yes) P-value

Diabetes

(No)

Diabetes

(Yes) P-value

Diabetes

(No)

Diabetes

(Yes) P-value

Age (yrs) 61.2 (7.1) 60.6 (6.3) 0.021 60.1 (6.6) 59.7 (6.4) 0.230 66.1 (6.8) 63.9 (6.9) <.001

BMI (kg/m2) 30.3 (6.4) 32.8 (6.5) <.001 28.1 (5.5) 31.2 (5.8) <.001 28.5 (5.7) 32.3 (6.3) <.001

WC (cm) 89.2 (13.2) 95.8 (13.5) <.001 84.4 (12.0) 93.3 (13.1) <.001 88.6 (13.3) 98.3 (14.4) <.001

WHR 0.81 (0.07) 0.84 (0.07) <.001 0.81 (0.08) 0.85 (0.08) <.001 0.82 (0.08) 0.87 (0.09) <.001

SBP (mmHg) 130.5 (17.6) 133.2 (16.5) <.001 124.7 (16.9) 128.9 (16.6) <.001 129.8 (17.8) 131.3 (17.1) 0.028

DBP (mmHg) 78.2 (9.2) 78.9 (9.3) 0.021 75.1 (9.0) 76.3 (9.4) 0.009 75.5 (9.3) 77.1 (9.0) <.001

Region, n (%) 0.027 0.937 0.088

Northeast 815 (17.4) 225 (19.9) 271 (12.0) 52 (11.4) 650 (25.8) 236 (24.7)

South 2281 (48.7) 562 (49.8) 948 (41.9) 188 (41.3) 511 (20.3) 217 (22.7)

Midwest 1071 (22.9) 244 (21.6) 92 (4.1) 17 (3.7) 759 (30.1) 254 (26.6)

West 516 (11.0) 97 (8.6) 952 (42.1) 198 (43.5) 598 (23.7) 247 (25.9)

Abbreviations: BMI: body mass index; WC: waist circumference; WHR: waist-to-hip ratio; SBP: systolic blood pressure; DBP: diastolic blood

pressure.

77

Table 4.5.2 Baseline characteristics of WHI-SHARe African American (AA, n=5,811) and Hispanic American (HA, n=2,718)

participants, and WHI-GARNET European American (EA, n=3,472) participants according to their incident ASCVD status

Mean (SD)

WHI-AA WHI-HA WHI-EA

ASCVD

(No)

ASCVD

(Yes) P-value

ASCVD

(No)

ASCVD

(Yes) P-value

ASCVD

(No)

ASCVD

(Yes) P-value

Age (yrs) 60.7 (6.8) 63.8 (7.3) <.001 59.8 (6.5) 63.3 (7.3) <.001 64.7 (6.9) 67.6 (6.5) <.001

BMI (kg/m2) 30.7 (6.5) 31.3 (6.4) 0.019 28.6 (5.7) 28.9 (5.3) 0.440 29.8 (6.3) 28.8 (5.6) <.001

WC (cm) 90.1 (13.4) 93.2 (14.1) <.001 85.7 (12.5) 88.5 (14.2) 0.004 91.4 (14.5) 90.8 (13.6) 0.217

WHR 0.81 (0.07) 0.83 (0.08) <.001 0.81 (0.08) 0.84 (0.10) <.001 0.83 (0.08) 0.84 (0.09) <.001

SBP (mmHg) 130.4 (17.1) 136.6 (19.1) <.001 124.8 (16.7) 133.1 (18.2) <.001 128.4 (17.2) 135.2 (17.9) <.001

DBP (mmHg) 78.3 (9.1) 78.7 (10.1) 0.261 75.3 (9.1) 75.7 (9.4) 0.554 75.6 (9.2) 76.8 (9.3) <.001

Region, n (%) 0.354 0.062 <.001

Northeast 924 (17.8) 116 (18.3) 295 (11.6) 28 (15.5) 609 (24.0) 277 (29.7)

South 2553 (49.3) 290 (45.7) 1076 (42.4) 60 (33.1) 523 (20.6) 205 (22.0)

Midwest 1158 (22.4) 157 (24.8) 103 (4.1) 6 (3.3) 781 (30.8) 232 (24.9)

West 542 (10.5) 71 (11.2) 1063 (41.9) 87 (48.1) 626 (24.7) 219 (23.5)

Abbreviations: BMI: body mass index; WC: waist circumference; WHR: waist-to-hip ratio; SBP: systolic blood pressure; DBP: diastolic blood

pressure.

78

Table 4.5.3 Significant interactions between genes of interest and components of dietary carbohydrate identified in WHI-AA

Gene name Carbohydrate

component Outcome

Method of

GRS RERI estimate RERI 95% CI RERI P-value FDR q-value

SLC22A3 Sugar ASCVD M2 0.64 (0.33, 0.94) 3.80e-5 0.081

PCSK9 White bread ASCVD M2 0.68 (0.36, 1.01) 3.80e-5 0.081

STK32B Cereal ASCVD M2 0.78 (0.43, 1.13) 1.24e-5 0.052

WDR12 Pasta ASCVD M1 0.76 (0.43, 1.09) 5.11e-6 0.044



WFS1 Pasta ASCVD M1 0.70 (0.39, 1.01) 1.04e-5 0.044

WFS1 Pasta ASCVD M4 0.70 (0.39, 1.01) 1.04e-5 0.044

Abbreviations: GRS: genetic risk score; M1-4: Method 1-4 (for GRS calculation); RERI: relative excessive risk of interaction; ASCVD:

atherosclerotic cardiovascular disease.

79

Table 4.5.4 Significant interactions between genes of interest and components of dietary carbohydrate identified in WHI-HA


