<%BANNER%>

Education Differences in Elevated Blood Glucose

Permanent Link: http://ufdc.ufl.edu/UFE0025059/00001

Material Information

Title: Education Differences in Elevated Blood Glucose Do They Vary by Race, Ethnicity and Sex?
Physical Description: 1 online resource (94 p.)
Language: english
Creator: Pavela, Gregory
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: diabetes, disparities, education, ethnicity, health, race
Sociology -- Dissertations, Academic -- UF
Genre: Sociology thesis, M.A.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The relationship between education and health has been well documented--increasing levels of education are associated with better health using various definitions of both education and health. This relationship has also been found to vary across demographic profiles. This study examines the relationship between education and blood glucose levels, whether the association varies across race and ethnicity and sex, and whether the association varies by level of blood glucose. Elevated levels of blood glucose can been classified into two stages: impaired fasting glucose (fasting glucose levels 100 to 125 mg/dL (5.6 to 6.9 mmol/L)) and diabetes (fasting glucose levels over 125 mg/dL(greater than or equal to 7.0 mmol/L)), with each stage associated with an array of health risks. This research is important because individuals with impaired fasting glucose are at increased risk for developing diabetes, and diabetics have an increased risk for developing cardiovascular disease--the leading cause of death in the United States. Data from the combined 1999-2006 National Health and Nutrition Examination Survey are analyzed. Logistic regression is used to test the relationship between level of blood glucose and whether the association varies across sex and race and ethnicity. A continuation ratio model is used to test for difference in effects of covariates between the two stages of blood glucose. Results indicate that education has a significant association with blood glucose levels, that this association is significantly different for males and females, that the association varies by race and ethnicity for females, and that the association with race and ethnicity varies by level of blood sugar for both males and females. For males, education has no effect on blood sugar levels after adjusting for age, B.M.I. and marital status. Having more than a high school degree is significantly associated with reduced levels of blood sugar for females. This effect is significantly different for men and women. Allowing for interactions between education and race and ethnicity suggest that blacks with less than a 9th grade education have a significantly reduced probability of developing elevated levels of blood glucose relative to other blacks, and being black has the effect of reducing the probability of having elevated blood sugar overall. Hispanics with a BA or more have a significantly increased probability of developing elevated blood glucose relative to other Hispanics, but being Hispanic has no significant association with elevated blood sugar levels. The association between race and blood glucose varies across levels of blood glucose for both men and women. For males, given that one has EBS, being either black or Hispanic significantly increases the risk of developing diabetes. For females, given that an individual has EBS, being black significantly increases the risk of developing diabetes. The race and ethnicity and blood glucose association is significantly different across stages of blood glucose for both males and females, changing from having either no effect or a weak protective effect on development of elevated blood glucose to having a significant effect on increasing the risk of developing diabetes, given EBS.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gregory Pavela.
Thesis: Thesis (M.A.)--University of Florida, 2009.
Local: Adviser: Henretta, John C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0025059:00001

Permanent Link: http://ufdc.ufl.edu/UFE0025059/00001

Material Information

Title: Education Differences in Elevated Blood Glucose Do They Vary by Race, Ethnicity and Sex?
Physical Description: 1 online resource (94 p.)
Language: english
Creator: Pavela, Gregory
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: diabetes, disparities, education, ethnicity, health, race
Sociology -- Dissertations, Academic -- UF
Genre: Sociology thesis, M.A.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The relationship between education and health has been well documented--increasing levels of education are associated with better health using various definitions of both education and health. This relationship has also been found to vary across demographic profiles. This study examines the relationship between education and blood glucose levels, whether the association varies across race and ethnicity and sex, and whether the association varies by level of blood glucose. Elevated levels of blood glucose can been classified into two stages: impaired fasting glucose (fasting glucose levels 100 to 125 mg/dL (5.6 to 6.9 mmol/L)) and diabetes (fasting glucose levels over 125 mg/dL(greater than or equal to 7.0 mmol/L)), with each stage associated with an array of health risks. This research is important because individuals with impaired fasting glucose are at increased risk for developing diabetes, and diabetics have an increased risk for developing cardiovascular disease--the leading cause of death in the United States. Data from the combined 1999-2006 National Health and Nutrition Examination Survey are analyzed. Logistic regression is used to test the relationship between level of blood glucose and whether the association varies across sex and race and ethnicity. A continuation ratio model is used to test for difference in effects of covariates between the two stages of blood glucose. Results indicate that education has a significant association with blood glucose levels, that this association is significantly different for males and females, that the association varies by race and ethnicity for females, and that the association with race and ethnicity varies by level of blood sugar for both males and females. For males, education has no effect on blood sugar levels after adjusting for age, B.M.I. and marital status. Having more than a high school degree is significantly associated with reduced levels of blood sugar for females. This effect is significantly different for men and women. Allowing for interactions between education and race and ethnicity suggest that blacks with less than a 9th grade education have a significantly reduced probability of developing elevated levels of blood glucose relative to other blacks, and being black has the effect of reducing the probability of having elevated blood sugar overall. Hispanics with a BA or more have a significantly increased probability of developing elevated blood glucose relative to other Hispanics, but being Hispanic has no significant association with elevated blood sugar levels. The association between race and blood glucose varies across levels of blood glucose for both men and women. For males, given that one has EBS, being either black or Hispanic significantly increases the risk of developing diabetes. For females, given that an individual has EBS, being black significantly increases the risk of developing diabetes. The race and ethnicity and blood glucose association is significantly different across stages of blood glucose for both males and females, changing from having either no effect or a weak protective effect on development of elevated blood glucose to having a significant effect on increasing the risk of developing diabetes, given EBS.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Gregory Pavela.
Thesis: Thesis (M.A.)--University of Florida, 2009.
Local: Adviser: Henretta, John C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0025059:00001


This item has the following downloads:


Full Text





EDUCATION DIFFERENCES IN ELEVATED BLOOD GLUCOSE: DO THEY VARY BY
RACE, ETHNICITY AND SEX?



















By

GREGORY PAVELA


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ARTS

UNIVERSITY OF FLORIDA

2009



































2009 Gregory Pavela


































To my parents









ACKNOWLEDGMENTS

I must thank my committee, and especially Dr. Henretta, for the inordinate amount of time

given to me and to this thesis. Dr. Henretta's help extends beyond careful attention to detail in

the titles of tables, formatting, and logic of argument and reaches all the way to handwritten

Statistical Analysis Software syntax to introduce me to the program. The process of writing this

thesis, in preparation for future work, has been the best and most useful part of my education

thus far. I can only hope that a little bit of Dr. Henretta's approach to problem solving has rubbed

off and ill improve my answers to future questions.










TABLE OF CONTENTS



A C K N O W L E D G M E N T S ..............................................................................................................4

LIST OF TABLES .............. ................ ........._..7.. ..........

LIST OF FIGURES ...................................................... .........8

ABSTRAC T ............................. ......... ................................9

CHAPTER

1 H E A L TH D ISPA R ITIE S ................................................................................................. 11

Introduction........................ ......... ......... ................ ................ 11
Conceptualizing Health Disparities .......................................13
Socioeconomic Status Health Disparities in Elevated Blood Sugar................ ...............15
Framework Linking Socioeconomic Status to EBS .................... .. ......16
Education as a Socioeconomic Status Measure and General Health Outcomes ..................17
Measuring Education. .. ................ ......... .................. ................. 19
Pathways Between Education and Health ................................. ...............21
E education and E ndogeneity .............................................................23
Fram work Linking Education to EBS.................................... .................... 24
Non-linear Effects of Education .................................................................................................25
Theories of Differential Effects ..................................................................................................25
G enetic and Early Life Effects ............................................... ............... 25
Selection and Personality D ifferences................................................................... 26
C cultural C capital D differences ....................................................................................... 27
Stress.......................................................................28

2 BACKGROUND TO DIABETES AND IMPAIRED FASTING GLUCOSE ......................31

D iab ete s ........................................ ............... 3 1
Impaired Fasting Glucose (Pre-Diabetes).................. .................................................. 33
Link Between Impaired Fasting Glucose and Diabetes.............................. ........... ....33
A Model and Critical Covariates for Risk of Diabetes.....................................................35
Age and Sex..................... ........................... 35
Race and Ethnicity ................ ....................... ..................... ..... .. ..............35
Obesity......................................................... ................. ........36
Marital Status.................................. ........37

3 DATA AND M ETHODOLOGY ................................................. ...............42

Objectives and Research Questions.............................................................. ...............42
Hypotheses ................ .................................................. ........ 42










D ata ......... .. ........ .... .. ................ .................................................... . 4 3
Background.............. ......... ........ ............... ...... ........ 43
Sample and Selection Procedures................................. ...............43
V ariables ................ ........ .... .. .. .............. .............................. ........ 45
Elevated Blood Glucose Level ............................................... ............... 45
D diagnosed D diabetes ................ .............. .............. ..... ............. ............... 45
Independent Variables ............................................. .... ..............46
Education................................................ ................ ........46
Race ............................................................................46
Age ...........................................................................46
M marital Status......................................... 47
Body Mass Index ........................_. ..................................47
Methods: Continuation Ratio Model ..................................... ......... ........47

4 RESULTS AND ANALYSIS ............................... ...........49

Data Description..................................................... .........49
Results for Men................ ........................... .....................53
Males, Stage 1: Impaired Fasting Glucose .................................................54
Males, Stage 2: Diabetes .................................................... ........55
Stage Differences for Males ................................................56
Results for Females................................. ...................... .........58
Females, Stage 1: Impaired Fasting Glucose ....................................... ........58
Females, Stage 2: Diabetes................................ .................59
Stage Differences for Females.......................... ............. ........60
Sex Differences................. ...................................... ................. ........ 61
Stage One.............................................61
Stage Two............................... ............................................62

5 DISCUSSION AND CONCLUSION ........................................ ... ...............78

Discussion................................................ .........78
Conclusion ...................... ......... .............. ........................................................83

APPENDIX: ADDITIONAL EDUCATIONAL ATTAINMENT .........................................85

LIST OF REFEREN CES ................................................................................. 87

BIOGRAPHICAL SKETCH ...................................................... ........93













6









LIST OF TABLES


Table page

1-1 Undiagnosed Diabetes By Educational Attainment and Race ....................................29

2-1 Obesity and Diabetes Prevalence Among U.S. Adults, by Sex and Age...........................41

4-1 Frequencies and Prevalence of Impaired Fasting Glucose* for Individuals Aged 20
and Over..................................... .................. .........64

4-2 Frequencies and Prevalence (Standard Error) of Diagnosed Diabetes for
Individuals Aged 20 and Over .............. .............. ..................... ......65

4-3 Prevalence (SE) of Diabetes as Measured from Either Diagnosis............... ................66

4-4 Prevalence (SE) of U.S. Population Aged 20 Years and Older with IFG* ................67

4-5 Prevalence (SE) of Diabetes in U.S. Population by Educational Attainment..................68

4-6 Proportions and Means of U.S. Born Population(SE) Over the Age of 24. ...................69

4-7 Results From a Logistic Regression Continuation Ratio Model for Males Greater
Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1)...............70

4-8 Results from a Logistic Regression Continuation Ratio Model for Males Greater
Than 24 Years Old for Having Diabetes Given EBS (Stage 2)................ ...............71

4-9 Results From a Logistic Regression Continuation Ratio Model for Females Greater
Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1)...............72

4-10 Results from a Logistic Regression Continuation Ratio Model for Feales Greater
Than 24 Years Old for Having Diabetes Given EBS (Stage 2)................ ...............73

A-i Percentage of Educational Attainment Within Five Category. Educational
Attainment and By Race/Ethncity in the U.S. Population.......................................85

A-2 Prevalence of Impaired Fasting Glucose (SE) of U.S. Born Population ...........................86









LIST OF FIGURES


Figure page

1-2 Conceptual Framework for the Association between Socioeconomic Status and
Elevated Blood Sugar, Adapted from Brown (2004).............................. ............ ...30

2-1 Conceptual Framework for Risk Factors for Diabetes, Adapted from Brown (2002). .....39

2-2 U.S. Diabetes Percentage by Age and Sex. Adapted from Centers for Disease
Control 2006 .........................................................40

4-1 Estimated Risk of Elevated Blood Sugar by Educational Attainment and Race For
M ales Over Age 45, Single, and Overweight ..................................... ......... ......74

4-2 Estimated Risk of Diabetes Given Elevated Blood Sugar by Educational Attainment
and Race For Males Over Age 45, Single, and Overweight............... .............75

4-3 Estimated Risk of Elevated Blood Sugar by Educational Attainment and Race For
Females Over Age 45, Single, and Overweight................................ ...........76

4-4 Estimated Risk of Diabetes Given Elevated Blood Sugar by Educational Attainment
and Race For Females Over Age 45, Single, and Overweight ......................................77











Abstract Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Master of Arts

EDUCATION DIFFERENCES IN ELEVATED BLOOD GLUCOSE: DO THEY VARY BY
RACE, ETHNICITY AND SEX

By

Gregory Pavela

AUGUST 2009

Chair: John Henretta
Major: Sociology

The relationship between education and health has been well documented--increasing

levels of education are associated with better health using various definitions of both education

and health. This relationship has also been found to vary across demographic profiles. This study

examines the relationship between education and blood glucose levels, whether the association

varies across race and ethnicity and sex, and whether the association varies by level of blood

glucose. Elevated levels of blood glucose can been classified into two stages: impaired fasting

glucose (fasting glucose levels 100 to 125 mg/dL (5.6 to 6.9 mmol/L)) and diabetes (fasting

glucose levels over 125 mg/dL(greater than or equal to 7.0 mmol/L)), with each stage associated

with an array of health risks. This research is important because individuals with impaired

fasting glucose are at increased risk for developing diabetes, and diabetics have an increased risk

for developing cardiovascular disease--the leading cause of death in the United States.

Data from the combined 1999-2006 National Health and Nutrition Examination Survey are

analyzed. Logistic regression is used to test the relationship between level of blood glucose and

whether the association varies across sex and race and ethnicity. A continuation ratio model is

used to test for difference in effects of covariates between the two stages of blood glucose.









Results indicate that education has a significant association with blood glucose levels, that

this association is significantly different for males and females, that the association varies by

race and ethnicity for females, and that the association with race and ethnicity varies by level of

blood sugar for both males and females.

For males, education has no effect on blood sugar levels after adjusting for age, body mass

index, and marital status. Having more than a high school degree is significantly associated with

reduced levels of blood sugar for females. This effect is significantly different for men and

women. Allowing for interactions between education and race and ethnicity suggest that blacks

with less than a 9th grade education have a significantly reduced probability of developing

elevated levels of blood glucose relative to other blacks, and being black has the effect of

reducing the probability of having elevated blood sugar overall. Hispanics with a BA or more

have a significantly increased probability of developing elevated blood glucose relative to other

Hispanics, but being Hispanic has no significant association with elevated blood sugar levels.

The association between race and blood glucose varies across levels of blood glucose for

both men and women. For males, given that one has EBS, being either black or Hispanic

significantly increases the risk of developing diabetes. For females, given that an individual has

EBS, being black significantly increases the risk of developing diabetes. The race and ethnicity

and blood glucose association is significantly different across stages of blood glucose for both

males and females, changing from having either no effect or a weak protective effect on

development of elevated blood glucose to having a significant effect on increasing the risk of

developing diabetes, given EBS.










CHAPTER 1
HEALTH DISPARITIES

Introduction

Differences in health outcomes by sex, race and socioeconomic status persist in American

society today. One important health outcome in which disparities exist is elevated blood sugar

(EBS), defined in this paper as having either impaired fasting glucose or diabetes (Borrell et. al.

2006). The significance of diabetes as a health outcome stems from its increasing incidence

within the total U.S. population, and its strong association with cardiovascular disease.

According to the Centers for Disease Control., the incidence of diabetes has increased 91% in the

past decade. (CDC 2008). The prevalence of diabetes has also been increasing. The CDC.reports

that from 1980 to 2006, the crude prevalence of diabetes has increased 132%, and this increase in

diabetes is similar regardless of age standardization, indicating that the increasing prevalence of

diabetes is not related to the changing age structure of the United States (CDC 2008). The

increase in prevalence could also be related to better treatment and longer survival times;

therefore incidence is a better measure to indicate that diabetes is of increasing concern.

While EBS is a health concern across the entire U.S. population, it affects different

segments of the population in different ways. Two key axes of differentiation in the blood

glucose disparity are race and ethnicity, and education. For example, Hispanics are twice as

likely to die from diabetes as are whites and those with higher levels of education are less likely

to have diabetes (Healthy People 2010, Borell 2006). Given the significance of studying diabetes

due to the increasing incidence of diabetes in the U.S. population and its differential impact on

key segments of the population, the objective of this research is to answer the following three

questions: (1) What is the association between education and the development of EBS; (2) does









this association vary by race and ethnicity and sex; and (3) does the association between race and

ethnicity, sex, and elevated blood sugar vary depending on the level of blood sugar?

In order to answer these questions and develop initial hypotheses, relevant health

disparities literature will first be reviewed. The definition of a health disparity and a framework

for general health outcomes will be provided. A key axis of health disparities is education, which

is a measure of socioeconomic status, thus socioeconomic disparities in health and EBS will be

reviewed. Following the outline of socioeconomic status disparities in health, socioeconomic

status will be integrated into the general framework of health outcomes (Figure 1-1) After the

general review of socioeconomic status and health disparities, education will become the focus

as a measure of socioeconomic status. The value of education as an socioeconomic status

measure, its measurement, its effects on EBS, the difficulties associated with studying the effects

of education on health, and theories of differential effects of education on health and EBS will

complete Chapter 1.

After reviewing health disparities by race and ethnicity and educational attainment, and

formulation of hypotheses in Chapter 1, a basic background to diabetes and impaired fasting

glucose (IFG) will be provided in Chapter 2, including their definition, incidence and prevalence,

and deleterious effects on the U.S. population. The empirical link between diabetes and IFG, as

well as the biological difference between the two will be outlined, followed by a discussion and

formulation of a model for the risk factors of EBS.

Once the critical covariates of EBS are reviewed in Chapter 2, Chapter 3 lists the

hypotheses formulated from the literature review of Chapters 1 and 2. Chapter 3 also discusses

the data and methodology used to test the hypotheses, including background to the NHANES,









sample analyzed, operationalizations of variables, and method of analysis are reviewed. Chapter

4 summarizes the results and Chapter 5 includes a discussion of the results and a conclusion.

Conceptualizing Health Disparities

A health disparity can be defined as "differences that occur by gender, race or ethnicity,

education or income, disability, geographic location, or sexual orientation (Healthy People

2010). However, this definition is only one among many definitions of what constitutes a

disparity (Smedley 2003, Whitehead 1991). Different definitions of what constitutes a health

disparity often involve differing foci of who or what is experiencing a disparity, such as race and

ethnicity or socioeconomic status. Most definitions of a disparity imply a reference group from

which comparisons are to be made. These comparisons can be made to the majority group, the

population mean, or to the healthiest group. (Adler 2008)

The difference between a disparity and a difference is not always clear, although a

disparity usually implies either something amenable to change or a form of injustice (Adler

2008, Herbert et al. 2008.). In this thesis I explicitly conceptualize the experience of racial and

ethnic minorities with impaired fasting glucose and diabetes as a disparity and not just a

difference. It is difficult to differentiate between a disparity and a difference in the case of EBS.

For example, genetic pre-dispositions towards development of diabetes in African-Americans

may be rooted in historical events. Biological explanations, commonly known as the "slavery

hypothesis", explain the higher incidence of hypertension in the African-American population

than in the general population by mortality conditions aboard slave transportation ships which

occasionally approached 20% ( Curtin 1992, Klein 2001). Such high mortality rates make trans-

Atlantic slave trading a plausible selective mechanism for individuals with certain metabolic

characteristics that may predispose them to metabolic diseases under other conditions. If this









explanation is valid, should the biological differences in diabetes be considered a genetic

difference, or a disparity rooted in unjust historical circumstances?

The causes of health disparities are numerous and interwoven with each other. Because

health itself is determined by numerous factors, including "physiological, psychological, cultural,

and social factors" (Goldberg 2004), the causes of health disparities can be found in the interplay

between these factors and the socio-historical context in which an individual or racial and ethnic

group exists. The U.S. is composed of numerous racial and ethnic groups, thus it is vitally

important to examine the extent and sources of racial and ethnic health disparities in the U.S.

population, keeping in mind that the goal of addressing health disparities is to improve the

overall health of society.

The complex factors to be taken into account when analyzing the sources of health

disparities can be organized using a general framework developed from combining models from

the Institute of Medicine and Office of Technology Assessment (Goldberg 2004). This

framework has three primary dimensions: health before care, health care access, and health care

delivery. Health before care refers to the variables that can influence health outside of the health

care system. It encompasses important factors that can influence health including individual

income, educational attainment, environment, personal characteristics, and overall social

conditions such as employment opportunities. Health before care is the focus of the analysis in

this thesis within the tri-partite framework for analyzing the origins of health disparities.

Health access refers to the ability of an individual to access and receive treatment for their

health condition. Language barriers, lack of financial resources, and a mistrust of the health care

system can all serve as barriers to affect health care access. Once one has gained access to the









system, Health care delivery is the third dimension on which health disparities may emerge and

includes variability of diagnosis, treatment, and communication.

Socioeconomic status is a key axis along which health behaviors, health care access and

delivery can vary. The next section discusses the role of socioeconomic status in EBS health

disparities.

Socioeconomic Status Health Disparities in Elevated Blood Sugar

Health disparities in diabetes by socioeconomic status have been well documented using

multiple measures of socioeconomic status, and the effects of socioeconomic status seem to vary

by sex (Smith 2007, Robbins et al. 2004. Stem et al. 1984). Those in the higher levels of

socioeconomic status indicators, including income, occupation, and education, tend to have

lower levels of diabetes. Along some socioeconomic status indicators, the gradient appears to be

growing (Smith 2007). Smith finds evidence that between National Health and Nutrition

Examination Survey II (NHANES) and NHANES IV, the gradient of diagnosed diabetes

between those with less than a high school degree and those with more than a high school degree

went from being insignificant to a gradient of diagnosed diabetes of about 4% (9.8% in the

lowest category of education vs. 6.0% in the highest category). Sex differences in the association

between socioeconomic status and diagnosed diabetes have been found for occupational status

but not for education (Robbins 2004). Robbins finds that higher levels of income, greater

occupational prestige, and higher levels of education were found to be associated with lower

probability of having diagnosed diabetes among women, but among men only education and

income were associated with lower probability of having diagnosed diabetes, not occupational

status.

As the socioeconomic status gradient in health appears to be widening for diabetes, the

proportion racial and ethnic groups with less than a high school degree is increasing. The









increasing proportions of racial and ethnic groups at lower levels of socioeconomic status may

undermine the success in equalizing the rates of undiagnosed diabetes between racial and ethnic

groups since the socioeconomic status gradient in undiagnosed diabetes is relatively constant

(Smith 2007). Table 1-1 reports undiagnosed diabetes rates in NHANES 1999-2006, as well as

"reference numbers" from Cowie (2002). Cowie analyzed the same data set as this study, so

Cowie's estimates serve as a comparison for estimates in this study. While Cowie's results differ

slightly from the results of this analysis, so do loannou's (2007) results without explanation.

Framework Linking Socioeconomic Status to Elevated Blood Sugar

A conceptual framework that links the socioeconomic status and diabetes should include

both the proximall" mechanisms such as health behaviors, access, and processes of care as

previously discussed, as well "distal" measures such as cultural patterns that mediate the

relationship between socioeconomic status and EBS and the proximal mechanisms that link

them. Brown et al. (2004) develop such a model, however it is intended to model the relationship

between socioeconomic status and health among persons with diabetes. This model, with some

adjustment, can also be used to link socioeconomic status with the development of EBS. Figure

1-1 displays the modified conceptual framework.

In the model developed by Brown et al. (2004), socioeconomic status encompasses

individual, household, and community characteristics that can shape the proximal mechanisms of

health behaviors, access to health care, and the process of health care. At the individual level

education, employment, and occupational prestige are likely to shape the proximal mediators

between socioeconomic status and health outcomes. At the household level, income and wealth

likely affect the proximal mediators for both adults and children. Finally, average community

income, education, and crime rates are a part of socioeconomic status that shape proximal

mediators.









Distal mediators and moderators are the effects of the characteristics of the individual,

provider, community, and health care system on the primary, proximal mechanisms. Distal

mediators may include level of acculturation and social-support at the individual level, language

concordance in patient-doctor relations at the provider level, and environmental safety at the

community level. Critical covariates to consider in any model with socioeconomic status are age,

sex, and race and ethnicity. As Brown (2004) acknowledges, an underlying assumption of their

model is a lack of endogeneity, or that socioeconomic status influences health, rather than health

influencing socioeconomic status. As will be discussed in the section on measuring education,

there are several issues to consider when modeling health outcomes. Early health events can

influence socioeconomic status attainment (Palloni 2006). Issues of endogeneity may also effect

health insurance, income, and occupational status. The model in this study does not consider the

effect of early life events on socioeconomic status attainment. To reduce the risk of endogeneity,

health insurance and income are excluded as possible explanatory variables between race and

ethnicity and EBS, so that the model being estimated is a reduced-form model.

While there are many possible measures of socioeconomic status, this analysis will focus

on education. The reasons for using education as a socioeconomic status proxy in health research

are outlined in the next section.

Education as a Socioeconomic Status Measure and General Health Outcomes

Higher levels of education are associated with better health, across multiple indicators of

health including including mortality, physical functioning, cardiovascular health, and cognitive

functioning (Kitagawa and Hauser 1973; Zimmer et al. 2002; Winkelby et al. 1992).

An early study on the relationship using mortality as a measurement of health was done by

Kitagawa and Hauser (Feinstein 1993). Using 1960 death certificates linked with census

information with census information, Kitagawa and Hauser measured the association between









levels of education and mortality. They found a strong inverse relationship up until the age of 65.

Over 65, the association between education and health was smaller. Recent evidence suggests

that the education gradient has been increasing in recent decades, with much of the gradient

driven by gains in life expectancy by the more highly educated in older age groups ( Meara and

Cutler 2008). During the past two decades, the life expectancy of the highly educated increased

by almost 3 years, compared to only half a year for the least educated. Increasing education

gradients, with the important exception of a narrowing education gradient among young black

men, occurred despite narrowing gradients across sex and ethnicity (Meara and Cutler 2008).

There is also evidence for sex differentials in the education gradient in mortality (Elo and

Preston 1995). Elo and Preston (1995) confirm the finding that education differentials tend

towards a maximum at older ages, but that these trends must be differentiated by sex. Females

aged 25-64 have experienced a recent narrowing of the education gradient in mortality and

females age 65 and over have experienced a static gradient. Men, however, have experienced a

broad pattern of widening education differentials in mortality since 1960 (Elo and Preston 1995).

Besides mortality, health can also be operationalized as physicalfunctioning, and here too,

research has shown a positive association with education. Zimmer et al. (2002) examined the

relative predictive power of both the parent's and the child's education on the level of reported

physical functioning of the parent, defined as reported difficulty in everyday tasks such as

sitting, crouching, and reaching for objects. They found that older adults who had more than a

primary level of education were "53% less likely to report a functional limitation in comparison

to those with low-level education" (Zimmer 2002).

Higher levels of education are also associated with lower risk of cardiovascular diseases

Using survey results from the Stanford Five-City project, Winkleby (1992) demonstrates that









those with the lowest levels of education tend to have the highest levels of cardiovascular risk

factors. For example, among men with less than a high school education, 47% reported being

cigarette smokers. Among men who had completed a college education 18% reported being

cigarette smokers. Females were found to have a similar gradient between education and

cigarette smoking, 41% of dropouts reported smoking, while 14% of college graduates reported

smoking. Across all levels of cardiovascular risk factors, men had higher levels of risk than

women did.

Not every measure of health has as clear cut an association with education as mortality and

physical functioning. Due to the nature of education itself, associations with cognitive

functioning in late-life and level of education may be spurious but is also possible that

associations between education and cognitive functioning could stem from lower "risk of chronic

and infectious diseases throughout the life course, quality of health care, occupational or

environmental exposures, or differences in health practices and lifestyle behaviors" (Cagney and

Lauderdale 2002). While using education as a measure of socioeconomic status when modeling

cognitive outcomes poses difficulties, using education as a measure of socioeconomic status to

model blood glucose outcomes is more appropriate.

Measuring Education

Aspects of a person's education include the quantity of education, credentials, and

selectivity of education received (Ross and Mirowsky 1999). The quantity of education is the

number of years of education achieved, implying that each year has equal importance. In

contrast, a focus on credentials argues that the value of an education comes from holding certain

degrees. Selectivity of education refers to the prestige of the institution attended. The selectivity

model could include aspects of both quantity and credentials: the institution offers a higher

quality education thus each year of education has a greater value, and the institute's degree is









more highly valued by society, granting greater access to resources such as occupations and

income. The selectivity model is more easily applied to college degrees. Measurement of

"prestige" below the college level is more difficult. Some private schools are certainly

recognized as prestigious, but there is no national ranking of the secondary schools attended by

the vast majority of the population.

Increasing the quantity of education should increase the stock of "human capital" one has,

according to human capital theory. School is seen as a place where students learn both specific

skills such as mathematics, but also general problem solving skills and the ability to negotiate

with others in the pursuit of a goal. Personality traits such as "self-directedness" are also

encouraged. Specific knowledge, general skills, and personal growth are thus substantive parts of

one's education that have a real impact on an individual, and the more exposure to that kind of

environment, the greater "human capital" one has. In turn, human capital "ultimately shapes

health and well-being" especially through the development of an internal locus of control (Ross

and Mirowsky 1999). Healthy behaviors are more likely to be practiced by somebody if they

believe a good part of their fate rests in their own hands.

The credential view of education argues that the substance of an education is minimal and

that the true value comes from possessing a degree. This approach does not predict a linear

association between years of schooling and various health outcomes. Rather, the possession of a

degree is the best indicator for good health. Furthermore, the effects of the degree would be

mediated by occupational status, assuming the positive association between good quality of jobs

and higher levels of health (Ross and Mirowsky 1999).

The final model of education's value is the selectivity model, which combines aspects of

both quantity and credential views of education. This model predicts that the quality of an









educational institution has health effects beyond the predicted effects of either quantity of

education or credentials. These increasing health returns could be due to either self-selecting

processes (the best students go to the best schools), or that the higher quality of the institution

(either substantively or credentially) leads to better job placement, resulting in higher levels of

health.

Research suggests that quantity of education has the largest impact on physical functioning

and perceptions of health (Ross and Mirowsky 1999). Ross and Mirowsky also find that the

credential model has no significant effect. Controlling for a healthy life-style, the selectivity of

college attended becomes insignificant, suggesting that the beneficial economic pathways

theorized for going to a high quality school are not the actual pathways through which education

operates. Instead, selective schools somehow promote more healthful lifestyles or select for

healthier individuals than less selective schools, although this intra-school difference is relatively

unimportant compared to the effects of continued education at any institution of higher

education.

Pathways Between Education and Health

Several pathways have been offered to explain the positive association between education

and health (Ross and Wu 1995, Adams 2002). These pathways include work and economic

conditions, social and psychological resources, a healthy lifestyle, and health care utilization

skills ("productive efficiency") (Ross and Wu 1995, Adams 2002).

Work and economic effects assume those who have a higher education are more likely to

have a job, with higher income, and greater opportunities for self-fulfillment. In 1991, 87% of

college graduate students were employed compared to 77% of those who had a high school

degree only, and 56% of people with eight years of education or less. Furthermore, those with

greater educational attainment stay unemployed for lesser periods of time (Ross and Wu 1995,









Moen 1999). Those who do not have the skills to enter the workforce face economic hardship

from lack of income, which in turn might affect health through the daily stresses of living life on

the edge and limited access to health care. Daily stresses may play a role in the in the health

advantage more highly educated individuals enjoy, including lower levels of blood sugar (Surwit

1992, Goetsch 1990). Goetsch (1990) found stress to have a hyperglycemic effect in both

laboratory and natural settings, with the greatest level of blood glucose range occurring on high

stress days.

Social and psychological resources are a second way in which education might improve

health. Education develops both a sense of personal control and social support network. The

common sense argument says that education's emphasis on problem solving and interpretation

increases one's perception of personal control. Thus when confronted with health problems, the

highly educated will confront them with both increased knowledge and increased attention to

changing unhealthy behaviors. Ironically, those who most need the tools of knowledge and the

sense of efficacy to confront health issues are those least likely to have them. Social support

systems are stronger for the college educated (Ross and Mirowsky 1999), and those with better

social support systems have better health outcomes, such as lower rates of mortality. Men with

few social connections experience a morality rate 2.3 times higher than those with better social

support (House, Landis, and Umberson 1988).

A third link between education and health is the tendency for the more highly educated to

practice healthier behaviors: they practice greater allocative efficiency. The greater one's

education, the less likely one is to smoke and the greater one's likelihood to drink moderately

(Ross and Wu 1995, Grossman 1997). Higher educational attainment is also associated with

regular exercise, lower body weight, and knowledge about blood pressure. Blacks and Hispanics









had different risk profiles than whites even after adjusting for educational attainment (Shea et al.

1991).

Those with greater levels of education may also have greater levels of productive

efficiency. Productive efficiency occurs when a more highly educated patient is better able to

recognize and express their symptoms to the doctor. In this way, the more highly educated are

able to "get more" out of their inputs into health care processes (Grossman 1997). Support has

been found for many of the theorized mechanisms (Ross and Wu 1995, Adams 2002). However,

endogenous processes may select healthier individuals for higher levels of educational

attainment.

Education and Endogeneity

Ross and Wu (1995), while developing excellent models of possible pathways through

which education can affect health outcomes, do little to address issues of reverse causation. Both

higher levels of education and better health may result from pre-existing factors. Issues of

endogeneity also arise for allocative and productive efficient explanations for education's

association with health. Individuals who are already more efficient accumulators of human

capital may already be healthier, or may have personality characteristic that predispose them to

defer gratification (Grossman 1997, Palloni 2006). It could also be the case that parental

background affects both health and educational attainment (Elo and Preston 1996). Despite these

difficulties in measuring education's effect on health, evidence of a causal relationship between

education and health has been found, with especially pronounced effects among women (Adams

2002).

A similar problem occurs when studying the effects of health insurance on health. Those

with health insurance may have been able to acquire health insurance because of better levels of

health, making health endogenous to the acquisition of health insurance. Acquisition of health









insurance may also indicate individuals who place a higher initial value on health than those who

don't acquire insurance. Thus good health and health insurance may both be a result of a higher

valuation of health. One possible method of dealing with endogeneity in health insurance when

examining its effect on health is to examine the effect of the acquisition of Medicare on health.

(McWilliams et al. 2007).

Income is another measure of socioeconomic status that is susceptible to issues of

endogeneity. Those with lower incomes may have worse health, but those with worse health

likely have lower incomes. Sudden health transitions have a significant effect on reducing

income levels, an important causal pathway through which health is endogenous to income

(Smith 2005). In contrast to income, education is a fairly consistent status within the usual life-

course ordering of events. Educational attainment is generally static after an individual has left

school, making it better than other indicators such as income and occupation in efforts to avoid

endogeneity (McWilliams 2002, Smith 2005).

Framework Linking Education to EBS

Many of the pathways through which education affects general health outcomes are likely

to affect blood sugar levels. Those with a higher level of education experience lower rates of

obesity, higher incomes, and greater levels of healthy behavior (Monteiro et al. 2001, Ross and

Wu 1995, Shea 1991). Given obesity's status as a risk factor for diabetes, it is expected that

increased education will reduce the probability of diabetes, with BMI as an important mediator.