component Outcome

Method of


AP3D1 GL T2D M1 0.70 (0.38, 1.02) 1.47e-5 0.062

AP3D1 GL T2D M4 0.70 (0.38, 1.02) 1.47e-5 0.061

SMARCA4 Sugar T2D M3 0.81 (0.47, 1.15) 2.46e-6 0.010

PSRC1 Added sugar T2D M3 0.69 (0.34, 1.04) 9.62e-5 0.096

SYCE1L Rice T2D M1 0.99 (0.63, 1.36) 9.85e-8 0.001

SYCE1L Rice T2D M3 0.75 (0.37, 1.13) 9.94e-5 0.096

SYCE1L Rice T2D M4 0.99 (0.63, 1.36) 9.85e-8 0.001

KLF12 Dark bread T2D M2 1.16 (0.76, 1.57) 1.74e-8 0.0001

KLF12 Dark bread T2D M3 0.95 (0.52, 1.40) 2.20e-5 0.047

KCNK16 Dark bread T2D M3 0.79 (0.41, 1.17) 4.36e-5 0.074

CETP Dark bread T2D M3 0.74 (0.37, 1.12) 8.95e-5 0.096

TWIST1 SSB T2D M2 0.91 (0.48, 1.33) 2.61e-5 0.074

HUNK SSB T2D M2 0.99 (0.54, 1.44) 1.68e-5 0.072

HUNK SSB T2D M3 1.20 (0.78, 1.61) 1.32e-8 0.0001

PSRC1 SSB T2D M3 0.81 (0.45, 1.17) 1.22e-5 0.034

PPARG Beans T2D M3 0.70 (0.35, 1.05) 1.01e-4 0.096

MNX1 Sugar ASCVD M3 1.00 (0.51, 1.49) 5.83e-5 0.080

80

ACOT6 Fiber ASCVD M1 1.06 (0.53, 1.57) 6.76e-5 0.096



ZNF217 Added sugar ASCVD M1 1.10 (0.57, 1.63) 4.76e-5 0.096

ZNF217 Added sugar ASCVD M4 1.10 (0.57, 1.63) 4.76e-5 0.096

CILP2 Added sugar ASCVD M3 1.19 (0.63, 1.74) 3.11e-5 0.053

C16orf74 White bread ASCVD M3 1.30 (0.66, 1.93) 6.61e-5 0.080

CLEC14A Cereal ASCVD M1 1.24 (0.70, 1.78) 6.15e-6 0.052



SOCS5P5 Beans ASCVD M1 1.05 (0.56, 1.54) 2.81e-5 0.093



CNKSR3 Beans ASCVD M3 1.43 (0.76, 2.10) 2.97e-5 0.053

VPS26A SSB ASCVD M1 1.35 (0.69, 2.02) 6.33e-5 0.096

VPS26A SSB ASCVD M4 1.35 (0.69, 2.02) 6.33e-5 0.096

PEPD SSB ASCVD M1 1.54 (0.78, 2.31) 7.98e-5 0.097

PEPD SSB ASCVD M4 1.54 (0.78, 2.31) 7.98e-5 0.097

Abbreviations: GRS: genetic risk score; M1-4: Method 1-4 (for GRS calculation); RERI: relative excessive risk of interaction; GL: glycemic load;

SSB: sugar-sweetened beverage; T2D: type 2 diabetes; ASCVD: atherosclerotic cardiovascular disease.

81

Table 4.5.5 Significant interactions between genes of interest and components of dietary carbohydrate identified in WHI-EA


component Outcome

Method of


HHIPL1 Sugar T2D M1 0.74 (0.45, 1.02) 5.49e-7 0.002

HHIPL1 Sugar T2D M4 0.74 (0.45, 1.02) 5.49e-7 0.002

ABCG5 GI T2D M2 0.63 (0.34, 0.92) 1.90e-5 0.041

ABCG5 GI T2D M3 0.68 (0.38, 0.98) 7.53e-6 0.020

ABCG5 White bread T2D M1 0.63 (0.35, 0.92) 1.51e-5 0.043




JADE2 Beans T2D M2 0.68 (0.39, 0.97) 3.31e-6 0.014

PCNXL2 SSB T2D M1 0.95 (0.62, 1.28) 1.76e-8 0.0001

PCNXL2 SSB T2D M2 0.67 (0.35, 0.99) 4.58e-5 0.078

PCNXL2 SSB T2D M3 0.82 (0.50, 1.14) 5.37e-7 0.005

PCNXL2 SSB T2D M4 0.95 (0.62, 1.28) 1.76e-8 0.0002

ABCB11 SSB T2D M2 0.66 (0.36, 0.97) 1.91e-5 0.041

ACAD10 SSB T2D M3 0.69 (0.41, 0.97) 1.47e-6 0.006

NINJ2 SSB T2D M3 0.65 (0.33, 0.97) 7.54e-5 0.092

PKMYT1 Whole grain T2D M3 0.58 (0.30, 0.86) 5.24e-5 0.074

82

BCAS3 Starch ASCVD M1 0.78 (0.44, 1.13) 1.01e-5 0.085

BCAS3 Starch ASCVD M4 0.78 (0.44, 1.13) 1.01e-5 0.085

GIP Potato ASCVD M2 0.61 (0.34, 0.89) 1.19e-5 0.100

Abbreviations: GRS: genetic risk score; M1-4: Method 1-4 (for GRS calculation); RERI: relative excessive risk of interaction.

83


Supplementary Table 4.6.1 Summary of genotyping, imputation and quality control procedures in each cohort

Abbreviations Study names Ethnicity Genotyping

platform

QC filters (for

genotyping)

Imputation

Method

QC filters (for

imputation)

WHI-SHARe

Women's Health

Initiative-SNP Health

Association Resource

African

Americans Affymetrix 6.0

min. sample call rate, 95%;

min. SNP call rate, 90%;

HWE, 1e-6;

MAF: 0.01.

MACH

MAF, MAF>0.01;

imputation quality,

r2>0.3.

WHI-

GARNET

Women's Health

Initiative-Genomics

and Randomized

Trials Network

European

Americans

Illumina

HumanOmni1-

Quad v1-0 B

min. sample call rate, 98%;

min. SNP call rate, 98%;

HWE, 1e-4;

MAF: none.

MACH

MAF, MAF>0.01;

imputation quality,

r2>0.3.

JHS Jackson Heart Study African

Americans Affymetrix 6.0

min. call rate, 95%;

HWE, 1e-6;

MAF, 0.01.

MACH

MAF, MAF>0.01;

imputation quality,

r2>0.3.

FHS Framingham Heart

Study

European

Americans

Affymetrix 500K

Affymetrix 50K

supplementary

array

min. call rate 97%;

HWE, 1e-6. MACH

MAF, MAF>0.01;

imputation quality,

r2>0.3.

Abbreviations: HWE: Hardy-Weinberg equilibrium; QC: quality control; MAF: minor allele frequency.

84

Supplementary Table 4.6.2 Replication results in the FHS Generation 3 cohort


component Outcome Method of GRS RERI estimate RERI 95% CI RERI P-value

HHIPL1 Sugar T2D M1 0.52 (-1.33, 2.38) 0.580

ABCG5 GI T2D M2 0.53 (-0.05, 1.11) 0.074

ABCG5 GI T2D M3 0.53 (-0.05, 1.11) 0.074

ABCG5 White bread T2D M1 0.04 (-1.05, 1.13) 0.939



JADE2 Beans T2D M2 -0.04 (-0.97, 0.90) 0.938

PCNXL2 SSB T2D M1 -1.75 (-3.68, 0.18) 0.076

PCNXL2 SSB T2D M2 -1.75 (-3.68, 0.18) 0.076

PCNXL2 SSB T2D M3 -1.75 (-3.68, 0.18) 0.076

ABCB11 SSB T2D M2 -1.75 (-3.68, 0.18) 0.076

ACAD10 SSB T2D M3 -1.75 (-3.68, 0.18) 0.076

NINJ2 SSB T2D M3 -1.75 (-3.68, 0.18) 0.076

PKMYT1 Whole grain T2D M3 -0.85 (-2.63, 0.93) 0.347

BCAS3 Starch ASCVD M1 -0.21 (-1.37, 0.94) 0.718

GIP Potato ASCVD M2 0.42 (-2.22, 3.06) 0.755

Abbreviations: GRS: genetic risk score; M1-4: Method 1-4 (for GRS calculation); RERI: relative excessive risk of interaction.