Higher levels of education are also associated with higher incomes, which in turn are associated

with lower rates of diabetes (Rabi et al. 2006). Healthy behaviors, including exercise are also

associated with lower rates of diabetes, thus as educational attainment increases, it would be

expected that rates of diabetes might decrease, with healthy behaviors as moderators ( Bums

2007). Given these clear associations between higher education and factors that reduce the risk









for diabetes, it is expected that higher levels of education will be negatively associated with risk

of EBS. However, this association may vary across different groups of the population for reasons

outlined below.

Non-linear Effects of Education

The effect of education may vary by level of education itself-- i.e. its effects are non-linear

(Cutler and Lleras-Muney 2007). That is, there may be a "heterogeneous effect for each year of

education" (Cutler and Lleras-Muney 2007:6) A linear association between education and health

has been found for some measures of health, including mortality, colorectal screenings, and use

of smoke detectors, while for other measures there is a non-linear association (Cutler and

Llerras-Muney 2007).

Smoking and obesity both tend to have non-linear associations with education. An

increased effect per each year of additional schooling is only seen in those who are more highly

educated. After 10 years of schooling however, health levels become linearly associated with

education levels (Cutler and Lleras-Muney 2007

Theories of Differential Effects

The association between education and is stronger among whites and Hispanics than for

blacks. This finding can be interpreted as evidence that education has a different "translation" for

health across different groups of people (Borrel 2006). There are several reasons why the

association between education and EBS might vary.

Genetic and Early Life Effects

Genetic pre-dispositions may account for race differences in prevalence of diabetes. If one

group is genetically "destined" to develop diabetes, environmental interventions would have a

lesser effect. For example, a population selected for a thrifty gene, once no longer needing the

thrifty gene for survival, may be at increased risk of developing risk factors for diabetes, such as









higher BMI. The thrifty gene hypothesis for Type 2 Diabetes states that certain populations

developed especially efficient metabolic mechanisms that are overwhelmed by modem progress

in food supplies and are therefore predisposed to obesity and Type 2 Diabetes (Neele 1968, Joffe

and Zimmet 1998). The concept has found support in the animal world as well as in human

populations particularly susceptible to diabetes.

More recent evidence also supports a "thrifty gene" hypothesis for perinatal malnutrition

leading to cell dysfunction, also known as the Barker hypothesis (Joffe and Zimmet 1998).The

Barker hypothesis conceptualizes poor fetal and early post-natal nutrition as "imposing

mechanisms of nutritional thrift upon the growing individual"(Hales and Barker 1992). In a fetal

environment with constrained growth, the fetus adopts survival mechanisms which elevate the

risk of diabetes in later life. Possible evidence of the Barker hypothesis includes the example of

the Naruruan islanders, who experienced increased rates of diabetes when malnutrition was

followed by affluence with the introduction of phosphate mining (Hales and Barker 1992). If the

Barker hypothesis is true, those populations which tend to experience less healthy fetal

environments may develop this syndrome in larger number, which could moderate the

relationship between diabetes and protective factors, such as education.

Other early life events may also play a large role in shaping both future socioeconomic

status attainment and health outcomes. While some researchers argue that education structures

the pathways that influence health, it is also likely that early health events help to structure

socioeconomic status attainment (Ross and Wu 1995, Palloni 2006).

Selection and Personality Differences

Personality differences might also account for a proportion of health disparities.

Personality, defined as the "distinctive and characteristic patterns of thought, emotion and

behavior that define an individual's personal style and influence his or her interaction with the









environment," was found to attenuate the relationship between education and all-cause mortality

in men by 34% in a study of 1989 French cohort (Navi, Kimiaki, Marmot et al. 2008) Not

surprisingly, those possessing the battery of personal characteristics deemed "anti-social"

personalities were more likely to die than those with "rational" and "healthy" personalities. The

same personality differences that account for a significant amount of health disparities might also

explain why education might have differing value for people. Those personal traits that have

been found to promote good health, such as placing high value on self-regulation of behavior and

autonomy may help those who possess them also "get more" out of their education by studying

harder, or staying calm under pressure-filled situations such as test-taking.

If personality traits tend to cluster around racial-ethnic groups it is possible that they might

account for some of the variation in the strengths of association between education and health.

Research on personality trait clustering, done in response to the use of personality tests in hiring

to determine if certain demographic groups were more likely to be denied employment based on

test results, indicates small but significant differences of mean scores on personality scales (Ones

and Anderson 2002). if certain demographic groups were more likely to be denied employment

based on test results Evidence has been found of small, but significant differences of mean scores

on personality scales. Differences in scores on "decisiveness," "emotional control," "stamina,"

and "warmth" have been found for blacks and whites in data from the UK (Ones and Anderson

2002)

Cultural Capital Differences

Those who favor the cultural capital explanation for race and ethnic differences point to the

different treatment of students and parents that may lead to these differences (Lareau and Horvat

1999, Morries 2005, 2007). If the value of education stems from the accumulation of skills

taught in the educational setting, different treatment may imply differing accumulation of skills,









altering the value of education. Evidence that the quantity of education is what matters, as

opposed to selectivity of education or credentials bestowed by education, support the idea that

education is a process of skill accumulation

Cultural capital explanations for varying effects of education argue that actors occupying

the multitude of social locations have, and are perceived to have, certain kinds of cultural capital,

and are treated accordingly. Those who are perceived to have the right cultural background will

be treated more favorably, and will be more likely to accumulate skills taught in school.

Advocates of the cultural capital framework point out that one measure of success in higher

education is grades, and grade distributions fall along patterns of social location, with Blacks

often underperforming compared to their Hispanic, Asian, and White peers (Massey 2003).

The currency of cultural capital that may vary from student to student includes a "wide

variety of resources including such things as verbal facility, general cultural awareness, aesthetic

preferences, information about the school system, and educational credentials" (Swartz 1997).

Other scholars have found evidence for different educational experiences based on race and

ethnicity. Black girls are treated differently than their peers based on perceptions of black

femininity within the context of hegemonic white femininity (Morries 2007).

Stress

Others argue that stressors within the social environment are an important source of health

disparities between classes of people (Surwit 1992, Williams 1997). Stress in the social

environment can occur when the psychological demands of the environment are perceived to be

greater than the perception of control over the environment. If certain groups of people

experience differing levels of stress in the educational environment or in their current lives, the

health value of education might be lessened.










Table 1-1. Undiagnosed Diabetes By Educational Attainment and Race, NHANES 1999-2006 Weighted Data
Percentage
Overall: 2.6%
Reference** 2.8%
Education
Less Than HS 4.6%
HS Degree 3.0%
More Than HS 1.8%

Race
Black 2.6%
White 2.6%
Hispanic 2.2%
* Undiagnosed diabetes defined as individuals who reported
having not been told by a doctor they have diabetes but who
had blood glucose levels >= 7.0 mmol/L
** Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fasting Glucose
in Adults in the U.S. Population" Diabetes Care. Vol. 29:1263-1268.












































Figure 1-1. Conceptual Framework for the Association between Socioeconomic Status and
Elevated Blood Sugar, Adapted from Brown (2004)









CHAPTER 2
BACKGROUND TO DIABETES AND IMPAIRED FASTING GLUCOSE

Diabetes

Type 2 diabetes ('diabetes' hereafter), defined as fasting plasma glucose level >126 mg/dl

(7.0 mmol/1), is associated with numerous health problems, including significantly increased risk

of mortality, often due to the cardiovascular complications that can result from having the

disease. Those with diabetes have higher rates of mortality across all age, racial-ethnic, and

gender groups (Bertoni et al. 2002). Using ICD-9 Medicare records, Bertoni et. al.(2002) found

that elders with diabetes suffer higher mortality rates, with a standardized mortality ratio of 1.41

compared to that of non-diabetic elders. Furthermore, national declines in the death rate due to

coronary heart disease have not been seen to the same extent in diabetic population (Gu, Cowei,

Harris 1998).

Diabetes is also associated with significantly increased functional and work disability, as

well as poor mental health. National Health and Nutrition Examination Survey data indicate that

diabetes is associated with increased probability of disability in functional activities, including

slower walking speed, decreased balance, and falling, and that the increased probability of

disability is likely to reduce the quality of life (Gregg et al. 2000). Those with diabetes report

more days of poor physical health (such as physical illness or injury) than matched respondents

(8.3 days compared to 3.0 days), and report more days of poor mental health (such as stress or

depression) than matched respondents (2.8 days compared to 1.8 days) (Valdmanis et al. 2001).

The prevalence of diabetes is increasing. From 1990 to 2000, the prevalence of self-

reported diagnosed diabetes in the U.S. increased 49%, from a 4.9 percent to 7.3 percent of the

U.S. population (Mokdad et al. 2001). Expanding the time frame to 1980-2000, the number of

people with physician-diagnosed diabetes in the United States increased more than two-fold,









from 5.8 million to 13.3 million (Kouznetsova 2007). Accompanying this increasing prevalence

is a dramatically increased incidence rate. In 33 U.S. states that had data for both periods, the

incidence rate increased 90% --from 4.8 per thousand in 1995--1997 to 9.1 in 2005--2007.

Increasing prevalence of a disease associated with increased mortality, increased functional

disability, and many other detrimental health effects has significant policy implications because

of its human and financial costs. Cost of illness estimates typically report either total cost to

society or excess cost of illness (Ettaro et al. 2004). Cost of illness estimates for diabetes of both

types are substantial, and total cost to society is growing with an estimated total cost to society

of diabetes of 100 billion dollars in 1995 (Ettaro 2004). The excess medical cost for an

individual with diabetes is estimated to be 2-5 times greater than for the non-diabetic. The

differential varies across groups, with those in the younger age groups typically experiencing the

largest differential in excess cost. Contributing to the excess cost are the higher hospitalization

rates for diabetics, which are approximately 2.5 times higher than for non-diabetics (Brown et.

al. 1999). The higher rate contributes to the estimated $3.8 billion in costs for in-patient diabetic

care from complications in 2001. Diabetes-related hospital visits cost the Medicare program

$1.3 billion, with estimated preventable costs of $366 million if proper primary care had been

provided prior to complications, such as diabetes, arising from diabetes (Economic and Health

Costs of Diabetes: HCUP Highlight 1 2005).

To compound the difficulties of higher health care costs, diabetics are more likely to

experience unemployment and are more likely to have incomes less than $20,000 than non-

diabetics. 71% of diabetics have annual incomes less than $20,000, compared to 59% of non-

diabetics. Minority status individuals are more likely have multiple hospitalizations than diabetic

non-minority individuals (Valdmanis 2001).









Impaired Fasting Glucose (Pre-Diabetes)

Pre-diabetes, also known as impaired fasting glucose, is a pre-diabetic state recently

defined in 2004 by the American Diabetes Association as individuals with impaired fasting

glucose levels between 100 mg per dl and 125mg per dl. The previous definition had included

only those individuals with IFG levels greater than 110 mg per dl. It is estimated that in 2002,

about 26% of the adult population has impaired fasting glucose, one of the leading risk factors

for developing diabetes mellitus (Cowie 2002).

Unlike diabetes, the prevalence of impaired fasting glucose has not dramatically increased.

A study comparing prevalence of IFG between 1988 to 1994 and 1999-2002 using data from

NHANES, the same data source used in this thesis, found a small increase in the prevalence of

IFG, from 26.3% to 26.9% (loannou et al. 2007). Results from a second study, also using

NHANES from the same year, found a slightly larger increase from 24.7% to 26.0% (Cowie

2002). While the results of these two studies are similar, they diverge more than would be

expected given the same data. It is not known why the estimates vary by 1-2%. Recall that rates

of diagnosed diabetes have risen significantly, from 5.1% to 6.5% during the same time frame as

has been reported for IFG, even as undiagnosed diabetes remained stable.

Impaired fasting glucose itself is also associated with increased risk of cardiovascular

disease, and those with pre-diabetes may also be at increased risk of mortality (Califf et al. 2008,

Peterson and McQuire 2005, Snehalatha et al. 2003). Therefore efforts to reduce the incidence of

CVD can target both diabetes and pre-diabetes.

Link Between Impaired Fasting Glucose and Diabetes

A major reason to set reduction of IFG prevalence as an important goal is impaired fasting

glucose's strong association with development of diabetes. Reported rates of progression from

IFG to diabetes have varied in the literature, with some research suggesting the percent of









individuals progressing from IFG to diabetes to be as high as 33% over a 6 year period, to as low

as 9.1% over a 12 year period, with non-white study populations generally reported to have

higher percentages of progression from IFG to diabetes. (Nichols et al. 2007) The recent

reduction in the IFG threshold, from 110 mg/dl to 100 mg/dl, also has an effect on the expected

rate of change. One U.S. based study reported that, over a 9 year period, 24.3% of individuals

who satisfied the old definition of IFG progressed to diabetes, whereas under the new definition

of impaired fasting glucose only 8.1% of individuals progressed from IFG to diabetes (Nichols et

al. 2007). Thus progression from IFG to diabetes is expected to occur at a slower rate under the

new definition. Other determinants of progression from IFG to diabetes include higher levels of

blood glucose, known hypertension, and high levels of triglycerides (Rasmussen et al. 2008).

The empirical link between IFG and diabetes, and the measurement of blood glucose along

a continuous scale for both states of glycemia is the major reason for analyzing the data using a

continuation ratio logistic regression model (CRM). This model, as explained in greater detail in

Chapter 4, allows the researcher to test for significant differences in effects of covariates

between stages of models. Some covariates may be more important for increasing risk of IFG

than diabetes, or some covariates may not be significant until one already has IFG.

While fasting plasma glucose levels follow a continuous spectrum that allow for analysis

via the CRM, it is important to recognize that the two conditions are separate. Diabetes is

characterized by the failure of beta cells to produce the required amount of insulin in the context

of insulin resistance, whereas IFG is characterized by the ability of the beta cells to

accommodate the required increase in insulin production in the context of insulin resistance

(Shabha 2004).









A Model and Critical Covariates for Risk of Diabetes

This study focuses on both IFG and diabetes and examines whether some of the primary

risk factors associated with IFG and diabetes are different. The CRM model, for accurate

comparisons, requires the inclusion of all covariates at each stage of the outcome level. The risk

factors for diabetes will determine what covariates to include in the final CRM model,

acknowledging that the risk factors for IFG may be different. Figure 2-1 outlines a model of risk

factors for the development of diabetes. Primary risk factors for the development of diabetes

include a family history of the disease and demographics, behavioral psychological, and clinical

factors (Brown 2002).

Age and Sex

Figure 2-2, constructed using data from the National Health Interview Survey and provided

by the Centers for Disease Control, illustrates the variation in diabetes prevalence between age-

groups. Diabetes prevalence increases with age for all races and sexes up until 75+ years, at

which point prevalence declines. Variation between age-groups is much greater than it is for sex

groups, across all racial categories, and is thus important to control for it. To help adjust for age's

possible non-linear association with risk of EBS, age is coded into 3 dummy variables: 24-35

years, 36-45 years, and 46+ years. Controlling for age also helps to account for the race and

education differences that may occur between age groups. Overall prevalence of diabetes in most

populations is equivalent across the sexes.

Race and Ethnicity

Blacks and Hispanics are both more likely to have diabetes than whites (Brancati et. al.

1996). The increasing proportion of the Hispanic population in the U.S. population and their

elevated risk of mortality compared to whites makes it is especially important to address these

health disparities in national efforts at reducing incidence and prevalence of diabetes. There are









significant differences in prevalence of both pre-diabetes and diabetes between racial-ethnic

groups over the age of 20, the ages analyzed here. Data from the National Institute of Health

demonstrate the disparity of diabetes between racial-ethnic groups: Among non-Hispanic whites,

8.7% aged 20 and over have diabetes. Among Non-Hispanic blacks however, 13.3% aged 20 and

over have diabetes. Among the Hispanic population, 9.5% aged 20 and over have diabetes.

While Asian Americans aren't included in the final analysis of this thesis due to limited sample

size, they are 1.5 times as likely to have diagnosed diabetes as non-Hispanic whites (National

Institute of Diabetes and Digestive and Kidney Diseases 2008).

Obesity

Obesity is a primary risk factor in the development of diabetes (Shai, Im and Jiang, R. et

al. 2006 ) Using data from the 2000 Behavioral Risk Factor Surveillance System, Mokdad et al.

derived estimates of diabetes prevalence between the obese and non-obese across numerous

characteristics. In each case, the obese group had a higher prevalence of diabetes than did the

non-obese group. Prevalence of obesity has been increasing, and some have called both obesity

and diabetes an "epidemic" among the U.S. population. (Mokdad et al. 2001). Table 2-1 reports

the percentage of obese persons by sex and by age groups.

Obesity is also strongly related to educational attainment, with the least educated having

the greatest likelihood of being obese (Borrell 2006). Since obesity is related to both education

and diabetes, it may operate as a mediating variable in the relationship between them. Despite

obesity's strong association with diabetes, controlling for obesity has been found to explain

some, but not all of the racial and ethnic differences in diabetes (Brancati 1996).

With regard to impaired fasting glucose, B.M.I. has been found to predictive for both

blacks and whites. (Klein et. al. 2004) Using data from the National Heart, Lung, and Blood

Institute, Klein (2004). found that baseline B.M.I was predictive of impaired fasting glucose for









black girls and rate of B.M.I. increase was predictive of impaired fasting glucose for white girls.

Whether the effect of B.M.I. remain constant through the progression of EBS remains unknown,

and therefore will be examined in this analysis.