85

Appendix

Acknowledgments to the Women’s Health Initiative (WHI)

The WHI program is funded by the National Heart, Lung, and Blood Institute, National

Institutes of Health, U.S. Department of Health and Human Services through contracts

HHSN268201100046C, HHSN268201100001C, HHSN268201100002C,

HSN268201100003C, HHSN268201100004C, and HHSN271201100004C.

We would like to acknowledge the investigators of each of the ancillary studies (AS90,

AS110, AS167, AS238, BA7, BA9, BA21, W5, W9, W10, and W18) and the following

WHI investigators:

Program Office: (National Heart, Lung, and Blood Institute, Bethesda, Maryland) Jacques

Rossouw, Shari Ludlam, Joan McGowan, Leslie Ford, and Nancy Geller Clinical.

Coordinating Center: Clinical Coordinating Center: (Fred Hutchinson Cancer Research

Center, Seattle, WA) Garnet Anderson, Ross Prentice, Andrea LaCroix, and Charles

Kooperberg.

Investigators and Academic Centers: (Brigham and Women's Hospital, Harvard Medical

School, Boston, MA) JoAnn E. Manson; (MedStar Health Research Institute/Howard

University, Washington, DC) Barbara V. Howard; (Stanford Prevention Research Center,

Stanford, CA) Marcia L. Stefanick; (The Ohio State University, Columbus, OH) Rebecca

Jackson; (University of Arizona, Tucson/Phoenix, AZ) Cynthia A. Thomson; (University

at Buffalo, Buffalo, NY) Jean Wactawski-Wende; (University of Florida,

86

Gainesville/Jacksonville, FL) Marian Limacher; (University of Iowa, Iowa City/Davenport,

IA) Jennifer Robinson; (University of Pittsburgh, Pittsburgh, PA) Lewis Kuller; (Wake

Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker; (University

of Nevada, Reno, NV) Robert Brunner; (University of Minnesota, Minneapolis, MN)

Karen L. Margolis Women’s Health Initiative Memory Study: (Wake Forest University

School of Medicine, Winston-Salem, NC) Mark Espeland.

Acknowledgments to the Jackson Heart Study (JHS)

The Jackson Heart Study is supported by contracts HHSN268201300046C,

HHSN268201300047C, HHSN268201300048C, HHSN268201300049C, and

HHSN268201300050C from the National Heart, Lung, and Blood Institute and the

National Institute on Minority Health and Health Disparity. The authors thank the

participants and the hard work of the Jackson Heart Study investigators and staff.

Acknowledgments to the Framingham Heart Study (FHS)

The Framingham Heart Study is conducted and supported by the National Heart, Lung, and

Blood Institute (NHLBI) in collaboration with Boston University (Contract No. N01-HC-

25195). This manuscript was not prepared in collaboration with investigators of the

Framingham Heart Study and does not necessarily reflect the opinions or views of the

Framingham Heart Study, Boston University, or NHLBI. The authors thank the

participants and the hard work of the Framingham Heart Study investigators and staff.

87

Bibliography

1. World Health Organization. Cardiovascular diseases (CVDs) Fact Sheet. May 2017.

Accessed at http://www.who.int/mediacentre/factsheets/fs317/en/.

2. World Health Organization. Diabetes Fact Sheet. November 2017. Accessed at

http://www.who.int/mediacentre/factsheets/fs312/en/.

3. Mathers CD and Loncar D. Projections of global mortality and burden of disease

from 2002 to 2030. PLoS Med. 2006;3:e442. PMC1664601.

4. Mozaffarian D, Benjamin EJ, Go AS, Arnett DK, Blaha MJ, Cushman M, Das SR,

de Ferranti S, Despres JP, Fullerton HJ, Howard VJ, Huffman MD, Isasi CR,

Jimenez MC, Judd SE, Kissela BM, Lichtman JH, Lisabeth LD, Liu S, Mackey RH,

Magid DJ, McGuire DK, Mohler ER, 3rd, Moy CS, Muntner P, Mussolino ME,

Nasir K, Neumar RW, Nichol G, Palaniappan L, Pandey DK, Reeves MJ,

Rodriguez CJ, Rosamond W, Sorlie PD, Stein J, Towfighi A, Turan TN, Virani SS,

Woo D, Yeh RW and Turner MB. Heart Disease and Stroke Statistics-2016 Update:

A Report From the American Heart Association. Circulation. 2016;133:e38-e360.

5. Centers for Disease Control and Prevention. National Diabetes Statistics Report:

Estimates of Diabetes and Its Burden in the United States, 2014. Atlanta, GA: US

Department of Health and Human Services. 2014.

http://www.who.int/mediacentre/factsheets/fs317/en/

http://www.who.int/mediacentre/factsheets/fs312/en/

88

6. Liu S, Buring JE, Sesso HD, Rimm EB, Willett WC and Manson JE. A prospective

study of dietary fiber intake and risk of cardiovascular disease among women. J Am

Coll Cardiol. 2002;39:49-56.

7. Liu S, Willett WC, Stampfer MJ, Hu FB, Franz M, Sampson L, Hennekens CH and

Manson JE. A prospective study of dietary glycemic load, carbohydrate intake, and

risk of coronary heart disease in US women. Am J Clin Nutr. 2000;71:1455-61.

8. Liu S and Manson JE. Dietary carbohydrates, physical inactivity, obesity, and the

'metabolic syndrome' as predictors of coronary heart disease. Current opinion in

lipidology. 2001;12:395-404.

9. Hu FB, Manson JE, Stampfer MJ, Colditz G, Liu S, Solomon CG and Willett WC.

Diet, lifestyle, and the risk of type 2 diabetes mellitus in women. N Engl J Med.

2001;345:790-7.

10. Willett W, Manson J and Liu S. Glycemic index, glycemic load, and risk of type 2

diabetes. Am J Clin Nutr. 2002;76:274S-80S.

11. Janket SJ, Manson JE, Sesso H, Buring JE and Liu S. A prospective study of sugar

intake and risk of type 2 diabetes in women. Diabetes care. 2003;26:1008-15.

12. Liu S, Manson JE, Stampfer MJ, Holmes MD, Hu FB, Hankinson SE and Willett

WC. Dietary glycemic load assessed by food-frequency questionnaire in relation to

plasma high-density-lipoprotein cholesterol and fasting plasma triacylglycerols in

postmenopausal women. Am J Clin Nutr. 2001;73:560-6.

13. Liu S, Manson JE, Buring JE, Stampfer MJ, Willett WC and Ridker PM. Relation

between a diet with a high glycemic load and plasma concentrations of high-

89

sensitivity C-reactive protein in middle-aged women. Am J Clin Nutr. 2002;75:492-

8.