Marital Status

Marital status is also associated with diabetes. Choi and Shi (2001) found that womenmn

who were single and 35 to 64 years old had a higher prevalence of diabetes than women of the

same age who were married". Marital status is also associated with better health outcomes

across other indicators of health (Shoenborn 2004). Using NHIS data pooled from the years

1999-2002, Shoenborn finds that regardlesses of population subgroup (age, sex, race, Hispanic

origin, education, income, or nativity) or health indicator (fair or poor health, limitations in

activities, low back pain, headaches, serious psychological distress, smoking, or leisure-time

physical inactivity), married adults were generally found to be healthier than adults in other

marital status categories". A notable exception in their study was obesity-- married men were

more likely to be obese than those in other martial groups. One possible explanation for all of

these findings is that marriage selects for better health, i.e. healthier individuals are more likely

to get married.

The relationship between education and marital status differs between men and women.

For men, each increasing year of education is associated with an increasing likelihood of

marriage, and this association has grown stronger from 1980-2000 (Rose 2003). For women,

there exists a "marriage penalty" after 12-16 years of education. Initially, women and men can

expect a positive association between education and likelihood of marriage. However, for

women, after 12-16 years of education the association becomes negative. This "marriage

penalty" for increasing levels of education lessened from 1980-2000. Thus the strength of

education's effect on likelihood of marriage seems to be growing for men and women alike.









Since the more highly educated (up to 12-16 years for women) are more likely to be married, and

marital status is associated with diabetes, it is important to control as another possible mediating

variable in the association between education and diabetes.


































Figure 2-1. Conceptual Framework for Risk Factors for Diabetes, Adapted from Brown (2002).









U.S. Diabetes Rate by Age and Sex


Figure 2-2. U.S. Diabetes Percentage by Age and Sex. Adapted from Centers for Disease
Control 2006









Table 2-1. Obesity and Diabetes Prevalence Among U.S. Adults, by Sex and Age BRFFSS, 2000*
Obesity Diabetes
%(SE) %(SE)
Total 19.8(.17) 7.3(.12)
Sex
Male 20.2(.26) 6.5(.18)
Female 19.4(.21) 8.2(.15)

Age
18-29 13.5(.33) 1.9(.13)
30-39 20.2(.36) 3.8(.18)
40-49 22.9(.41) 5.8(.27)
50-59 25.6(.47) 10.9(.37)
60-69 22.9(.50) 14.5(.44)
>=70 15.5(.41) 14.9(.42)
*Adapted from Mokdad 2008.









CHAPTER 3
DATA AND METHODOLOGY

Objectives and Research Questions

The objective of this thesis is to answer the following three questions: (1) What is the

association between education and the development of EBS; (2) does this association vary by

race and ethnicity and sex; and (3) does the association between race and ethnicity and elevated

blood sugar vary depending on the level of blood glucose?

Hypotheses

Seven hypotheses follow from the research questions and past research. First, it is

hypothesized that there is a negative correlation between level of education and probability of

having EBS: as level of education increases, the probability of having EBS decreases (HI). This

hypothesis is based on the research that has consistently found an association between higher

levels of education and better health. Second, it is hypothesized that the association between

education and EBS varies by race and ethnicity. The effect of education on probability of having

EBS will be weaker among Blacks (H2). This hypothesis is based on the research that suggests

there may be a "thrifty gene" that predisposes blacks to EBS and that blacks may not possess the

same cultural capital as whites, limiting the effect of education. It is hypothesized that education

has similar effects between Hispanics and whites (H3). This hypothesis is based on the previous

research that found similar effects of education for whites and Hispanics but not for blacks. It is

hypothesized that the effect of education will not differ by sex (H4). This hypothesis is based on

research that suggests the effect of socioeconomic status varies by sex for occupational status,

but not for level of education. Finally, it is hypothesized that the effect of race and ethnicity and

education varies by level of blood glucose. This hypothesis is based on the fact that diabetes and

IFG are two different, although closely linked, disease states. Past research and descriptive









statistics of the sample analyzed suggest that being black will be associated with lower rates of

IFG but higher rates of diabetes (H5). It is expected that being Hispanic will not have a different

association with either IFG or diabetes (H6). It is expected that education will have a stronger

association with EBS than with diabetes (H7).

Data

Background

The NHANES (National Health and Nutrition Examination Survey has been the primary

tool of the National Center for Health Statistics of collecting information on the health of the

U.S. population since 1960. NHANES data have been released in two year cycles since 1999 to

allow the survey to adjust more quickly to the needs of researchers studying a diversity of health

issues within the US population. The NHANES is especially useful to researchers because the

survey collects laboratory measures as well as self-reported information on health statuses and

other relevant dimensions. Pooled data from the 1999-2006 NHANES are analyzed in this study.

Sample and Selection Procedures

The target population of the survey is the civilian, non-institutionalized US population,

with low-income and minority oversamples The survey population is selected through a stratified

multistage probability sample. There are four primary stages of sample selection: "1) selection of

Primary Sampling Units (PSUs), which are counties or small groups of contiguous counties; 2)

segments within PSUs (a block or group of blocks containing a cluster of households); 3)

households within segments; and 4) one or more participants within each household. A total of

15 PSU's are visited during a 12-month time period" (National Center for Health Statistics

2009). Once a household has been selected, it is notified via mail and a NHANES representative

makes direct contact with the household to determine if it contains eligible participants. All

eligible interviewees are also asked to participate in the medical examination component of









NHANES. Each respondent is randomly assigned to either a morning or afternoon examination

session. The morning participants were tested for their fasting levels of various blood chemicals

such as glucose. Because the survey sample population varied at each level of the NHANES

(questionnaire, medical examination component, and fasting subsample), it is important to use

the proper weighting variables when doing analyses. NHANES is based on a clustered sampling

design and respondents have different probabilities of selection. Failing to adjust for the

sampling design through the use of weights may bias estimates and overstate significance levels

(National Center for Health Statistics 2009).

Blood sugar levels in the NHANES were assessed using fasting plasma glucose, two hour

glucose tolerance test, and serum insulin in participants aged 12 years and older. The present

analysis uses the fasting plasma glucose measure. This measurement comes from the smallest

sub-sample group in the NHANES, and following NHANES documentation, the weighting for

the smallest subsample is used. Thus the fasting subsample weight variable is appropriate for the

analysis. Since data were pooled across four survey cycles, the weight variable had to be

adjusted to account for the differing survey years. The NHANES provides a 4 year sample

weight to be used with analysis for both 1999-2000 and 2001-2002, and provides a 2 year weight

specific to either 2003-2004 or 2005-2006. Combined survey cycles should be representative of

the population at the midpoint of the combined survey period and the sum of rescaled weights

should match the survey population at the midpoint of each period. The 4 and 2 year weights

used in this analysis were made directly comparable by assigning half the 4 year sample weight

for respondents in 1999-2002, and 14 of the sample weight for respondents 2004-2006 (National

Center for Health Statistics 2009).









The target sample for analysis consists of US born persons aged 24 and over, although

descriptive tables 4-1 through 4-5 report the prevalence of IFG and diabetes in the U.S.

population over the age of 20 in order to make a more precise comparison with previously

published statistics. Limiting the sample to those born in the US in the regression analysis helps

to prevent measuring education received outside the US, and controls for the better health of the

immigrant population, also known as the "healthy immigrant" effect (Kennedy et al. 2006).

Variables

Elevated Blood Glucose Level

For the regression analysis, the dependent variable elevated blood sugar is defined as

having fasting blood glucose levels greater than or equal to 100 mg/DL, reporting by the

respondent that he or she has been told by a doctor the individual has diabetes, or currently

taking medication to reduce blood sugar levels. This cut-off point was chosen because it is the

new standard definition of having impaired fasting glucose. Elevated blood glucose is coded

dichotomously, with 1= having EBS and 0 as not having EBS.

Diagnosed Diabetes

While the regression analysis uses the broadest definition of diabetes, several tables report

descriptive statistics using more narrow definitions, such as diagnosed only or undiagnosed

diabetes. Tables that report diagnosed diabetes define diabetes as a self-report of a doctor's

diagnosis of diabetes, without making use of the blood glucose analysis. Tables that report

undiagnosed diabetes report individuals as undiagnosed if they report blood sugar levels greater

than 125 mg/dL but do not report having been told by a doctor they have diabetes.









Independent Variables


Education

The education variable used in this analysis was measured as an ordinal variable by

handing a respondent a card, reading the categories to them as necessary, and recording the

respondent's choice. The respondent could choose from first through 12th grade, high school

graduate, GED or equivalent, some college, AA degree, technical degree, BA, MA, or PhD.

Codes were also used for Refusal or Don't Know. For analysis, education will be recorded into

five categories: less than 9th grade, less than high school, high school degree, some college, and

BA or greater. Although data for blood glucose and education is available for all respondents

aged 12 and over, analysis is limited to those age 24 and over because this is the standard

minimum age at which someone is likely to earn a degree beyond that of the BA.

Race

The race variable used in this analysis was constructed from two variables that are

respondent self-reports on questions of race and ethnicity. From these questions, the main race

variable was constructed containing the following racial-ethnic categories: Mexican American,

Other Hispanic, Non-Hispanic White, Non-Hispanic Black, and Other. These categories are

collapsed for the final analysis into Hispanic, Non-Hispanic White, and Non-Hispanic Blacks.

Age

Age was ascertained by asking the respondent "How old are you", with a resulting

response range (for the final sample used in the analysis) between 24 and 85 years of age.

Anything greater than 85 was put into the 85+ category. (N=599). Age is coded into three

categories: age 24-35, age 36-45, and age 46 or older.









Marital Status

Marital status was computed from both the question asking the respondent their marital

status, and imputing data from responses that made reference to marital status. Marital status is

coded as a dummy variable in the analysis, with 1 being married, 0 if not married.

BMI

BMI was calculated using measurements from the medical examination portion of the

NHANES survey. This variable was coded as overweight for any individual over 24.9 kg divided

by meters of height squared and normal for any individual less than 25 kg divided by meters of

height squared.

Methods: Continuation Ratio Model

The continuation ratio model is an appropriate model when the ordered categories follow a

progression of stages. Stages of elevated blood glucose are an appropriate outcome to model

using a continuation ratio as it is possible to rank the stages from "low" to "high", and generally

one must proceed through one stage of blood glucose before entering into another. This study

uses the continuation ratio model because the objectives require an analysis of the effects of

selected covariates across the different stages of blood glucose.

Pooled data from the 1999-2006 NHANES are analyzed using binomial logistic regression

to determine racial and ethnic differences in the magnitude of education's effect on blood sugar

levels. For this analysis, there exist three different stages of blood sugar to be compared: normal,

elevated, and diabetic stages of blood sugar. The first comparison, or "stage" will be made

between those with normal blood sugar and those at the first "cut-off' point of blood glucose

levels of 100 mg/dL or more. The second comparison, or "stage" will be between those who

have diabetes versus those who do not, among the population of individuals with EBS. Finally,

interaction terms of the form stage*'x', where x is an independent variable, can be used to test for









differences of effects of the variables included in the model across the different stages of the

outcome variable. A significant interaction term indicates that the effect of the particular variable

differs depending on the stage of the outcome variable.

Models for males and females are estimated separately because an overall Wald Test

indicated that at least one coefficient was significantly different for males and females (F=2.62,

p<.05). Thus our fourth hypothesis, that of no interaction between sex and education in the

association with EBS is not supported. Significant sex differences will be reported in Tables 4-7

and 4-8. The first model will use EBS (having either impaired fasting glucose or diabetes) as its

outcome measure. The second regression model will compare those with diabetes, defined as

having a blood glucose level greater than or equal to 126 mg/dL, taking medication to control

blood sugar, or having been told by a doctor the individual has diabetes, with those who have

EBS but do not have diabetes.









CHAPTER 4
RESULTS AND ANALYSIS

Data Description

Table 4-1 summarizes weighted and un-weighted estimates of impaired fasting glucose

percentages for the U.S. population over the age of 20 across sex and race and ethnicity. While

later analyses use National Health and Nutrition Examination data for ages 24 and older, this

table uses data for ages 20 and over to facilitate comparison with previous studies. This table

includes a column for a "reference" parameter, derived from another article that used data from

the 1999-2002 NHANES for estimates of impaired fasting glucose and diabetes in the U.S.

population (Cowie 2002). Cowie's estimates generally differ from the present analysis by about

1%. These differences likely are the result of different weighting procedures used by Cowie

(2002) to derive estimates from a combined sample.

Data for each two year release cycle of the NHANES is presented, as well as a weighted

average constructed using all eight years of data. The weighted population average of IFG across

all eight years of data is 26.6%. Results from this thesis' analysis replicate the finding that rates

of impaired fasting glucose have not risen dramatically, changing from 25.3% for 1999-2002 to

27.9% in 2003-2006, although a 2.5% increase in the U.S population suggests that well over

half-a million people more in 2006 have impaired fasting glucose than in 2002. Overall, men are

consistently found to have higher percentages of IFG than women. The eight year average of

percentage of men with IFG is 34.5%, versus 21.8% for women. Men across all three

race/ethnicities also have a higher percentage of IFG than women. The eight year average

percentages of white, black, and Mexican-American men are 34.0%, 22.0%, and 35.3%,

respectively. The eight year average percentages of women with IFG are 21.2%, 18.9%, and

19.0%, respectively. With respect to race, whites and Hispanics consistently have a higher









percentage of their populations with IFG than blacks. Whites have an eight year average of

27.3% and Hispanics have a similar 27.7% average. The eight year average for blacks is 20.3%.

This table suggests that there are racial and ethnic differences as well as sex differences in

the U.S. population in prevalence of impaired fasting glucose. IFG puts the individual at greater

risk for developing diabetes. Do these apparent differences between sex and racial and ethnic

groups persist into the next stage of elevated blood sugar, diabetes? Table 4-2 presents weighted

and unweighted estimates of percentages of individuals with diagnosed diabetes. Data for each

two year release cycle of the NHANES is presented, as well as a weighted average constructed

using all eight years of data. The weighted population average of diagnosed diabetes is 7.l1%.

Overall, men and women have similar levels of diagnosed diabetes. The eight year average of

percentage of men with diagnosed diabetes is 7.l1%, versus 7.2% for women. Blacks and

Hispanics consistently have a higher percentage of diagnosed diabetes than Whites. Whites have

an eight year average of 6.3%, whereas blacks have the highest eight year average, 10.9%, and

Hispanics have an eight year average of 7.3%. Within racial and ethnic categories, only white

men have a higher percentage of diagnosed diabetes than woman, 6.5% vs. 6.l1%. For blacks and

Hispanics, women appear to have a higher percentage of diagnosed diabetes than their male

counterparts: 9.9% vs. 11.7% among blacks and 6.2% vs. 8.4% among Hispanics.

The apparent difference in diagnosed diabetes by race and ethnicity reverses the trend seen

in Table 4-1, where whites consistently had a higher percentage of individuals with IFG.

However as previously discussed, racial and ethnic differences of undiagnosed diabetes appear

minimal. (Please see Table 1-1) Thus it is not surprising that in Table 4-3, which presents the

percentage of persons with diabetes, including diagnosed, undiagnosed, and those taking

medication, the patterns seen in Table 4-2 persist.









Data for each two year release cycle of the NHANES are presented in Table 4-3, as well as

a weighted average constructed using all eight years of data. The weighted population average of

diabetes is 9.9%. (versus 7.1% with diagnosed diabetes). Overall, men and women have similar

levels of diabetes, although men have a slightly higher percentage of individuals with diabetes

than women with the expanded definition of diabetes. Blacks and Hispanics consistently have a

higher percentage of diagnosed diabetes than Whites. Whites have an eight year average of

6.3%, whereas blacks have the highest eight year average, 10.9%, and Hispanics have an eight

year average of 7.3%. Within racial and ethnic categories, only white men have a higher

percentage of diagnosed diabetes than woman, 10.3% vs. 7.7%. For blacks and Hispanics,

women appear to have a higher percentage of diabetes than their male counterparts: 9.9% vs.

11.7% among blacks and 6.2% vs. 8.4% among Hispanics. Thus there seem to be two

differences between levels of elevated blood sugar (IFG and diabetes): both Table 4-2 and Table

4-3 reverse the trend seen in Table 1, where whites consistently had a higher percentage of

individuals with IFG.; and there are only small sex differences for diabetes, unlike the sex

differences found at the IFG level of blood sugar.

Tables 4-1 through 4-3 show some evidence for different effects of race and ethnicity and

sex across different levels of blood sugar. Males, especially white males, seem to have a higher

percentage of individuals with IFG, however this sex effect is not as pronounced at the diabetic

stage. Race effects also appear to differ: blacks have the lowest 8 year average percentage of

individuals with IFG (20.3%) but the highest 8 year average percentage of individuals with

diabetes (13.2%).

Tables 4-4 and 4-5 present descriptive statistics for another independent variable of

interest, education, and the weighted percentages of individuals with both IFG and diabetes









within 3 categories of educational attainment: less than high school, high school degree, and

college or more. Estimated percentages of both IFG and diabetes tend to decrease as level of

education increases. Overall, among those in the lowest level of education, 31% and 15.2% are

estimated to have IFG and diabetes, respectively. These estimates decrease to 24.3% and 7.7%,

respectively, for those individuals in the highest category.

At the level of IFG, there are some apparent race differences: the estimated percentage of

whites decreases with each increasing category of education; moving from 34.8% among those

with less than a high school education to 29.8% among those with a high school degree, to

24.6% among those with some college education or more. Among Blacks, for both men and

women, there appears to be little effect of increasing education on the estimated percentages of

individuals with IFG. For Hispanics, the highest estimated percentage of individuals with IFG is

in the lowest category (30.6%), and the lowest estimated percentage is for individuals with a

high school degree (19.5%).

At the level of diabetes, there are again some apparent race differences: the estimated

percentage of whites decreases with each increasing category of education; moving from 14.6%

among those with less than a high school education, 9.9% among those with a high school

degree, and 7.3% among those with some college education or more. Among blacks there also

appears to be a trend of decreasing percentages of individuals with diabetes, moving from

17.5%, to 14.2%, to 9.9%. For Hispanics, the highest estimated percentage of individuals with

diabetes is in the lowest category of education (14.1%), and the lowest estimated percentage is

for individuals with a high school degree (8.2%). For Hispanic women, the downward trend

continues on into the highest category of education, with 6.7% of Hispanic women estimated to

have diabetes. However for Hispanic men, having some college education or more seems to have









little effect on percentage of individuals with diabetes compared to Hispanic men with a high

school degree.