14. Ding EL, Song Y, Manson JE, Hunter DJ, Lee CC, Rifai N, Buring JE, Gaziano

JM and Liu S. Sex hormone-binding globulin and risk of type 2 diabetes in women

and men. N Engl J Med. 2009;361:1152-63. 2774225.

15. Hardy K, Brand-Miller J, Brown KD, Thomas MG and Copeland L. The

Importance of Dietary Carbohydrate in Human Evolution. Q Rev Biol.

2015;90:251-268.

16. Atkinson FS, Foster-Powell K and Brand-Miller JC. International tables of

glycemic index and glycemic load values: 2008. Diabetes care. 2008;31:2281-3.

2584181.

17. Jenkins DJ, Wolever TM, Taylor RH, Barker H, Fielden H, Baldwin JM, Bowling

AC, Newman HC, Jenkins AL and Goff DV. Glycemic index of foods: a

physiological basis for carbohydrate exchange. Am J Clin Nutr. 1981;34:362-6.

18. Augustin LS, Kendall CW, Jenkins DJ, Willett WC, Astrup A, Barclay AW, Bjorck

I, Brand-Miller JC, Brighenti F, Buyken AE, Ceriello A, La Vecchia C, Livesey G,

Liu S, Riccardi G, Rizkalla SW, Sievenpiper JL, Trichopoulou A, Wolever TM,

Baer-Sinnott S and Poli A. Glycemic index, glycemic load and glycemic response:

An International Scientific Consensus Summit from the International Carbohydrate

Quality Consortium (ICQC). Nutr Metab Cardiovasc Dis. 2015;25:795-815.

19. Stern MP. Do non-insulin-dependent diabetes mellitus and cardiovascular disease

share common antecedents? Ann Intern Med. 1996;124:110-116.

90

20. Chan KH, Huang YT, Meng Q, Wu C, Reiner A, Sobel EM, Tinker L, Lusis AJ,

Yang X and Liu S. Shared molecular pathways and gene networks for

cardiovascular disease and type 2 diabetes mellitus in women across diverse

ethnicities. Circulation Cardiovascular genetics. 2014;7:911-9.

21. Ding EL, Song Y, Malik VS and Liu S. Sex differences of endogenous sex

hormones and risk of type 2 diabetes: a systematic review and meta-analysis. Jama.

2006;295:1288-99.

22. Goto A, Chen BH, Song Y, Cauley J, Cummings SR, Farhat GN, Gunter M, Van

Horn L, Howard BV, Jackson R, Lee J, Rexrode KM and Liu S. Age, body mass,

usage of exogenous estrogen, and lifestyle factors in relation to circulating sex

hormone-binding globulin concentrations in postmenopausal women. Clinical

chemistry. 2014;60:174-85.

23. Kalyani RR, Franco M, Dobs AS, Ouyang P, Vaidya D, Bertoni A, Gapstur SM

and Golden SH. The association of endogenous sex hormones, adiposity, and

insulin resistance with incident diabetes in postmenopausal women. The Journal of

clinical endocrinology and metabolism. 2009;94:4127-35. 2775652.

24. Chen BH, Brennan K, Goto A, Song Y, Aziz N, You NC, Wellons MF, Manson JE,

White DL, Butch AW and Liu S. Sex hormone-binding globulin and risk of clinical

diabetes in American black, Hispanic, and Asian/Pacific Islander postmenopausal

women. Clinical chemistry. 2012;58:1457-66. 4088986.

25. Laaksonen DE, Niskanen L, Punnonen K, Nyyssonen K, Tuomainen TP, Valkonen

VP, Salonen R and Salonen JT. Testosterone and sex hormone-binding globulin

91

predict the metabolic syndrome and diabetes in middle-aged men. Diabetes care.

2004;27:1036-41.

26. Fortunati N, Catalano MG, Boccuzzi G and Frairia R. Sex Hormone-Binding

Globulin (SHBG), estradiol and breast cancer. Molecular and cellular

endocrinology. 2010;316:86-92.

27. Key T, Appleby P, Barnes I and Reeves G. Endogenous sex hormones and breast

cancer in postmenopausal women: reanalysis of nine prospective studies. Journal

of the National Cancer Institute. 2002;94:606-16.

28. Allen NE, Key TJ, Dossus L, Rinaldi S, Cust A, Lukanova A, Peeters PH, Onland-

Moret NC, Lahmann PH, Berrino F, Panico S, Larranaga N, Pera G, Tormo MJ,

Sanchez MJ, Ramon Quiros J, Ardanaz E, Tjonneland A, Olsen A, Chang-Claude

J, Linseisen J, Schulz M, Boeing H, Lundin E, Palli D, Overvad K, Clavel-

Chapelon F, Boutron-Ruault MC, Bingham S, Khaw KT, Bueno-de-Mesquita HB,

Trichopoulou A, Trichopoulos D, Naska A, Tumino R, Riboli E and Kaaks R.

Endogenous sex hormones and endometrial cancer risk in women in the European

Prospective Investigation into Cancer and Nutrition (EPIC). Endocrine-related

cancer. 2008;15:485-97. 2396334.

29. Lee JS, LaCroix AZ, Wu L, Cauley JA, Jackson RD, Kooperberg C, Leboff MS,

Robbins J, Lewis CE, Bauer DC and Cummings SR. Associations of serum sex

hormone-binding globulin and sex hormone concentrations with hip fracture risk

in postmenopausal women. The Journal of clinical endocrinology and metabolism.

2008;93:1796-803. 2386277.

92

30. Perry JR, Weedon MN, Langenberg C, Jackson AU, Lyssenko V, Sparso T,

Thorleifsson G, Grallert H, Ferrucci L, Maggio M, Paolisso G, Walker M, Palmer

CN, Payne F, Young E, Herder C, Narisu N, Morken MA, Bonnycastle LL, Owen

KR, Shields B, Knight B, Bennett A, Groves CJ, Ruokonen A, Jarvelin MR,

Pearson E, Pascoe L, Ferrannini E, Bornstein SR, Stringham HM, Scott LJ,

Kuusisto J, Nilsson P, Neptin M, Gjesing AP, Pisinger C, Lauritzen T, Sandbaek

A, Sampson M, Zeggini E, Lindgren CM, Steinthorsdottir V, Thorsteinsdottir U,

Hansen T, Schwarz P, Illig T, Laakso M, Stefansson K, Morris AD, Groop L,

Pedersen O, Boehnke M, Barroso I, Wareham NJ, Hattersley AT, McCarthy MI

and Frayling TM. Genetic evidence that raised sex hormone binding globulin

(SHBG) levels reduce the risk of type 2 diabetes. Human molecular genetics.

2010;19:535-44. 2798726.

31. Moran LJ, Brinkworth GD, Martin S, Wycherley TP, Stuckey B, Lutze J, Clifton

PM, Wittert GA and Noakes M. Long-Term Effects of a Randomised Controlled

Trial Comparing High Protein or High Carbohydrate Weight Loss Diets on

Testosterone, SHBG, Erectile and Urinary Function in Overweight and Obese Men.