Tables 4-1 through 4-5 tables have provided evidence for differing effects of sex, race and

ethnicity, and education in their respective associations with levels of blood sugar. The next

section presents formal tests to see if these apparent trends have statistical significance

Table 4-6 is the final descriptive statistics table, and includes the means and proportions of

variables used in the continuation ratio model analysis. For percentages of educational

attainment across five levels of education for the entire NHANES sample, rather than just the US

born segment, see Table 1, Appendix A.

Results for Men

Table 4-7 displays the logistic regression results for males. These regression results

represent the effects of the covariates across two different stages of blood sugar. The regression

coefficients of the covariates reported in Stage 1 describe the effects of their respective variables

on the log odds of having either impaired fasting glucose or diabetes compared to normal levels.

The regression coefficients of the covariates reported in Stage 2 describe the effects of their

respective variables on the log odds of having diabetes, given that the individual has elevated

blood sugar. Coefficients with a significance level of less than .05 are considered significant.

Four different models were tested: (1) a model with just race and age, (2) a model that

additionally adjusts for educational attainment, (3) a model which additionally adjusts for marital

status and BMI, and finally (4) a model which allows interaction between race and ethnicity and

education. The column labeled as "Stage Diff' reports the results of significant stage interactions

for the covariates. This interaction signals that the effect of the covariate is significantly different

from one level of blood sugar to the next. For males, stage differences were tested using equation

three because of the lack of significant education and race and ethnicity interactions in equation









four. For females, stage differences were tested using equation four, which included the

interaction between education and race and ethnicity.

Although the education and race and ethnicity interaction effects are insignificant for males

they are included for the sake of model comparisons with females. The same four models were

estimated for males and females at the two stages of elevated blood sugar: impaired fasting

glucose and diabetes. The column labeled as "Sex Diff' report the results of significant sex

interactions for each of the two stages of blood sugar. Significant sex differences correspond to

equation four, which allowed for education and race and ethnicity interactions.

Males, Stage 1: Impaired Fasting Glucose

Model 1 examines the main effects of race controlling for age. Being black has a

significant effect on reducing the log odds of developing EBS relative to whites (b=-0.28,

p<.05). Being in either of the two youngest age categories is also significantly associated with

reduced probability of developing EBS.

Model 2 additionally adjusts for level of education. Having a BA or more is the only level

of education found to be statistically different in its effect on the log odds relative to having a

high school education. Having a BA or more reduces the log odds of having either IFG or

diabetes by 0.33. Being black continues to have a significant association with reduced log odds

of developing EBS.

Model 3 additionally adjusts for marital status and BMI. Being married has no significant

association with EBS relative to non-married individuals, however having a BMI of 26 or over

is associated with a significant increase in the log odds of developing EBS (b=0.94, p<.01).

Being in either of the younger age groups significantly reduces the probability of having EBS

(b=-1.53, p<.01 for ages 24-35 and b=-0.86, p<.01 for ages 36-45). There are no significant

effects of race, education, or marital status in the full main effects model for males. Thus for









males, our first hypothesis of a direct association between education and probability of having

EBS is not supported.

Model 4, the last model, allows for an interaction effect between race and ethnicity and

education. These interaction terms are insignificant, indicating that our second hypothesis of an

interaction between being black and educational attainment that reduces the effect of education

on probability of having EBS is not supported. Our third hypothesis that education's association

with EBS is similar for Hispanics and whites is supported by the lack of interaction. The

interaction terms, although insignificant, are included in the final model for proper comparisons

of males with the final model for females.

Males, Stage 2: Diabetes

Four models were again tested for stage 2. The regression coefficients of the covariates

reported in stage 2 describe the effects of their respective variables on the log odds of having

diabetes, given that the individual already has EBS.

Model 1 tests just the effects of race and age. Being black is found to have a significant

association with increased log odds of having diabetes (b=.56, p<.01). This is a reversal blacks

association with reduced probability of having elevated blood sugar. However statistical tests of

stage differences are not conducted until equation 3, the full model. Being Hispanic is also found

to have a significant association with an increased probability of developing diabetes, given EBS

(b=.70, p<.01) Being in the youngest age category has a significant association with reduced

probability of developing diabetes given one has EBS (b=-1.13, p<.01).

Model 2 additionally adjusts for level of education. Unlike in Stage 1, no level of

education is found to be statistically different in its association with the log odds of developing

diabetes relative to having a high school education, nor are any of the effects of education found

to be significantly different depending on stage of blood sugar or race and ethnicity. After









adjusting for education, being in the youngest category relative to the oldest category continues

to be associated with reduced probability of having diabetes given one has EBS (b=-1.2-, p<.01).

Model 3, the full main effects model for males, additionally adjusts for marital status and

B.M.I. Being married has no significant association with probability of having diabetes, given

IFG. Having a BMI in the "overweight" or "obese" categories has a significant association with

increased log odds of developing diabetes relative to those with a "normal" B.M.I. (b=0.52,

p<.05). Being black or Hispanic continues to have a significant association with increased

probability of developing diabetes given EBS.

As in stage 1, model 4, the final model, allows for an interaction between race and

ethnicity and education. These interaction terms continue to be insignificant but are included in

the final model for proper comparisons of males with the final model for females.

Stage Differences for Males

Statistical tests for difference of effect of covariates within the statistical model for males

were conducted using equation 3, the full main effects model. Three variables were found to vary

significantly by stage for males: being black, being Hispanic, and BMI. The lack of an

interaction between education and stage of blood glucose does not support our seventh

hypothesis that education's association with EBS would be stronger than with diabetes, given

IFG.

In stage 1, model 3, being black had a significant association with reduced probability of

developing EBS. In stage 2, model 3, being black is found to have a significant association with

increased probability of developing diabetes, given EBS. This finding supports the fifth

hypothesis that the effect of being black will vary by level of blood glucose. The proportional

odds model is limited in that it forces the effect of the independent variable to have the same

effect for every level of the outcome variable. The ability of the continuation ratio model to both









estimate and test differences across stages indicates its superiority to the proportional odds model

in this particular analysis. The continuation ratio model provides more flexibility in the effect of

the independent variable and allows a third statistical inference about the effect of being black:

that these two coefficients are significantly different from each other across the stages of blood

glucose. The results of the stage interaction tests are evidence that the effect of being black has a

significantly different effect depending on the level of blood glucose. Being black reduces the

probability of developing EBS. However, being black has a significantly different effect given

EBS on increasing the probability of diabetes.

A second variable found to have significantly different effects across stages of blood

glucose is being Hispanic. In stage 1, equation 3, being Hispanic has no significant effect on

estimated probability of EBS. In stage 2, equation 3, being Hispanic has a significant effect on

increasing the probability of developing diabetes. Thus our sixth hypothesis of no interaction

between being Hispanic and level of blood glucose is rejected. The lack of statistical difference

in stage 1 highlights the difference between a CRM stage test of difference and a standard

inference about regression coefficients: CRM tests for difference of effect by level of the

dependent variable whereas standard inference tests difference from zero. Thus even though both

coefficients may be insignificantly different from zero, the two coefficients may be different

from each other.

The third variable found to have a significantly different effect by level of blood glucose is

BMI. In stage 1, model 3, being overweight has a significant effect on increasing the probability

of developing EBS (b=0.94, p<.01) In stage 2, model 3, being overweight has a significant

effect on increasing the probability of diabetes, given EBS. BMI thus appears to have a

significantly smaller effect on developing diabetes given EBS then on developing EBS for males.









Results for Females

Table 8 presents the regression results for females across the two different stages of blood

sugar. The regression coefficients of the covariates reported in Stage 1 describe the effects of

their respective variables on the log odds of having either impaired fasting glucose or diabetes

relative to the population that does not have impaired fasting glucose or diabetes. The regression

coefficients of the covariates reported in Stage 2 describe their effects on the log odds of having

diabetes, given that the individual has impaired fasting glucose. Four models were tested: (1) a

model with just race and age, (2) a model that additionally adjusts for educational attainment (3)

a model which additionally adjusts for marital status and B.M.I. and finally (4) a model which

allows for an interaction between race and ethnicity and education.

Females, Stage 1: Impaired Fasting Glucose

In model 1 with just race and age, being either black or Hispanic has a significant

association with reduced log odds of having either IFG or diabetes relative to whites. Like males,

being in either non-referent age group, Age 20-35 and Age 35-45, is found to have a significant

association with reduced log odds of having either IFG or diabetes relative to being over the age

of 45.

Model 2 additionally adjusts for level of education. Each level of education for females in

stage 1 was found to have a significantly different effect on log odds of developing EBS relative

to having a high school degree. Having less than a high school degree has a significant

association with increased log odds, while having more than a high school degree had a

significant effect on decreased log odds.

Model 3 additionally adjusts for marital status and B.M.I. Being married has no significant

association with EBS probability, however being overweight or obese has a significant

association with increased log odds of developing EBS relative to those with a overweight









B.M.I. (b= 1.33, p<.01). All levels of education continue to be associated with probability of

developing diabetes.

Model 4 allows for an interaction between race and ethnicity and education. The two

younger categories continue to have a significant association with reduced log odds of having

elevated blood sugar. Having less than a high school education is associated with a higher

probability of having EBS and having more than a high school education is associated with a

lower probability of having EBS, supporting the first hypothesis. An adjusted Wald test was

performed to see if the slopes for educational attainment vary by race and ethnicity. The omnibus

test provides strong evidence that there is an interaction effect (F=3.14, p<.01). The interaction

between being black and having less than a 9th grade education lends support to the second

hypothesis that being black lessens the association between education and probability of having

EBS. An interaction for Hispanics having a BA or more is significant (b=1.6, p<.01). Thus there

is evidence that being Hispanic is associated with a reduced assoication of having a BA or more

on the log odds of having EBS. This does not support the third hypothesis of no interaction

between education and Hispanics.

Females, Stage 2: Diabetes

The same four models were tested for stage 2. Again, the regression coefficients of the

covariates reported in stage 2 describe the statistical effects of their respective variables on the

log odds of having diabetes, given that the individual already has EBS.

Model 1 tests just the effects of race and age. Unlike in stage 1, being black is found to

have a significant association with increased log odds of having diabetes relative to whites, given

one has EBS (b=.80, p<.01).. Being Hispanic does not have a significantly different effect on the

progression from IFG to diabetes. Unlike that of having either IFG or diabetes, age is not

associated with progression from IFG to diabetes.









Model 2 additionally adjusts for level of education. No level of education is found to be

statistically different in its effect on the log odds relative to having a high school, nor are any of

the effects of education found to be significantly different depending on stage of blood sugar or

race and ethnicity.

Model 3 additionally adjusts for marital status and B.M.I. For females, being married is

associated with a decreased probability of developing diabetes given EBS (b=-0.52, p<.01)..

Being overweight or obese is again associated with increased log odds of developing diabetes

relative to those with a "normal" B.M.I. (b=0.74, p<.01). Being in the youngest categories

continues to be associated with reduced log odds of having elevated blood sugar relative to those

in the oldest age category, given one has EBS

Model 4 allows for an interaction effect between race and ethnicity and education however

no interactions are found to be significant.

Stage Differences for Females

Four covariates were found to have statistical effects that differed by level of blood

glucose: being black, being in the youngest age category, being married, and BMI. The seventh

hypothesis of a stronger association between education and lower levels of blood glucose is not

supported by the lack of an interaction between education and level of blood glucose.

In stage 1, model 4, the effect of being black on probability of EBS was insignificant. In

stage 2, model 4, being black was still insignificant in its effect on diabetes. However, the two

estimates are significantly different from each-other, supporting the fifth hypothesis of an

interaction between being black and stage of blood glucose. Being black has no statistically

significant effect on probability of either EBS or diabetes, but the effect in stage 2 is significantly

greater than in stage 1. The lack of an interaction between being Hispanic and level of blood

glucose in the association with probability of EBS supports the sixth hypothesis that the









association between EBS probability and being Hispanic does not vary with level of blood

glucose

In stage 1, model 4, being married has no association with EBS. In stage 2, model 4, being

married has a significant association with a reduced the probability of developing diabetes (b= -

0.77, p<.05). In stage 1, model 4, being overweight has a significant association with an

increased probability of EBS (b=1.32, p<.01). Like males, the direction of association of BMI

with EBS remains constant into stage two, increasing the probability of diabetes given EBS

(b=0-.77, p<.05). Again, the effect of BMI is significantly different and weaker for females in

stage 2 than in stage 1.

Sex Differences

Stage One

All tests of sex interaction were done using equation four. Three covariates were found to

have significantly different effects by sex for stage 1: having more than a high school education,

having a BA or more, and BMI. Having more than a high school degree was not significantly

associated with EBS for males, however it significantly reduced the probability of EBS for

females. The same pattern was found for having a BA or more. These findings suggest that

educational attainment has a stronger association for females than for males in reduced

probability of EBS for whites.

The third covariate found to differ by sex in stage 1 was BMI. For females, the effect of

being overweight was stronger than for males (beta(males)=0.94, beta(females)=1.32). This

suggests that, like high educational attainment, higher BMI has a stronger association with an

increased probability of EBS for females than for males.









Stage Two

Two sex differences in the effects of covariates on the probability of diabetes given EBS

were found in stage 2: the effect of marital status and an interaction term for Hispanics and

having a BA or more.For males, being married does not have a significant effect on the

probability of developing diabetes given IFG. For females however, being married has a

significant effect on reducing the probability of diabetes (b=-.54, p<.01).

The interaction term is not significantly different from zero for either males of females, but

appears to have a different direction of effect: having a BA or more for male Hispanics is

associated with higher probability of diabetes but having a BA or more for female Hispanics is

associated with a lower probability. While the term approaches significance for females, neither

term is significantly different from zero.

Figures 4-1 through 4-4 present estimated risk of EBS and diabetes given IFG for

individuals in the oldest age category, single, and overweight based on equation four in tables 4-

7 and 4-8. In Figure 4-1, one of the clearest rends is that for Hispanic males-the estimated risk

of EBS per 1,000 decreases with each increasing level of education, except for having more than

a high school degree. In Figure 4-2, Hispanics do not show the same nearly stepwise benefit

from increasing education. Whites however have a large difference in estimated relative risk of

developing diabetes. Having less than a 9th grade education for white males is estimated to

increase the probability of developing diabetes by 23% relative to a high school degree whereas

having a BA or more reduces the probability by 14%.

Figure 4-3 describes the risk of having EBS for females who are overweight, single, and

over age 45. Among Hispanics, there is a "U' shaped trend in the estimated relative risk of

developing EBS, with the lowest risk among Hispanics with a high school degree and the highest

risk in the two lower categories of education and in the highest category of education. In Figure









4-4, this 'U'-shaped trend disappears for relative risk of having diabetes for Hispanic females.

Hispanics with a BA or more are estimated to have a 76% lower probability of having diabetes,

given EBS, than Hispanics with a high school degree. Having a BA or more for blacks and

whites is also estimated to reduce probabilities of developing diabetes relative to having a high

school degree.










Table 4-1. Frequencies and Prevalence of IFG* for Individuals Aged 20 and Over Without Diagnosed Diabetes** in the
U.S. Population, NHANES 1999-2006, Aged 20 and Over Without Diagnosed Diabetes in the U.S. Population
NHANES 1999-2006,Weighted Data


Combined Reference*** Weighted Weighted Weighted
Frequencies Frequencies Frequencies Frequencies Frequencies Prevalence Prevalence Prevalence Prevalence
1999-2000 2001-2002 2003-2004 2005-2006 1999-2002 1999-2002 1999-2002 2003-2006 1999-2006

Total 479(25.9%) 617(28.0%) 523(26.6%) 605(30.5%) 1096(27.1%) 26.0% 25.3% 27.9% 26.6%

Men 282(32.6%) 380(36.1%) 290(30.7%) 365(37.9%) 662(34.5%) 33.0% 31.8% 34.2% 34.5%
Women 197(20.0%) 237(20.6%) 233(22.9%) 240(23.6%) 434(20.4%) 20.0% 19.4% 22.1% 21.8%

White 223(26.6%) 356(30.1%) 314(29.4%) 311(31.1%) 579(28.7%) 27.0% 26.2% 28.5% 27.3%
Men 140(34.5%) 215(38.0%) 181(35.2%) 203(40.0%) 355(36.6%) 34.0% 33.0% 35.0% 34.0%
Women 83(19.21%) 141(22.9%) 133(24.0%) 108(22.0%) 224(21.4%) 21.0% 20.0% 22.4% 21.2%

Black 64(19.9%) 71(18.9%) 82(22.5%) 120(27.3%) 135(19.4%) 17.0% 17.3% 23.0% 20.3%
Men 29(20.6%) 39(21.6) 36(21.0%) 62(30.1%) 68(21.1%) 19.2% 19.1% 24.7% 22.0%
Women 35(19.4) 32(16.5%) 46(23.8%) 58(24.7%) 67(17.9%) 15.0% 15.8% 21.7% 18.9%

Hisp. 148(28.6%) 154(31.4) 94(23.7%) 121(30.8%) 302(29.9%) 30.0% 29.2% 26.4% 27.7%
Men 88(36.8%) 106(45.5) 52(26.3%) 67(36.6%) 194(41.1%) 41.1% 41.2% 30.3% 35.3%
Women 60(21.6%) 48(18.7) 42(21.1%) 54(25.7%) 108(20.2%) 18.0% 15.6% 22.0% 19.0%


* Impaired Fasting Glucose defined as individuals with glucose levels 100-125 mg/dL( 5.6 to 6.9 mmol/L)
**Individual who had been told by doctor they had diabetes were excluded from the sample
*** Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fasting Glucose in Adults in the U.S. Population"
Diabetes Care. Vol. 29:1263-1268.