PLoS One. 2016;11:e0161297.

32. Goto A, Song Y, Chen BH, Manson JE, Buring JE and Liu S. Coffee and caffeine

consumption in relation to sex hormone-binding globulin and risk of type 2 diabetes

in postmenopausal women. Diabetes. 2011;60:269-75. 3012180.

33. van Gemert WA, Schuit AJ, van der Palen J, May AM, Iestra JA, Wittink H, Peeters

PH and Monninkhof EM. Effect of weight loss, with or without exercise, on body

93

composition and sex hormones in postmenopausal women: the SHAPE-2 trial.

Breast cancer research : BCR. 2015;17:120. 4557857.

34. Smith RN, Mann NJ, Braue A, Makelainen H and Varigos GA. The effect of a high-

protein, low glycemic-load diet versus a conventional, high glycemic-load diet on

biochemical parameters associated with acne vulgaris: a randomized, investigator-

masked, controlled trial. Journal of the American Academy of Dermatology.

2007;57:247-56.

35. Longcope C, Feldman HA, McKinlay JB and Araujo AB. Diet and sex hormone-

binding globulin. The Journal of clinical endocrinology and metabolism.

2000;85:293-6.

36. Newcomb PA, Klein R, Klein BE, Haffner S, Mares-Perlman J, Cruickshanks KJ

and Marcus PM. Association of dietary and life-style factors with sex hormones in

postmenopausal women. Epidemiology. 1995;6:318-21.

37. Selva DM, Hogeveen KN, Innis SM and Hammond GL. Monosaccharide-induced

lipogenesis regulates the human hepatic sex hormone-binding globulin gene. The

Journal of clinical investigation. 2007;117:3979-87. 2066187.

38. Design of the Women's Health Initiative clinical trial and observational study. The

Women's Health Initiative Study Group. Controlled clinical trials. 1998;19:61-109.

39. Anderson GL, Manson J, Wallace R, Lund B, Hall D, Davis S, Shumaker S, Wang

CY, Stein E and Prentice RL. Implementation of the Women's Health Initiative

study design. Annals of epidemiology. 2003;13:S5-17.

94

40. Patterson RE, Kristal AR, Tinker LF, Carter RA, Bolton MP and Agurs-Collins T.

Measurement characteristics of the Women's Health Initiative food frequency

questionnaire. Annals of epidemiology. 1999;9:178-87.

41. Henderson MM, Kushi LH, Thompson DJ, Gorbach SL, Clifford CK, Insull W, Jr.,

Moskowitz M and Thompson RS. Feasibility of a randomized trial of a low-fat diet

for the prevention of breast cancer: dietary compliance in the Women's Health Trial

Vanguard Study. Preventive medicine. 1990;19:115-33.

42. Kristal AR, Feng Z, Coates RJ, Oberman A and George V. Associations of

race/ethnicity, education, and dietary intervention with the validity and reliability

of a food frequency questionnaire: the Women's Health Trial Feasibility Study in

Minority Populations. American journal of epidemiology. 1997;146:856-69.

43. Block G, Woods M, Potosky A and Clifford C. Validation of a self-administered

diet history questionnaire using multiple diet records. Journal of clinical

epidemiology. 1990;43:1327-35.

44. Schakel SF, Sievert YA and Buzzard IM. Sources of data for developing and

maintaining a nutrient database. Journal of the American Dietetic Association.

1988;88:1268-71.

45. Neuhouser ML, Tinker LF, Thomson C, Caan B, Horn LV, Snetselaar L, Parker

LM, Patterson RE, Robinson-O'Brien R, Beresford SA and Shikany JM.

Development of a glycemic index database for food frequency questionnaires used

in epidemiologic studies. The Journal of nutrition. 2006;136:1604-9.

95

46. Benjamini Y and Hochberg Y. Controlling the False Discovery Rate - a Practical

and Powerful Approach to Multiple Testing. J Roy Stat Soc B Met. 1995;57:289-

300.

47. Tingley D, Yamamoto T, Hirose K, Keele L and Imai K. mediation: R Package for

Causal Mediation Analysis. J Stat Softw. 2014;59.

48. Imai K, Keele L and Tingley D. A general approach to causal mediation analysis.

Psychological methods. 2010;15:309-34.

49. Imai K, Keele L and Yamamoto T. Identification, Inference and Sensitivity

Analysis for Causal Mediation Effects. Stat Sci. 2010;25:51-71.

50. R Core Team. R: A language and environment for statistical computing. R

Foundation for Statistical Computing, Vienna, Austria. 2016.

51. Reynolds RC, Lee S, Choi JY, Atkinson FS, Stockmann KS, Petocz P and Brand-

Miller JC. Effect of the glycemic index of carbohydrates on Acne vulgaris.

Nutrients. 2010;2:1060-72. 3257617.

52. Augustin LS, Galeone C, Dal Maso L, Pelucchi C, Ramazzotti V, Jenkins DJ,

Montella M, Talamini R, Negri E, Franceschi S and La Vecchia C. Glycemic index,

glycemic load and risk of prostate cancer. International journal of cancer.

2004;112:446-50.

53. Strain G, Zumoff B, Rosner W and Pi-Sunyer X. The relationship between serum

levels of insulin and sex hormone-binding globulin in men: the effect of weight loss.

The Journal of clinical endocrinology and metabolism. 1994;79:1173-6.

96

54. Daka B, Rosen T, Jansson PA, Rastam L, Larsson CA and Lindblad U. Inverse

association between serum insulin and sex hormone-binding globulin in a

population survey in Sweden. Endocrine connections. 2013;2:18-22. 3680959.

55. Prentice RL, Caan B, Chlebowski RT, Patterson R, Kuller LH, Ockene JK,

Margolis KL, Limacher MC, Manson JE, Parker LM, Paskett E, Phillips L, Robbins

J, Rossouw JE, Sarto GE, Shikany JM, Stefanick ML, Thomson CA, Van Horn L,

Vitolins MZ, Wactawski-Wende J, Wallace RB, Wassertheil-Smoller S, Whitlock

E, Yano K, Adams-Campbell L, Anderson GL, Assaf AR, Beresford SA, Black HR,

Brunner RL, Brzyski RG, Ford L, Gass M, Hays J, Heber D, Heiss G, Hendrix SL,

Hsia J, Hubbell FA, Jackson RD, Johnson KC, Kotchen JM, LaCroix AZ, Lane DS,

Langer RD, Lasser NL and Henderson MM. Low-fat dietary pattern and risk of

invasive breast cancer: the Women's Health Initiative Randomized Controlled

Dietary Modification Trial. Jama. 2006;295:629-42.

56. Monroe KR, Murphy SP, Henderson BE, Kolonel LN, Stanczyk FZ, Adlercreutz H

and Pike MC. Dietary fiber intake and endogenous serum hormone levels in

naturally postmenopausal Mexican American women: the Multiethnic Cohort

Study. Nutrition and cancer. 2007;58:127-35.