Table 4-2. Frequencies and Prevalence (SE) of Diagnosed Diabetes for Individuals Aged 20 and Over
in the U.S. Population, NHANES 1999-2006. Weighted Data


Combined Reference** Weighted Weighted Weighted
Frequencies Frequencies Frequencies Frequencies Frequencies Prevalence Prevalence Prevalence Prevalence
1999-2000 2001-2002 2003-2004 2005-2006 1999-2002 1999-2002 1999-2002 2003-2006 1999-2006

Total 480(9.8%) 511(9.5%) 545(10.8%) 509(10.2%) 2045(10.1%) 6.5% 6.5%2 7.7% 7.1%

Men 233(10.3%) 248(9.8%) 269(11.1%) 247(104%) 997(10.4%) 6.7% 6.7% 7.4% 7.1%
Women 247(9.5%) 263(9.2%) 276(10.5%) 262(10.1%) 1048(9.8%) 6.3% 6.3% 8.1% 7.2%

White 146(6.6%) 220(7.7%) 243(9.0%) 192(7.7%) 801(7.8%) 5.6% 5.6% 6.9% 6.3%
Men 84(7.9%) 112(8.4%) 130(10.1%) 92(7.6%) 418(8.5%) 6.1% 6.2% 6.7% 6.5%
Women 62(5.4%) 108(7.1%) 113(8.0%) 100(7.8%) 383(7.1%) 5.0% 5.1% 7.1% 6.1%

Black 130(14.3%) 118(11.7%) 124(12.5%) 169(15.1%) 541(13.4%) 10.0% 10.0% 11.7% 10.9%
Men 55(13.4%) 53(11.1%) 53(11.1)% 89(16.4%) 250(13.1%) 8.2% 8.2% 11.4% 9.9%
Women 75(15.0%) 65(12.2%) 71(13.7%) 80(13.8%) 291(13.6%) 11.4% 11.4% 12.0% 11.7%

Hisp. 151(11.8%) 127(11.4%) 146(14.8%) 113(11.3%) 537(12.3%) 6.5% 6.5% 7.9% 7.3%
Men 71(12.0%) 59(11.0%) 73(15.2%) 51(10.7%) 254(12.2%) 5.4% 5.3% 7.0% 6.2%
Women 80(11.6%) 68(11.9%) 73(14.5%) 62(11.7%) 282(12.3%) 7.8% 7.7% 9.0% 8.4%


*Physician diagnosis only of diabetes
** Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fasting Glucose in Adults in the U.S.
Diabetes Care. Vol. 29:1263-1268.


Population"










Table 4-3. Prevalence (SE) of Diabetes as Measured from Either Diagnosis
Medication Usage or Blood Analysis for Individuals Aged 20
and Over in the U.S. population, NHANES 1999-2006
Weighted Data

1999-2000 2001-2002 2003-2004 2005-2006 1999-2006

Total 8.5%(.01) 10.0%(.01) 10.3%(.01) 10.6%(.01) 9.9%(<.01)

Men 9.1%(.01) 11.9%(.01) 11.4%(.01) 9.6%(.01) 10.6%(.01)
Women 7.9%(.01) 8.3%(.01) 9.3%(.01) 11.5%(.01) 9.3%(.01)

White 7.7%(.02) 9.1%(.01) 9.2%(.01) 9.7%(.01) 9%(.01)
Men 9.6%(.02) 11.8%(.02) 11.3%(.01) 8.4%(.02) 10.3%(.01)
Women 5.9%(.01) 6.5%(.01) 7.3%(.01) 10.9%(.01) 7.7%(.01)

Black 9.6%(.01) 13.2%(.02) 14.2%(.02) 15%(.02) 13.2%(.01)
Men 5.1%(.01) 11.5%(.02) 12.9%(.03) 15.3%(.02) 11.6%(.01)
Women 12.8%(.02) 14.6%(.03) 15.2%(.03) 14.8%(.02) 14.4%(.01)

Hisp. 10.3(.01) 10%(.01) 11.6%(.03) 12.4%(.01) 11.3%(.01)
Men 8.8%(.03) 9.5%(.02) 11.4%(.03) 11.4%(.03) 10.2%(.01)
Women 11.6%(.05) 12.6%(.02) 11.8%(.04) 13.6%(.03) 12.4%(.02)
N=1,854 N=2,205 N=1,977 N=1,982 N=8,018


*Diabetes includes individual self-reports of physician diagnosis,
medication usage, and analysis of glucose levels from blood samples
with diabetic levels >=7.0 mmol per L (>=126 mg per dL)












Table 4-4. Prevalence (SE) of U.S. Population Aged 20 Years and Older with IFG*
by Educational Attainment, NHANES 1999-2006: Weighted Data

Impaired Fasting Glucose
Less Than High School High School Degree College or More

Total 31%(.013) 27.8%(.015) 24.3%(.010)

Men 36%(.019) 31.8%(.020) 31.1%(.013)
Women 26.4%(.015) 24.1%(.015) 18.1%(.011)

White 34.8%(.013) 29.8%(.017) 24.6%(.013)
Men 39.6%(.030) 34.0%(.024) 31.7%(.015)
Women 30.5%(.028) 25.9%(.018) 18%(.014)

Black 21.8%(.021) 23.4%(.026) 19.1%(.015)
Men 22.4%(.028) 25.0%(.036) 20.2%(.027)
Women 21.2%(.212) 21.9%(.030) 18.3%(.019)

Hisp. 30.6%(.016) 19.5%(.031) 21.1%(.022)
Men 38.7%(.023) 23.9%(.039) 27.6%(.031)
Women 22.2(.021)% 14.6%(.037) 15.0%(.032)
N=2,440 N=1,872 N=3,667


* Impaired Fasting Glucose defined as individuals
with glucose levels 5.6 to 6.9 mmol per L (100-125 mg per dL)










Table 4-5. Prevalence (SE) of Diabetes in U.S. Population by Educational Attainment


Less Than High Scho
15.2%(.01)


12.8%(.01)
17.3%(.02)


14.6%(.02)
14.3%(.02)
15%(.02)


17.5%(.02)
14.1%(.02)
20.8%(.03)


14.1%(.01)
10.8%(.02)
17.5%(.03)
*N=2,450


ol High School Degr
10.5(.01)


9.9%(.01)
11.1%(.01)


9.9%(.01)
9.5%(.02)
10.2%(.01)


14.2%(.02)
10.6%(.02)
17.4%(.03)


8.4%(.02)
8.5%(.03)
8.2%(.03)
N=1,880


*ee College or More
7.7%(.01)


10.1%(.01)
5.6%(.01)


7.3%(.01)
9.9%(.01)
4.9%(.01)


9.9%(.01)
10.0%(.02)
9.8%(.02)


8.6%(.02)
10.6%(.03)
6.7%(.03)
N=3,673


*Diabetes includes individual self-reports of physician diagnosis,
medication usage, and analysis of glucose levels from blood samples
with diabetic levels >=7.0 mmol per L (>=126 mg per dL)


Total


Men
Women


White
Men
Women


Black
Men
Women


Hisp.
Men
Women










Table 4-6. Proportions and Means of U.S. Born Population(SE) Over the Age of 24. Weighted
Data NHANES 1999-2006
White Black Hispanic
Age(In Years) 48(.42) 43.3(.50) 40.7(1.6)

Education
Less Than 9th Grade 4% 5% 10%
Less Than HS Degree 9% 24% 21%
High School Degree 28% 24% 22%
More Than High School 32% 33% 37%
BA or More 27% 13% 10%

Marital Status
Married 62% 33% 57%



BMI
Over 64% 74% 74%
Under 2% 2% 2%
Normal 34% 24% 24%










Table 4-7. Results From a Logistic Regression Continuation Ratio Model for Males Greater Than 24 Years Old for Having EBS
Given Normal Blood Glucose (Stage 1):NHANES 1999-2006


Constant
Race
(White=Ref)
Black
Hispanic
Age (>45=ref)
Age24-35
Age 36-45
Edu.
(HSDG=Ref)
< 9th Grade
> High School
BA or More
Married (l=Yes)
BMI(Normal=ref)
Over
Educ*Race
Black* LT9th
Black* LTHS
Black* MTHS
Black* BAOM
Hisp.*LT9th
Hisp.LTHS
Hisp.*MTHS
Hisp.*BAOM
*=p<.05; **p<.01;


Model 1
Coef.
0.27


-0.28
0.14

-1.6
-0.84


SE
0.07**


0.12*
0.16

0.13**
0.12**


Model 2
Coef.
0.33


SE
0.09**


-0.36 0.12**
0.07 0.16


-1.56
-0.84


0.04
0.13
0.04
-0.33


0.13**
0.13**


Model 3
Coef.
0.62


-0.38
-0.05

-1.54
-0.86


0.19 0.14
0.16 0.22
0.11 0.04
.13* -0.26
-0.12


0.94


SE
.15**


.13**
0.19

0.13**
.13**


0.19
0.14
0.1
0.13
0.11


10** 0.94** 0.09


0.02 0.46
-0.16 0.32
-0.51 0.28


x=sig. diff. by stage; #=sig diff. by sex


Sex diff.


Stage diff


Model 4
Coef
0.61


-0.17
-0.36


-1.53
-0.86


0.02
0.19
0.09
-0.25
-0.11


SE
0.15**


0.19


0.21
0.18
0.11
0.13
0.11


-0.09
1.12
0.7
0.14
0.22


0.4
0.75
0.63
0.22
0.71










Table 4-8. Results from a Logistic Regression Continuation Ratio Model for Males Greater Than 24 Years Old for Having Diabetes
Given EBS (Stage 2):NHANES 1999-2006
Model 1 Model 2 Model 3 Model 4 Sex diff. Stage diff.
Coef SE Coef SE Coef SE Coef SE
Constant -1 0.09 -1.1 0.17 -1.01 0.22 -1.01 0.23**


Race (White=Ref)
Black
Hispanic
Age (>45=ref)
Age24-35
Age 36-45
Edu. (HSDG=Ref)
< 9th Grade
> HS
BA or More
Married (l=Yes)
BMI(Normal=ref)
Over

Educ*Race
Black* LT9th
Black* LTHS
Black* MTHS
Black* BAOM
Hisp.*LT9th
Hisp.LTHS
Hisp.*MTHS
Hisp.*BAOM
*=p<.05; **p<.01;


0.56 .18**
0.7 0.33*


-1.13
-0.42


.35**
0.25


0.52
0.65

-1.14
-0.4

0.14
0.12
0.27
-0.19


.19*
0.32*

0.35**
0.26

0.27
0.25
0.22
0.22


0.54 .21*
0.62 .25*


-1.2
-0.48

0.19
0.09
0.23
-0.18
0.05


0.56 0.40
0.25 0.57


0.36**
0.26


0.29
0.25
0.22
0.23
0.17


0.52 .22*


-1.23
-0.48

0.3
-0.04
0.23
-0.2
0.05


.37**
0.27

0.34
0.39
0.25
0.25
0.17


0.51 0.23*


-0.14 0.56
0.32 0.45
-0.35 0.53
0.12 0.67
-0.41 0.88
0.76 0.73
0.57 0.68
0.72 0.97


x=sig. diff. by stage; #=sig diff. by sex










Table 4-9. Results From a Logistic Regression Continuation Ratio Model for Females Greater Than 24 Years Old for Having EBS
Given Normal Blood Glucose (Stage 1):NHANES 1999-2006


Constant
Race
(White=Ref)
Black
Hispanic
Age (>45=ref)
Age24-35
Age 36-45
Edu.
(HSDG=Ref)
< 9th Grade
> HS
BA or More
Married (l=Yes)
BMI(Normal=ref)
Over
Educ*Race
Black* LT9th
Black* LTHS
Black* MTHS
Black* BAOM
Hisp.*LT9th
Hisp.LTHS
Hisp.*MTHS
Hisp.*BAOM


Model 1
Coef.
-0.43


0.35
0.43


-1.95
-0.42


SE


Model 2
Coef.


-0.28


.12** 0.23
.14** 0.21


.18** -1.9
.08** -0.93


0.41
0.39
-0.27
-0.56


SE
0.09**


Model 3
Coef.
0.22


SE
0.11*


Model 4
Coef
0.23


Sex diff. Stage diff.
SE
0.12


.12* -0.03 0.12 0.05 0.23
0.15 0.14 0.17 -0.39 0.33


0.18** -1.9
0.14** 0.88


.17*
.16*
.12*
.14**


0.47
0.36
-0.32
-0.51
-0.15


1.33


.18** -1.9 .19*
.14** -0.89 0.15


.19*
.16*
.13*
.14**
0.09


0.7
0.39
-0.35
-0.57
-0.15


.23**
0.21
0.15*
.16**
0.1


.13** 1.32


-0.94
-0.22
0.01
0.13


0.47*
0.34
0.31
0.37


-0.27 0.46


0.5
0.53
1.6


0.37
0.46
0.55**


*=p<.05; **p<.01; x=sig. diff. by stage; #=sig diff. by sex


.08**










Table 4-10. Results from a Logistic Regression Continuation Ratio Model for Feales Greater Than 24 Years Old for Having
Diabetes Given EBS (Stage 2):NHANES 1999-2006


Constant
Race
(White=Ref)
Black
Hispanic
Age (>45=ref)
Age24-35
Age 36-45
Edu.
(HSDG=Ref)
< 9th Grade
> HS
BA or More
Married (l=Yes)
BMI(Normal=ref)
Over
Educ*Race
Black* LT9th
Black* LTHS
Black* MTHS
Black* BAOM
Hisp.*LT9th
Hisp.LTHS
Hisp.*MTHS
Hisp.*BAOM


Model 1
Coef.
-0.87


0.84
0.38

-0.36
0.58


SE
.10**


Model 2
Coef.
-0.78


.15** 0.83
0.2 0.3


Model 3
SE Coef.
0.15 -0.35


.15** 0.59
0.19 0.39


0.4 -0.27 0.42 -0.33
0.3 -0.56 0.3 -0.49


0.16
0.06
-0.49
-0.59


0.34
0.26
0.26
0.34


0.16
-0.07
-0.5
-0.55
-0.52

0.74


Model 4
SE Coef.
0.18 -0.36


.15** 0.64
0.21 0.92

0.39 -0.34
0.32 -0.49


0.35
0.27
.25*
0.34
0.18**


0.29
-0.17
-0.5
-0.36
-0.54


.29* 0.77


-0.18
0.21
0.08
-0.84
-1.30
-0.13
-0.44
-1.90


Sex diff. Stage diff
SE
0.19


0.34
0.50

0.41
0.33


0.40
0.36
0.30
0.37
0.18**

.30*


0.77
0.56
0.54
0.69
0.86
0.68
0.67
1.20


*=p<.05; **p<.01; x=sig. diff. by stage; #=sig diff. by sex











Estimated Risk (per 1,000) of

Elevated Blood Glucose


BAOM

>HSDG

HSDG


<9th


600 800 1000




<9th


*WHITE 652 690 648 668 589
*HISPANIC 801 758 562 618 555
SBLACK 565 562 555 450 470




Estimated Relative Risk (Compared

to High School)


BAOM

>HSDG

HSDG


<9th

0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60

<9th HSDG BAOM
*WHITE 1.01 1.06 1.00 1.03 0.91
SHISPANIC 1.42 1.35 1.00 1.10 0.99
* BLACK 1.02 1.01 1.00 0.81 0.85


Figure 4-1. Estimated Risk of EBS by Educational Attainment and Race For Males Over Age
45: A) Estimated Risk and B) Estimated Relative Risk


I I


HSDG


.1I


>HSDG


BAOM


mm
















BAOM
>HSDG

HSDG


< 9th


Estimated Risk (per 1,000) of

Elevated Blood Glucose


300 400


* WHITE 330 259 267 314 230
*HISPANIC 295 490 319 510 440
*BLACK 428 458 389 361 371




Estimated Relative Risk (Compared

to High School


BAOM

>HSDG

HSDG


<9th

0.00 0.50 1.00 1.50 2.00

<9th HSDG BAOM
WHITE 1.23 0.97 1.00 1.18 0.86
EHISPANIC 0.93 1.54 1.00 1.60 1.38
SBLACK 1.10 1.18 1.00 0.93 0.95


<9th


0 100 200


HSDG


>HSDG


Figure 4-2. Estimated Risk of Diabetes Given EBS by Educational Attainment and Race: A)
Estimated Risk and, B) Estimated Relative Risk




BAOM











Estimated Risk (per 1,000) of
Diabetes Given EBS


-I-I


BAOM

>HSDG

HSDG


< 9th


I I


0 100 200 300 400 500 600 700 800




<9th


*WHITE 717 650 557 470 416
HISPANIC 567 675 460 505 705
SBLACK 510 611 570 485 460
A


Estimated Relative Risk (Compared

to High School

BAOM

>HSDG

HSDG


< 9th

0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80

<9th HSDG BAOM
*WHITE 1.29 1.17 1.00 0.84 0.75
HISPANIC 1.23 1.47 1.00 1.10 1.53
mBLACK 0.90 1.07 1.00 0.85 0.81
B
Figure 4-3. Estimated Risk of EBS by Educational Attainment and Race For Females Over Age
45, Single, and Overweight: A) Estimated Risk and B) Estimated Relative Risk


HSDG


BAOM


.1~


>HSDG


mm











Estimated Risk (per 1,000) of

Diabetes Given EBS


BAOM

>HSDG

HSDG


< 9th


- pM


0 100 200 300 400 500 600 700




<9th


7WHITE 483 371 411 297 327
*HISPANIC 389 565 636 406 154
SBLACK 596 579 570 465 285




Estimated Relative Risk (Compared

to High School

BAOM
>HSDG
HSDG
< 9th

0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40

<9th HSDG BAOM
*WHITE 1.17 0.90 1.00 0.72 0.80
*HISPANIC 0.61 0.89 1.00 0.64 0.24
mBLACK 1.05 1.02 1.00 0.82 0.50

Estimated Risk of Diabetes Given EBS by Educational Attainment and Race For
Females Over Age 45, Single, and Overweight: A)Estimated Risk and B) Estimated
Relative Risk


I wr


HSDG


>HSDG


BAOM


A























B
Figure 4-4.