57. Reimer RA and McBurney MI. Dietary fiber modulates intestinal proglucagon

messenger ribonucleic acid and postprandial secretion of glucagon-like peptide-1

and insulin in rats. Endocrinology. 1996;137:3948-56.

58. Soh NL and Brand-Miller J. The glycaemic index of potatoes: the effect of variety,

cooking method and maturity. European journal of clinical nutrition. 1999;53:249-

54.

97

59. Jenkins DJ, Wolever TM, Jenkins AL, Lee R, Wong GS and Josse R. Glycemic

response to wheat products: reduced response to pasta but no effect of fiber.

Diabetes care. 1983;6:155-9.

60. Rasmussen O, Winther E, Arnfred J and Hermansen K. Comparison of blood

glucose and insulin responses in non-insulin dependent diabetic patients. Studies

with spaghetti and potato taken alone and as part of a mixed meal. European journal

of clinical nutrition. 1988;42:953-61.

61. Hermansen K, Rasmussen O, Arnfred J, Winther E and Schmitz O. Differential

glycaemic effects of potato, rice and spaghetti in type 1 (insulin-dependent) diabetic

patients at constant insulinaemia. Diabetologia. 1986;29:358-61.

62. Juntunen KS, Niskanen LK, Liukkonen KH, Poutanen KS, Holst JJ and Mykkanen

HM. Postprandial glucose, insulin, and incretin responses to grain products in

healthy subjects. Am J Clin Nutr. 2002;75:254-62.

63. Wolever TM, Jenkins DJ, Kalmusky J, Giordano C, Giudici S, Jenkins AL,

Thompson LU, Wong GS and Josse RG. Glycemic response to pasta: effect of

surface area, degree of cooking, and protein enrichment. Diabetes care.

1986;9:401-4.

64. Granfeldt Y, Bjorck I and Hagander B. On the importance of processing conditions,

product thickness and egg addition for the glycaemic and hormonal responses to

pasta: a comparison with bread made from 'pasta ingredients'. European journal of

clinical nutrition. 1991;45:489-99.

65. Juanola-Falgarona M, Salas-Salvado J, Buil-Cosiales P, Corella D, Estruch R, Ros

E, Fito M, Recondo J, Gomez-Gracia E, Fiol M, Lapetra J, Lamuela-Raventos RM,

98

Serra-Majem L, Pinto X, Munoz MA, Ruiz-Gutierrez V, Alfredo Martinez J,

Castro-Quezada I and Bullo M. Dietary Glycemic Index and Glycemic Load Are

Positively Associated with Risk of Developing Metabolic Syndrome in Middle-

Aged and Elderly Adults. Journal of the American Geriatrics Society.

2015;63:1991-2000.

66. Salmeron J, Ascherio A, Rimm EB, Colditz GA, Spiegelman D, Jenkins DJ,

Stampfer MJ, Wing AL and Willett WC. Dietary fiber, glycemic load, and risk of

NIDDM in men. Diabetes care. 1997;20:545-50.

67. Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL and Willett WC.

Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in

women. Jama. 1997;277:472-7.

68. Schulze MB, Liu S, Rimm EB, Manson JE, Willett WC and Hu FB. Glycemic index,

glycemic load, and dietary fiber intake and incidence of type 2 diabetes in younger

and middle-aged women. Am J Clin Nutr. 2004;80:348-56.

69. Livesey G, Taylor R, Livesey H and Liu S. Is there a dose-response relation of

dietary glycemic load to risk of type 2 diabetes? Meta-analysis of prospective

cohort studies. Am J Clin Nutr. 2013;97:584-96.

70. Bhupathiraju SN, Tobias DK, Malik VS, Pan A, Hruby A, Manson JE, Willett WC

and Hu FB. Glycemic index, glycemic load, and risk of type 2 diabetes: results from

3 large US cohorts and an updated meta-analysis. Am J Clin Nutr. 2014;100:218-

32. 4144100.

99

71. Barclay AW, Petocz P, McMillan-Price J, Flood VM, Prvan T, Mitchell P and

Brand-Miller JC. Glycemic index, glycemic load, and chronic disease risk--a meta-

analysis of observational studies. Am J Clin Nutr. 2008;87:627-37.

72. Mirrahimi A, de Souza RJ, Chiavaroli L, Sievenpiper JL, Beyene J, Hanley AJ,

Augustin LS, Kendall CW and Jenkins DJ. Associations of glycemic index and load

with coronary heart disease events: a systematic review and meta-analysis of

prospective cohorts. Journal of the American Heart Association. 2012;1:e000752.

3541617.

73. Yu D, Shu XO, Li H, Xiang YB, Yang G, Gao YT, Zheng W and Zhang X. Dietary

carbohydrates, refined grains, glycemic load, and risk of coronary heart disease in

Chinese adults. American journal of epidemiology. 2013;178:1542-9. 3888273.

74. Hardy DS, Hoelscher DM, Aragaki C, Stevens J, Steffen LM, Pankow JS and

Boerwinkle E. Association of glycemic index and glycemic load with risk of

incident coronary heart disease among Whites and African Americans with and

without type 2 diabetes: the Atherosclerosis Risk in Communities study. Annals of

epidemiology. 2010;20:610-6. 3085981.

75. Sieri S, Krogh V, Berrino F, Evangelista A, Agnoli C, Brighenti F, Pellegrini N,

Palli D, Masala G, Sacerdote C, Veglia F, Tumino R, Frasca G, Grioni S, Pala V,

Mattiello A, Chiodini P and Panico S. Dietary glycemic load and index and risk of

coronary heart disease in a large italian cohort: the EPICOR study. Archives of

internal medicine. 2010;170:640-7.

100

76. Fan J, Song Y, Wang Y, Hui R and Zhang W. Dietary glycemic index, glycemic

load, and risk of coronary heart disease, stroke, and stroke mortality: a systematic

review with meta-analysis. PLoS One. 2012;7:e52182. 3527433.

77. Sieri S, Brighenti F, Agnoli C, Grioni S, Masala G, Bendinelli B, Sacerdote C,

Ricceri F, Tumino R, Giurdanella MC, Pala V, Berrino F, Mattiello A, Chiodini P,

Panico S and Krogh V. Dietary glycemic load and glycemic index and risk of

cerebrovascular disease in the EPICOR cohort. PLoS One. 2013;8:e62625.

PMC3662695.

78. Kaushik S, Wang JJ, Wong TY, Flood V, Barclay A, Brand-Miller J and Mitchell

P. Glycemic index, retinal vascular caliber, and stroke mortality. Stroke; a journal

of cerebral circulation. 2009;40:206-12.

79. Shikany JM, Tinker LF, Neuhouser ML, Ma Y, Patterson RE, Phillips LS, Liu S

and Redden DT. Association of glycemic load with cardiovascular disease risk

factors: the Women's Health Initiative Observational Study. Nutrition.

2010;26:641-7. 2866821.

80. Ford ES and Liu S. Glycemic index and serum high-density lipoprotein cholesterol

concentration among us adults. Archives of internal medicine. 2001;161:572-6.

81. Denova-Gutierrez E, Huitron-Bravo G, Talavera JO, Castanon S, Gallegos-Carrillo

K, Flores Y and Salmeron J. Dietary glycemic index, dietary glycemic load, blood

lipids, and coronary heart disease. Journal of nutrition and metabolism. 2010;2010.

2911609.

82. Goff DC, Jr., Lloyd-Jones DM, Bennett G, Coady S, D'Agostino RB, Sr., Gibbons

R, Greenland P, Lackland DT, Levy D, O'Donnell CJ, Robinson JG, Schwartz JS,

101

Shero ST, Smith SC, Jr., Sorlie P, Stone NJ, Wilson PW and American College of

Cardiology/American Heart Association Task Force on Practice G. 2013

ACC/AHA guideline on the assessment of cardiovascular risk: a report of the

American College of Cardiology/American Heart Association Task Force on

Practice Guidelines. J Am Coll Cardiol. 2014;63:2935-59. PMC4700825.

83. Goff DC, Jr., Lloyd-Jones DM, Bennett G, Coady S, D'Agostino RB, Gibbons R,

Greenland P, Lackland DT, Levy D, O'Donnell CJ, Robinson JG, Schwartz JS,

Shero ST, Smith SC, Jr., Sorlie P, Stone NJ, Wilson PW, Jordan HS, Nevo L, Wnek

J, Anderson JL, Halperin JL, Albert NM, Bozkurt B, Brindis RG, Curtis LH,

DeMets D, Hochman JS, Kovacs RJ, Ohman EM, Pressler SJ, Sellke FW, Shen

WK, Smith SC, Jr., Tomaselli GF and American College of Cardiology/American

Heart Association Task Force on Practice G. 2013 ACC/AHA guideline on the

assessment of cardiovascular risk: a report of the American College of

Cardiology/American Heart Association Task Force on Practice Guidelines.

Circulation. 2014;129:S49-73.

84. Grambsch PM and Therneau TM. Proportional Hazards Tests and Diagnostics

Based on Weighted Residuals. Biometrika. 1994;81:515-526.

85. Guenther PM, Reedy J and Krebs-Smith SM. Development of the Healthy Eating

Index-2005. J Am Diet Assoc. 2008;108:1896-901.

86. Pounis G, Castelnuovo AD, Costanzo S, Persichillo M, Bonaccio M, Bonanni A,

Cerletti C, Donati MB, de Gaetano G and Iacoviello L. Association of pasta

consumption with body mass index and waist-to-hip ratio: results from Moli-sani

and INHES studies. Nutrition & diabetes. 2016;6:e218. 4973136.

102

87. Estruch R, Ros E, Salas-Salvado J, Covas MI, Corella D, Aros F, Gomez-Gracia E,

Ruiz-Gutierrez V, Fiol M, Lapetra J, Lamuela-Raventos RM, Serra-Majem L, Pinto

X, Basora J, Munoz MA, Sorli JV, Martinez JA, Martinez-Gonzalez MA and

Investigators PS. Primary prevention of cardiovascular disease with a

Mediterranean diet. N Engl J Med. 2013;368:1279-90.

88. Martinez-Gonzalez MA, Salas-Salvado J, Estruch R, Corella D, Fito M, Ros E and

Predimed I. Benefits of the Mediterranean Diet: Insights From the PREDIMED

Study. Prog Cardiovasc Dis. 2015;58:50-60.

89. Shay CM, Van Horn L, Stamler J, Dyer AR, Brown IJ, Chan Q, Miura K, Zhao L,

Okuda N, Daviglus ML, Elliott P and Group IR. Food and nutrient intakes and their

associations with lower BMI in middle-aged US adults: the International Study of

Macro-/Micronutrients and Blood Pressure (INTERMAP). Am J Clin Nutr.

2012;96:483-91. PMC3417211.

90. Chiavaroli L, Kendall CWC, Braunstein CR, Blanco Mejia S, Leiter LA, Jenkins

DJA and Sievenpiper JL. Effect of pasta in the context of low-glycaemic index

dietary patterns on body weight and markers of adiposity: a systematic review and

meta-analysis of randomised controlled trials in adults. BMJ Open. 2018;8:e019438.

PMC5884373.

91. Huang M, Li J, Ha MA, Riccardi G and Liu S. A systematic review on the relations

between pasta consumption and cardio-metabolic risk factors. Nutr Metab

Cardiovasc Dis. 2017;27:939-948.

92. Shaw JE, Sicree RA and Zimmet PZ. Global estimates of the prevalence of diabetes

for 2010 and 2030. Diabetes research and clinical practice. 2010;87:4-14.

103

93. Ye EQ, Chacko SA, Chou EL, Kugizaki M and Liu S. Greater whole-grain intake

is associated with lower risk of type 2 diabetes, cardiovascular disease, and weight

gain. The Journal of nutrition. 2012;142:1304-13.

94. Hu FB. Resolved: there is sufficient scientific evidence that decreasing sugar-

sweetened beverage consumption will reduce the prevalence of obesity and obesity-

related diseases. Obesity reviews : an official journal of the International

Association for the Study of Obesity. 2013;14:606-19.

95. Qi Q, Chu AY, Kang JH, Jensen MK, Curhan GC, Pasquale LR, Ridker PM, Hunter

DJ, Willett WC, Rimm EB, Chasman DI, Hu FB and Qi L. Sugar-sweetened

beverages and genetic risk of obesity. N Engl J Med. 2012;367:1387-96. 3518794.

96. Qi Q, Kilpelainen TO, Downer MK, Tanaka T, Smith CE, Sluijs I, Sonestedt E,

Chu AY, Renstrom F, Lin X, Angquist LH, Huang J, Liu Z, Li Y, Asif Ali M, Xu

M, Ahluwalia TS, Boer JM, Chen P, Daimon M, Eriksson J, Perola M, Friedlander

Y, Gao YT, Heppe DH, Holloway JW, Houston DK, Kanoni S, Kim YM,

Laaksonen MA, Jaaskelainen T, Lee NR, Lehtimaki T, Lemaitre RN, Lu W, Luben

RN, Manichaikul A, Mannisto S, Marques-Vidal P, Monda KL, Ngwa JS, Perusse

L, van Rooij FJ, Xiang YB, Wen W, Wojczynski MK, Zhu J, Borecki IB, Bouchard

C, Cai Q, Cooper C, Dedoussis GV, Deloukas P, Ferrucci L, Forouhi NG, Hansen

T, Christiansen L, Hofman A, Johansson I, Jorgensen T, Karasawa S, Khaw KT,

Kim MK, Kristiansson K, Li H, Liu Y, Lohman KK, Long J, Mikkila V,

Mozaffarian D, North K, Pedersen O, Raitakari O, Rissanen H, Tuomilehto J, van

der Schouw YT, Uitterlinden AG, Zillikens MC, Franco OH, Shyong Tai E, Ou

Shu X, Siscovick DS, Toft U, Verschuren WM, Vollenweider P, Wareham NJ,

104

Witteman JC, Zheng W, Ridker PM, Kang JH, Liang L, Jensen MK, Curhan GC,

Pasquale LR, Hunter DJ, Mohlke KL, Uusitupa M, Cupples LA, Rankinen T, Orho-

Melander M, Wang T, Chasman DI, Franks PW, Sorensen TI, Hu FB, Loos RJ,

Nettleton JA and Qi L. FTO genetic variants, dietary intake and body mass index:

insights from 177,330 individuals. Human molecular genetics. 2014;23:6961-72.