CHAPTER 5
DISCUSSION AND CONCLUSION

Discussion

The objective of this thesis was to answer the following three questions: (1) What is the

association between education and the development of elevated blood sugar, (2) does this

association vary by race and ethnicity and sex, and (3) does the association between race and

ethnicity, sex, and elevated blood sugar vary depending on the level of blood sugar? Seven

hypotheses emerged from these three research questions.

First, it was hypothesized that as the level of education increased, the probability of having

EBS would decrease. The results of the analysis suggest that there is indeed a positive

association between the development of EBS and educational attainment for females but not for

males. Females with less than a high school degree have a significantly increased probability for

developing EBS, and females with more than a high school education have a significantly

decreased probability of developing EBS. Moreover, the effect of education is found to vary

significantly by sex. The effect of having more than a high school education on probability of

EBS is significantly greater for females than for males. This finding supports previous research

that suggests increasing socioeconomic status attainment has a stronger effect for females than

for males for blood glucose related outcomes (Stern et al. 1984).

Only limited little evidence is found in support of the second and third hypotheses of an

interaction between race and educational attainment. Two interactions were found to be

significant for females, however interpretation of these terms is difficult. In stage 1, being black

significantly reduced the probability of diabetes given less than a 9th grade education. Figure 4-

3 aids in the interpretation of this interaction term by estimating probabilities by level of

education. A downward trend in the relative risk of developing diabetes as educational









attainment increases can be observed for white females; however no such trend is observed for

black females. Thus for females, there is some evidence in support of the third hypothesis. Past

research has suggested that a downward trend of diabetes with increasing education for whites

exists, but not for blacks (Borrell 2006). One possibility is that these associations happened by

chance-i.e. noise in the data. Only 57 black females had less than a 9th grade education across

all eight years, making a chance significant finding plausible. Another possibility is that one of

the previous reasons given for expecting variation in the effect of education plays a role for

blacks. It could be the effect of genes: black females may be predisposed to development of EBS

due to a "thrifty gene", thereby reducing the protective effect of education. Pre-natal

environments (Barker hypothesis) may be worse for females, and the mechanism their bodies

adapt to scarcity may predispose them to risk if their post-natal environment overwhelms their

metabolic mechanisms for dealing with glucose. Cultural capital explanations may also play the

role: it could be that black females and the schools they attend are ill-equipped to aid in their

accumulation of human capital that pays dividends in later health outcomes. Unfortunately, these

data makes it impossible to disentangle the multitude of early-life and current life events, as well

as direct measurement of concepts such as "capital", precluding a direct test of any of these

hypotheses.

The second significant interaction was among Hispanic females with a BA or more. In this

case, being Hispanic and having a BA or more had a significant effect on increasing the

probability of EBS. This interaction does not support the third hypothesis that the statistical

effect of education would be similar for whites and Hispanics. Again, there are several possible

but un-testable explanations for this interaction term. It could be noise in the data: Only Hispanic

39 females had a BA or more across all eight years, and 47% of these highly educated Hispanics









had EBS. Again, early life events could have an influence on this outcome: Hispanic females

with a BA probably have greater access to food resources than lesser educated Hispanic females.

If genetic predispositions or early-life events primed a "thrifty" gene, then greater access to food

resources may not necessarily predict better health outcomes in terms of blood glucose and blood

glucose-related health risks. Another explanation is that having a BA or more is a proxy for level

of acculturation. Acculturation has been found to have varying effects on the probability of

developing diabetes. Some studies suggest that among Mexican-Americans, greater acculturation

is associated with lower risk of diabetes, whereas other studies suggest that higher levels of

acculturation are associated with greater risk of diabetes for non-Mexican Hispanics (Hazuda

1988, Kandula 2008).

The association with acculturation may vary by ethnic origin and may also vary by time.

As American society becomes increasingly overweight, acculturation to such a lifestyle may be

increasingly associated with risk of diabetes. Evidence of the reduced effect of having a BA or

more on lowering the probability of EBS for Hispanics lends support to the possibility that

acculturation to a US lifestyle may play a role in increasing risk of EBS. The effect of the

interaction between Hispanics and higher levels of education on increasing the probability of an

adverse outcome runs counter to traditional ideas of a Hispanic paradox. Typically, Hispanics

have been found to have lower mortality rates than would be expected given the socioeconomic

status. This has often been attributed to "cultural" differences, although recent research indicates

the effect is likely due to reverse migration (Palloni 2004). Since the health outcome of interest is

EBS and not morality, then the findings aren't a contradiction. Rather, it lends support to the

reverse migration explanation insofar as the interaction term is valid. Future research should









adjust for levels of acculturation to see if it accounts for some of the counter-intuitive effects of

education on Hispanic females at the extreme end of educational attainment.

Racial/ethnic differences and sex differences in associations with EBS also exist. No

evidence is found in support of the fourth hypothesis of no sex and education interactions. There

is strong support for the fifth hypothesis however. Among males and females, blacks are at lower

risk of developing impaired fasting glucose relative to whites, even after adjustments for age,

education, marital status and body mass index. This seems counter-intuitive given the higher risk

of diabetes for blacks, but is a finding borne out in other research (Cowie 2002).

With regards to the third objective, what makes the finding of the protective effect of being

black on IFG especially interesting is the significant reversal of the effect of being black on

developing diabetes, given one has EBS. Among blacks, this significant difference of effect is

found for both males and females.

There is mixed evidence for the sixth hypothesis. Among males, being Hispanic has a

significantly different effect depending on the level of blood glucose. No such interaction is

found for females. Unlike blacks, being Hispanic is associated with an increased probability of

EBS at both stages of blood glucose. Given EBS, the statistical effect of being Hispanic on the

probability of having diabetes is significantly different, and stronger than association being

Hispanic has when modeling EBS alone.

The seventh hypothesis stated that as the level of blood glucose increased from EBS to

diabetes given IFG, the association of education with the probability of EBS would decrease.

This hypothesis was not supported as education is found to have no statistical effect on

increasing the probability of developing diabetes, given EBS. It is important to remember,

however, that this doesn't mean education has no association with the likelihood of developing









diabetes, rather, only that it is not associated with increased probability of progression from EBS

to diabetes.

For males, BMI is the only other variable with a different effect across stages of EBS. The

difference of effect of BMI is also seen in females. The effect of BMI on increasing the

probability of developing diabetes is significant for both males and females and for both stages

of blood sugar. However, the effect of BMI on increasing risk of IFG is significantly greater than

the effect of BMI on increasing risk of diabetes. These findings suggest that while BMI is an

important risk factor for EBS, other factors help to "push" an individual across the edge into

diabetes. One avenue of further research would be to differentiate between "overweight" and

"obesity" in the CRM model to test if obesity retains an equally significant effect on risk of

diabetes. Another possibility would be to examine race by BMI interactions to see if that extra

something need for BMI to push people into being diabetic is being in a particular racial and

ethnic group. These BMI findings also demonstrates one advantage of using BMI coded

ordinally: it allows one to capture the varying effects of stages of BMI on risk of EBS.

For females, the effect of BMI is also found to be significantly different depending on

level of blood glucose. Unique to females however, is the significant reversal of the effect of

being married. Being married has no significant effect on probability of EBS but has a significant

effect on reducing the probability of diabetes given EBS.

A potential limitation of this analysis is "over-controlling"-covariates such as low

educational attainment are arguably a part of the "minority experience". If one conceptualizes

race as a marker for a wide range of social experiences that influence health outcomes, then even

if every conceivable experience was controlled for and explained the relationship, a health

disparity would remain so long as race and ethnicity exerted constraints upon an individual's life









experience. This conceptualization views the health-related factors that contribute to the health

disparity of elevated blood sugar to "lie along a causal pathway by which race and ethnicity

affects health and ... have their roots in an injustice."(Herbert 2008). By adjusting for BMI, age,

and marital status, this paper can be criticized for "over-controlling" when looking for significant

racial differences. However, one objective of the analysis is to test for different effects of the

variables across different stages of elevated blood sugar, thus not including them in attempts to

avoid "over-controlling" would preclude this objective.

Conclusion

This thesis began with the conceptualization of EBS as a racial and ethnic health disparity,

implying that differences in health outcomes related to blood glucose can and should be

addressed. EBS is an especially important health outcome to address in the U.S today: it is

increasing in incidence and prevalence and is a risk factor for the leading cause of death in the

U.S. today, cardiovascular disease.

Disparities such as that associated with EBS can occur as a result of health behaviors,

health care access, and health care process. This paper has conceptualized education as one

possible factor that might explain some of the racial disparity in EBS. Education certainly has an

effect within all three origins of health disparities. Those who are educated practice healthier

lifestyles, have better jobs, greater levels of health insurance, and can afford better quality health

care. If the effect of education had been found to be significantly different by race and ethnicity,

then further research into why the effect varied would be warranted in an attempt to maximize

the beneficial effect of education for all. This research only found the effect of education to vary

at 2 intersections of race and ethnicity and education: Hispanics with a BA or more are at an

increased risk of EBS and blacks with less than a 9th grade education are at a decreased risk.

Whether these variations are due to noise, early life differences, different experiences within the









educational setting, or different abilities to translate educational resources into better health

behaviors access and process requires further research.

A strength of this study is its conceptualization of elevated blood glucose as occurring

along a spectrum, the categorization of the spectrum into separate stages, and the testing for

differences of effect of the covariates across the stages of blood glucose. Using a CRM, the

effect of race, education, BMI, and marital status were all found to have significantly different

effects given the level of blood glucose. Of these, perhaps the finding that BMI has a

significantly stronger association with the probability ofEBS than on diabetes given EBS

provides the most immediate recommendations for health interventions. While obesity has been

recognized a primary factor in the increasing prevalence of diabetes, it plays an even more

important role in the development of EBS as a whole. Continuing efforts at reducing the

increasing rates of obesity should also include the reduction of elevated blood sugar as a central

obj ective.









APPENDIX
ADDITIONAL EDUCATIONAL ATTAINMENT

Table A-1. Percentage of Educational Attainment Within Five Category. Educational Attainment and
By Race/Ethncity in the U.S. Population. NHANES 1999-2006 Weighted Data
Less Than Less Than High School Some College/ BA+
9th Grade High School Degree Training

Black 5.9% 26.2% 24.1% 31.0% 12.6%

Hispanic 8.4% 19.5% 25.7% 33.4% 12.6%

White 3.5% 9.9% 27.7% 31.0% 27.8%









Table A-2. Prevalence (SE) of U.S. Born Population Aged 20 years and Older With IFG* By
Educational Attainment, NHANES 1999-2006, Weighted Data
Less Than High School High School Degree College or More


Total 31.30%

Men 34.5%
Women 28.4%


White
Men
Women

Black
Men
Women

Hisp.
Men
Women


29.10%

33.6%
24.8%

30.2%
35.9%
25.7%

24.2%
26.1%
22.5%


34.5%
38.4%
30.8%

22.5%
23.0%
22.0%


30.8% 21.6%
38.4% 25.4%
25.2% 17.6%o
N=1,410 N=1,603 N=3,170
* Impaired Fasting Glucose defined as individuals
with glucose levels 5.6 to 6.9 mmol per L (100-125 mg
per dL)


24.30%

30.9%
18.5%

24.8%
31.7%
18.3%

19.2%
21.0%
18.1%

18.9%
19.6%
18.2%









LIST OF REFERENCES


Adams, Scott J. 2002. "Educational Attainment and Health: Evidence from a Sample of Older
Adults." Education Economics. 10(1): 97-110.

Agency for Healthcare Research and Quality. 2005. "Economic and Health Costs of Diabetes:
HCUP Highlight 1". AHRQ Publication No. 05-0034, Retrieved on March 18th 2009
from http://www.ahrq.gov/data/hcup/highlightl/highl.htm

Allison, Paul. 1999. Logistic Regression Using SAS: Theory and Application. Cary, North
Carolina: SAS Institute, Inc.

American Diabetes Association. 2004."Position Statements: Diagnosis and classification of
diabetes mellitus" Diabetes Care. 27(Suppl. 1):S5-S10.

Ardern, Chris and Peter Katzmarzy. 2007. "Geographic and Demographic Variation in the
Prevalence of the Metabolic Syndrome in Canada." Canadian Journal of Diabetes
31(1):34-46.

Bertoni, Alain, Julie Krop, Gerard Anderson and Frederick Brancati. 2002."Diabetes-Related
Morbidity and Mortality in a National Sample of U.S. Elders." Diabetes Care 25:471-
475

Brancati, Frederick, P.K. Whelton, L.H. Kuller and M.J. Klag. 1996. "Diabetes mellitus, race,
and socioeconomic status:a population-based study." Annals ofEpidemiology. 6:67-73.

Brown, JB. G.A. Nichols, H.S. Glauber and A.W. Bakst. 1999. "Type 2 diabetes: incremental
medical care costs during the first 8 years after diagnosis." Diabetes Care 22(7): 1116-
1124.

Borrell, Luisa, Florence J. Dallo and Kellee White. 2006. "Education and Diabetes in a Racially
and Ethnically Diverse Population." American Journal of Public Health 96:1637-1642.

Burns, N.; Finucane, F.; Hatunic, M 2007. "Early-onset type 2 diabetes in obese white subjects
is characterized by a marked defect in beta cell insulin secretion, severe insulin resistance
and a lack of response to aerobic exercise training." Diabetologia 50(7):1362-1364

Caliif, RM, M. Boolell, S.M. Haffner, M.A Bethel, J. McMurray, A. Duggal and R.R. Holman.
2008. "Prevention of diabetes and cardiovascular disease in patients with impaired
glucose tolerance: rationale and design of the Nateglinide And Valsartan in Impaired
Glucose Tolerance Outcomes Research (NAVIGATOR) Trial." American Heart Journal
4:623-32.

Cagney, K.A. and Diane S. Lauderdale. "Education, Wealth, and Cognitive Function in Later
Life." Journals of Gerontology 57(2): 163-172.









Centers for Disease Control and Prevention. 1990. "Morbidity and Mortality Weekly Report:
Current Trends in Regional Variation in Diabetes Mellitus Prevalence -- United States,
1988 and 1989." Retrieved on February 14th, 2009 from
http://www.cdc.gov/mmwr/preview/mmwrhtml/00001830.htm

Centers for Disease Control and Prevention. 2008a "Crude and Age-Adjusted Percentage of
Civilian, Non-institutionalized Population with Diagnosed Diabetes, United States, 1980-
2006" Retrieved on February 14th, 2009 from
http://www.cdc.gov/diabetes/statistics/prev/national/figage.htm

------. 2008b. "Morbidity and Mortality Weekly Report: State-Specific Incidence of Diabetes
Among Adults --- Participating States, 1995--1997 and 2005--2007."
Retrieved on April 10th, 2009 from
http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5743a2.htm

Choi, B.C. and F. Shi. 2001." Risk factors for diabetes mellitus by age and sex: results of the
National Population Health Survey." Diabetologia 44(10):1221-1231.

Cohen, D.A., B. Finch, A. Bower and A. Sastry. 2005. "Collective efficacy and obesity: The
potiental influence of social factors on health." Social Science and Medicine. 62(3):769-
778.

Cowie, C. 2006. "Prevalence of diabetes and impaired fasting glucose in adults in the U.S.
population: National Health and Nutrition Examination Survey 1999-2002". Diabetes
Care 29(6):1263-1268.

Cutler, David M. and Adriana Lleras-Muney. 2007."Education and Health: Evaluating Theories
and Evidence. NBER Working Paper No. W12352 Retrieved on April 20th, 2009 from
http://ssrn.com/abstract=913315

Elo, I. and S. Preston. 1996. Educational differences in mortality: United States, 1979- 1985"
Social Science and Medicine 42: 47-57.

Ettaro, Lorraine, Thomas Songer, Ping Zhang, and Michael Englegau. 2004. "Cost-of-illness
studies in diabetes mellitus." Pharmacoeconomics 22(3): 149-64.

Ford, Earl, Wayne Giles and William Dietz. 2002. "Prevalence of the Metabolic Syndrome
Among US Adults Findings From the Third National Health and Nutrition Examination
Survey." Journal of the American Medical Association. 287:356-359.

Feinstein, Jonathon S. 1993. "The Association Between Socioeconomic Status and Health: A
Review of the Literature." The Milbank Quarterly, 71: 279-322.

Grossman, M. and R. Kaestner. 1997. "Effects of education on health." J. R. Behrman and N.
Stacey. (Eds) The Social Benefits ofEducation. Ann Arbor, MI: University of Michigan
Press.










V L Goetsch, D.J. Wiebe, L.G. Veltum, and B. Van Dorsten. 1990. "Stress and blood glucose in
type II diabetes mellitus" Behaviour Research and Therapy. 28(6):531-537.

Goldberg, J., W. Hayes and J. Huntley. 2004. "Understanding Health Disparities." Health Policy
Institute of Ohio

Gregg Beckles, Leveille Williamson and Narayan Engelgau. 2000. "Diabetes and physical
disability among older U.S. adults." Diabetes Care. 23:1272-7

Herbert, P.L., J.E. Sisk, and E.A. Howell. 2008. "When does a difference become a disparity?
Conceptualizing racial and ethnic disparities in health." Health Affairs 27(2): 374-3 82.

Hermann, Nabi, Mika Kivimaki, Michael G Marmot, Jane Ferrie Marie Zins Pierre
Ducimetiere, Silla M Consoli, and Archana Singh-Manoux 2008."Does personality
explain social inequalities in mortality?" International Journal ofEpidemiology. 37:591-
602.

loannou, George N., Chris L. Bryson, and Edward Boyko. 2007 "Prevalence and trends of
insulin resistance, impaired fasting glucose, and diabetes." Journal of Diabetes and its
Complications. 21:363-370.