4271061.

97. Sotos-Prieto M, Luben R, Khaw KT, Wareham NJ and Forouhi NG. The

association between Mediterranean Diet Score and glucokinase regulatory protein

gene variation on the markers of cardiometabolic risk: an analysis in the European

Prospective Investigation into Cancer (EPIC)-Norfolk study. The British journal of

nutrition. 2014;112:122-31. 4174631.

98. Qi Q, Downer MK, Kilpelainen TO, Taal HR, Barton SJ, Ntalla I, Standl M,

Boraska V, Huikari V, Kiefte-de Jong JC, Korner A, Lakka TA, Liu G, Magnusson

J, Okuda M, Raitakari O, Richmond R, Scott RA, Bailey ME, Scheuermann K,

Holloway JW, Inskip H, Isasi CR, Mossavar-Rahmani Y, Jaddoe VW, Laitinen J,

Lindi V, Melen E, Pitsiladis Y, Pitkanen N, Snieder H, Heinrich J, Timpson NJ,

Wang T, Yuji H, Zeggini E, Dedoussis GV, Kaplan RC, Wylie-Rosett J, Loos RJ,

Hu FB and Qi L. Dietary Intake, FTO Genetic Variants, and Adiposity: A

Combined Analysis of Over 16,000 Children and Adolescents. Diabetes.

2015;64:2467-76.

99. Sonestedt E, Hellstrand S, Schulz CA, Wallstrom P, Drake I, Ericson U, Gullberg

B, Hedblad B and Orho-Melander M. The association between carbohydrate-rich

foods and risk of cardiovascular disease is not modified by genetic susceptibility to

105

dyslipidemia as determined by 80 validated variants. PLoS One. 2015;10:e0126104.

4405383.

100. Villegas R, Goodloe RJ, McClellan BE, Jr., Boston J and Crawford DC. Gene-

carbohydrate and gene-fiber interactions and type 2 diabetes in diverse populations

from the National Health and Nutrition Examination Surveys (NHANES) as part

of the Epidemiologic Architecture for Genes Linked to Environment (EAGLE)

study. BMC genetics. 2014;15:69. 4094781.

101. Palmer CN, Kimber CH, Doney AS, Proia AS, Morris AD, Gaetani E, Quarta M,

Smith RC and Pola R. Combined effect of inflammatory gene polymorphisms and

the risk of ischemic stroke in a prospective cohort of subjects with type 2 diabetes:

a Go-DARTS study. Diabetes. 2010;59:2945-8. 2963555.

102. Block G, Woods M, Potosky A and Clifford C. Validation of a Self-Administered

Diet History Questionnaire Using Multiple Diet Records. J Clin Epidemiol.

1990;43:1327-1335.

103. Schakel SF, Sievert YA and Buzzard IM. Sources of Data for Developing and

Maintaining a Nutrient Database. J Am Diet Assoc. 1988;88:1268-1271.

104. Neuhouser ML, Tinker LF, Thomson C, Caan B, Van Horn L, Snetselaar L, Parker

LM, Patterson RE, Robinson-O'Brien R, Beresford SAA and Shikany JM.

Development of a glycemic index database for food frequency questionnaires used

in epidemiologic studies. J Nutr. 2006;136:1604-1609.

105. Liu EY, Buyske S, Aragaki AK, Peters U, Boerwinkle E, Carlson C, Carty C,

Crawford DC, Haessler J, Hindorff LA, Marchand LL, Manolio TA, Matise T,

Wang W, Kooperberg C, North KE and Li Y. Genotype imputation of Metabochip

106

SNPs using a study-specific reference panel of ~4,000 haplotypes in African

Americans from the Women's Health Initiative. Genetic epidemiology.

2012;36:107-17. 3410659.

106. MacArthur J BE, Cerezo M, Gil L, Hall P, Hastings E, Junkins H, McMahon A,

Milano A, Morales J, Pendlington Z, Welter D, Burdett T, Hindorff L, Flicek P,

Cunningham F, and Parkinson H. The new NHGRI-EBI Catalog of published

genome-wide association studies (GWAS Catalog). Nucleic Acids Research, 2017,

Vol. 45 (Database issue): D896-D901.

107. Burdett T (EBI) HPN, Hastings E (EBI), Hindorff LA (NHGRI), Junkins HA

(NHGRI), Klemm AK (NHGRI), MacArthur J (EBI), Manolio TA (NHGRI),

Morales J (EBI), Parkinson H (EBI) and Welter D (EBI). The NHGRI-EBI Catalog

of published genome-wide association studies. Available at: www.ebi.ac.uk/gwas.

Accessed December 11, 2017, version v1.0.1.

108. Keku E, Rosamond W, Taylor HA, Jr., Garrison R, Wyatt SB, Richard M, Jenkins

B, Reeves L and Sarpong D. Cardiovascular disease event classification in the

Jackson Heart Study: methods and procedures. Ethnicity & disease. 2005;15:S6-

62-70.

109. Knol MJ, VanderWeele TJ, Groenwold RH, Klungel OH, Rovers MM and Grobbee

DE. Estimating measures of interaction on an additive scale for preventive

exposures. Eur J Epidemiol. 2011;26:433-8. PMC3115067.

110. Bass JDSwcfAJ, Dabney A and Robinson D (2015). qvalue: Q-value estimation for

false discovery rate control. R package version 2.10.0,

http://github.com/jdstorey/qvalue.

www.ebi.ac.uk/gwas

http://github.com/jdstorey/qvalue

107

111. Pederson RA and McIntosh CH. Discovery of gastric inhibitory polypeptide and its

subsequent fate: Personal reflections. J Diabetes Investig. 2016;7 Suppl 1:4-7.

PMC4854497.

112. Harrison KA, Druey KM, Deguchi Y, Tuscano JM and Kehrl JH. A novel human

homeobox gene distantly related to proboscipedia is expressed in lymphoid and

pancreatic tissues. J Biol Chem. 1994;269:19968-75.

113. Jiang W, Shi Y, Zhao D, Chen S, Yong J, Zhang J, Qing T, Sun X, Zhang P, Ding

M, Li D and Deng H. In vitro derivation of functional insulin-producing cells from

human embryonic stem cells. Cell Res. 2007;17:333-44.

dietary carbohydrates, associated biomarker, and genetic

Documents