U.S. Department of Health and Human Services. Healthy People 2010. 2nd ed. With
Understanding and Improving Health and Objectives for Improving Health. 2 vols.
Washington, DC: U.S. Government Printing Office, November 2000.

Kitagawa, Evelyn M. and Philip Hauser. 1968. "Education Differentials in Mortality by Cause of
Death :United States, 1960." Demography. 5:318-353.

Kouznetsova, Maria, Xiaoyu Huang, Jing Ma, Lawrence Lessner and David 0. Carpenter. 2007.
"Increased Rate of Hospitalization for Diabetes and Residential Proximity of Hazardous
Waste Sites." Environmental Health Perspectives 115(1):75-79.

Mayfield, J.A. and Deb P. Whitecotton. 1999. "Work disability and diabetes." Diabetes Care.
22(7):1105-9.

Meara, E.R., S. Richards, and D.M. Cutler. 2008. "The Gap Gets Bigger: Changes in Mortality
and Life Expectancy, by Education, 1981-2000." Health Affairs 27(2): 350-360.

Mcguire, D.K. and J.L. Petersen. 2005. "Impaired glucose tolerance and impaired fasting
glucose--a review of diagnosis, clinical implications and management." Diabetes and
Vascular Disease Research 2(1):9-15.

McWilliams, Michael, Meara Ellen and Alan M. Zaslavsky. 2007. "Health of Previously
Uninsured Adults After Acquiring Medicare Coverage." JAMA. 298(24): 2886-2894.









Mirowski, John, and Catherine Ross. 2003. Education, Social Status, andHealth. New York.
New York: Aldine de Gruyter

Moen, E. 1999 "Education, ranking, and competition forjobs". Journal of Labor Economics.
17:694-723.

Mokdad Ali H., Barbara A Bowman, and Earl S. Ford. 2001. "The Continuing Epidemics of
Obesity and Diabetes in the United States." Journal of the American Medical
Association. 286(10):1195-1200

Monteiro, Carols, Conde Wolney, and Barry Popkin. 2001. "Independent Effects of Income and
Education on the Risk of Obesity in the Brazilian Adult Population." Journal of Nutrition
131:881-886.

National Center for Health Statistics. 2009. "Sample Design." Retrieved April 12th 2009 from
http://www.cdc.gov/nchs/tutorials/Nhanes/SurveyDesign/SampleDesign/intro.htm

National Institute of Diabetes and Digestive and Kidney Diseases. National Diabetes Statistics,
2007 fact sheet. Bethesda, MD: U.S. Department of Health and Human Services,
National Institutes of Health, 2008.

Nichols, Gregory A., Teresa Hillier, and Jonathon Brown. 2007. "Progression from Newly
Acquired Impaired Fasting Glucose to Type 2 Diabetes." Diabetes Care 3:228-233.

Ones, Deniz S. and Neil Anderson. 2002. "Gender and Ethnic Group Differences on Personality
Scales in Selection: Some British Data." Journal of Occupational and Organizational
Psychology". Journal of Occupational & Organizational Psychology. 75(3):255-276.

Peterson, J.L, and D.K. McGuire. 2005. "Impaired glucose tolerance and impaired fasting
glucose-a review of diagnosis, clinical implications and management". Diabetes and
Vascular Disease Research. 2(1):9-15.

Palloni, Alberto and Elizabeth Arias. 2004. "Paradox Lost: Explaining the Hispanic Adult
Mortality Advantage." Demography, Vol. 41(3) pp. 385-415

Palloni, Alberto. 2006. "Reproducing Inequalities: Luck, Wallets, and the Enduring Effects of
Childhood Health." Demography. 43:587-615.

Rabi, Doreen, Alun Edwards, Danielle Souther, Larence Svenson, Peter Sargious, Peter Norton,
Eric Larsen, and William Ghali. 2006. "Association of socio-economic status with
diabetes prevalence and utilization of diabetes care services." BMC Health Services
Research 6:124.

Preston, S.H., and I.T. Elo. 1995. Are Educational Differentials in Adult Mortality Increasing in
the United States? Journal ofAging and Health 7(4): 476-496.










Rao, Shobha S., Phillip Disraeli, and Tamara McGregror. 2004. "Impaired Glucose Tolerance
and Impaired Fasting Glucose." American Family Physician. Retrieved on November 20,
2008 from http://www.aafp.org/afp/20040415/1961.html

Robbins, J.M., V. Vaccarino, H. Zhang, and S.V. Kasl. 2005. "Socioeconomic Status and
Diagnosed Diabetes Incidence." Diabetes Research and Clinical Practice. 68(3):230-
236.

Rose, Elaina. 2003. "Does Education Really Disadvantage Women in the Marriage Market ?"
University of Washington Institute for Economic Research Working Paper No. UWEC-
2003-15. Retrieved April 1 2009 from
SSRN: http://ssrn.com/abstract=423360 or DOI: 10.2139/ssm.423360

Rosenbloom, A.L., J.R. Joe, R.S. Young, and W.E. Winter. 1999. Emerging epidemic of type 2
diabetes in youth." Diabetes Care, 22(2):345-35

Ross Catherine E. and Chia-Lang Wu. 1995. "The Links Between Education and Health"
American Sociological Review 60(5): 719-745.

Ross Catherine E., and Chia-Lang Wu. "Education, Age, and the Cumulative Advantage in
Health." Journal of Health and Social Behavior 37(1): 104-120.

Ross, Catherine E., and John Mirowsky. 1999. "Refining the Association between Education and
Health: The Effects of Quantity, Credential, and Selectivity." Demography 36:4:445-
460.

Shai, Im, R. Jiang, and J.E. Manson. 2006. "Ethnicity, obesity, and risk of type 2 diabetes in
women: a 20-year follow-up study. Diabetes Care. 29:1585-90

Shea, S., A. Stein, C. Basch, P. Lantigua,, C. Maylahn, D. Strogatz and L. Novick. 1991
"Independent associations of educational attainment and ethnicity with behavioral risk
factors for cardiovascular disease." American Journal ofEpidemiology 134:567-582

Shoenborn, C.A. 2004. "Marital status and health: United States, 1999-2002." Advanced Data
351:1-32.

Smedley, Brian D., Adrieene Stith, and Alan R. Nelson. 2003. Committee on Understanding and
Eliminating Racial and Ethnic Disparities in Health Care. Washington D.C. : The
National Academies Press

Smith, James. 2007. "Diabetes and the Rise of the SES Health Gradient". NBER Working Paper
No. W12905. Retrieved April 1 2009 from http://ssrn.com/abstract=963738

Smith, James. 2005. "Unraveling the socioeconomic status: Health Connection" Population and
Development Review. 30:108-132.













Snehalatha, C., A. Ramachandran, K. Satyavani, S.Sivasankari, and V. Vijay. 2003 "Clustering
of cardiovascular risk factors in impaired fasting glucose and impaired glucose
tolerance." International Journal of Diabetes in Developing Countries 23:59-61.

Stunkard A.J. and Thorkild I.A. Sorensen 1996. Obesity and Socioeconomic Status: A
Complex Relation." The New England Journal of Medicine 329:1036-1037.

Valdmanis, D.W. Smith and M.R. Page. 2001."Productivity and economic burden associated
with diabetes." American Journal of Public Health, 91(1):129-130.

Whitehead, M. 1992."The concepts and principles of equity and health." International Journal of
Health Services 22(3):429-45.

Winkelby, M.A., D.E. Jatulis, E. Frank, and S.P. Fortmann. 1992. "Socioeconomic status and
health: how education, income, and occupation contribute to risk factors for
cardiovascular disease." American Journal of Public Health. 82(6): 816-820.

Zimmer, Zachary, Albert Hermalin, and Hui-Sheng Ling. 2002. "Whose Education Counts? The
Added Impact of Adult-Child Education on Physical Functioning of Older Taiwanese"
The Journals of Gerontology Series B: Psychological Sciences and Social Sciences
57:S23-S32









BIOGRAPHICAL SKETCH

Gregory Pavela earned his Bachelor of Arts from the University of Virginia with majors in

sociology and history. He is currently enrolled as a graduate student at the University of Florida

and received his Master of Arts in sociology from the University of Florida in 2009. He plans to

continue his research on diabetes and health disparities.









EDUCATION DIFFERENCES IN ELEVATED BLOOD SUGAR: DO THEY VARY BY
RACE, ETHNICITY AND SEX?

Gregory Pavela
(540) 645-0579
Sociology and Criminology & Law
John Henreta
Master of Arts
July 2009

This thesis develops our understanding of the risk factors associated with impaired fasting

glucose (pre-diabetes) and diabetes. Diabetes is a risk factor for cardiovascular disease, the

leading cause of death in the U.S., and as such it is important to understand how known risk

factors such as race and ethnicity, obesity, and education interact with each other and with the

level of disease to increase the risk of development of diabetes. The goal of such an

understanding is to reduce the overall prevalence of the disease and to reduce racial and ethnic

disparities in diabetes.





PAGE 1

1 EDUCATION DIFFERENCES IN ELEVATED BLOOD GLUCOSE : DO THEY VARY BY RACE, ETHNICITY AND SEX? By GREGORY PAVELA A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Gregory Pavela

PAGE 3

3 To my parents

PAGE 4

4 ACKNOWLEDGMENTS I must thank my committee, and especially Dr. Henretta, for the inordinate amount of tim e the titles of tables, formatting, and logic of argument and reache s all the way to handwritten Statistical Analysis Software syntax to introduce me to the p rogram. The process of writing this thesis, in preparation for future work, has been the best and most useful part of my education off and ill improve my a nswers to future questions.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURE S ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ..... 9 CHAPTER 1 HEALTH DISPARITIES ................................ ................................ ................................ ....... 11 Introduction ................................ ................................ ................................ ............................. 11 Conceptualizing Health Disparities ................................ ................................ ........................ 13 Socioeconomic Status Health Disparities in Elevated Blood Sugar ................................ ....... 15 Framework Linking Socioeconomic S tatus to EBS ................................ ............................... 16 Education as a Socioeconomic Status Measure and General Health Outcomes .................... 17 Measuring Education ................................ ................................ ................................ .............. 19 Pathways Between Education and Health ................................ ................................ .............. 21 Education and Endogeneity ................................ ................................ ................................ .... 23 Framework Linking Education to EBS ................................ ................................ ................... 24 Non linear Effects of Education ................................ ................................ ............................. 25 Theories of Differential Effe cts ................................ ................................ .............................. 25 Genetic and Early Life Effects ................................ ................................ ........................ 25 Selection and Personality Differences ................................ ................................ ............. 26 Cultural Capital Differences ................................ ................................ ............................ 27 Stress ................................ ................................ ................................ ................................ 28 2 BACKGROUND TO DIABETES AND IMPAIRED FASTING GLUCOSE ...................... 31 Diabetes ................................ ................................ ................................ ................................ .. 31 Impaired Fasting Glucose (Pre Diabetes) ................................ ................................ ............... 33 Link Between Impaired Fasting Glucose and Di abetes ................................ .......................... 33 A Model and Critical Covariates for Risk of Diabetes ................................ ........................... 35 Age and Sex ................................ ................................ ................................ ..................... 35 Race and Ethnicity ................................ ................................ ................................ ........... 35 Obesity ................................ ................................ ................................ ............................. 36 Marital Status ................................ ................................ ................................ ................... 37 3 DATA AND MET HODOLOGY ................................ ................................ ........................... 42 Objectives and Research Questions ................................ ................................ ........................ 42 Hypotheses ................................ ................................ ................................ .............................. 42

PAGE 6

6 Data ................................ ................................ ................................ ................................ ......... 43 Background ................................ ................................ ................................ ...................... 43 Sample and Selection Procedures ................................ ................................ .................... 43 Variables ................................ ................................ ................................ ................................ 45 Elevated Blood Glucose Level ................................ ................................ ........................ 45 Diagnosed Diabetes ................................ ................................ ................................ ......... 45 Independent Variables ................................ ................................ ................................ ............ 46 Education ................................ ................................ ................................ ......................... 46 Race ................................ ................................ ................................ ................................ 46 Age ................................ ................................ ................................ ................................ .. 46 Marital Status ................................ ................................ ................................ ................... 47 Body Mass Index ................................ ................................ ................................ ............. 47 Methods: Continuation Ratio Model ................................ ................................ ...................... 47 4 RESULTS AND ANALYSIS ................................ ................................ ................................ 49 Data Description ................................ ................................ ................................ ..................... 49 Results for Men ................................ ................................ ................................ ....................... 53 Males, Stage 1: Impaired Fasting Glucose ................................ ................................ ...... 54 Males, Stage 2: Diabetes ................................ ................................ ................................ 55 Stage Differences for Males ................................ ................................ ................................ ... 56 Results for Females ................................ ................................ ................................ ................. 58 Females, Stage 1: Impaired Fasting Glucose ................................ ................................ .. 58 Females, Stage 2: Diabetes ................................ ................................ .............................. 59 Stage Differences for Females ................................ ................................ ................................ 60 Sex Differences ................................ ................................ ................................ ....................... 61 Stage One ................................ ................................ ................................ ......................... 61 Stage Two ................................ ................................ ................................ ........................ 62 5 DISCUSSION AND CONCLUSION ................................ ................................ .................... 78 Discussion ................................ ................................ ................................ ............................... 78 Conclusion ................................ ................................ ................................ .............................. 83 APPENDIX: ADDITIONAL EDUCATIONAL ATTAINMENT ................................ ................ 85 LIST OF REFERENCES ................................ ................................ ................................ ............... 87 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ......... 93

PAGE 7

7 LIST OF TABLES Table page 1 1 Undiagnosed Diabetes By Educational Attainment and Race ................................ .......... 29 2 1 Obesity and Diabetes Prevalence Am ong U.S. Adults, by Sex and Age ........................... 41 4 1 Fr e quencies and Prevalence of I mpaired F asting G lucose for Individuals Aged 20 and Over ................................ ................................ ................................ ............................. 64 4 2 Frequencies and Prevalence (S tandard E rror ) of Diagnosed Diabetes for Individuals Aged 20 and Over ................................ ................................ ........................... 65 4 3 Prevalence (SE) of Diabetes as Measured from Either Diagnosis ................................ .... 66 4 4 Prevalence (SE) of U.S. Population Aged 20 Years and Older with IFG* ..................... 67 4 5 Prevalence (SE) of Diabetes in U.S. Population by Educational Attainment .................... 68 4 6 Proportions and Means of U.S. Born Population(SE) Over the Ag e of 24. ...................... 69 4 7 Results From a Logistic Regression Continuation Ratio Model for Males Greater Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1) .................. 70 4 8 Results from a Logistic Regression Continuation Ratio Model for Males Greater Than 24 Years Old for Having Diabetes Given EBS (Stage 2 ) ................................ ......... 71 4 9 Results From a Logistic Regression Continuation Ratio Model for Females Greater Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1) .................. 72 4 10 Results from a Logistic Regression Continuation Ratio Model for Feales Greater Than 24 Years Old for Having Diabetes Given EBS (Stage 2) ................................ ......... 73 A 1 Perc entage of Educational Attainment Within Five Category. Educational Attainment and By Race/Ethncity in the U.S. Population. ................................ ................ 85 A 2 Prevalence of Impaired Fasting Glucose (SE) of U.S. Born Popul ation ........................... 86

PAGE 8

8 LIST OF FIGURES Figure page 1 2 Conceptual Framework for the Association between Socioeconomic Status and Elevated Blood Sugar, Adapted from Brown (2004) ................................ ......................... 30 2 1 Conceptual Framework for Risk Factors for Diabetes, Adapted from Brown (2002). ..... 39 2 2 U.S. Diabetes Percentage by Age and Sex. Adapted from Centers for Disease Control 2006 ................................ ................................ ................................ ...................... 40 4 1 Es timated Risk of Elevated Blood Sugar by Educational Attainment and Race For Males Over Age 45, Single, and Overweight ................................ ................................ .... 74 4 2 Esti mated Risk of Diabetes Given E levated Blood Sugar by Educational Attainment and Race For Males Over Age 45, Single, and Overweight ................................ .............. 75 4 3 E stimated Risk of E levated Blood Sugar by Edu cational Attainment and Race For Females Over Age 45, Single, and Overweight ................................ ................................ 76 4 4 Esti mated Risk of Diabetes Given E levated Blood Sugar by Educational Attainment and Race For Females Over Age 45, Single, and Overweight ................................ ......... 77

PAGE 9

9 Abstract Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Master of Arts EDUCATION DIFFERENCES IN ELEVATED BLOOD GL UC OSE: DO THEY VARY BY RACE, ETHNICITY AND SEX By Gregory Pavela AUGUST 2009 Chair: John Henretta Major: Sociology The relationship between education and health has been well documented -increasing levels of education are associated with better health us ing various definitions of both education and health. This relationship has also been found to vary across demographic profiles. This study examine s the relationship between education and blood glucose levels, whether the association varies across race and ethnicity and sex, and whether the association varies by level of blood glucose. Elevated levels of blood glucose can be en classified into two stages: impaired fasting glucose (f asting glucose levels 100 to 125 mg / dL (5.6 to 6.9 mmol/L) ) and diabetes (f as ting glucose levels over 125 mg/dL (greater than or equal to 7.0 mmol/L) ), with each stage associated with an array of health risks. This research is important because i ndividuals with impaired fasting g lucose are at increased risk for developing diabetes, and diabetics have an increased risk for developing cardiovascular disease -the leading cause of death in the United States. Data from the combined 1999 2006 National Health and Nutrition Examination Survey are analyzed Logistic regression is used to tes t the relationship between level of blood glucose and whether the association varies across sex and race and ethnicity A continuation ratio model is used to test for difference in effects of covariates between the two stages of blood glucose.

PAGE 10

10 Results ind icate that education has a significant association with blood glucose levels, that this association is significantly different for males and females, that the association varies by race and ethnicity for females, and that the association with race and ethn icity varies by level of blood sugar for both males and females. For males, education has no effect on blood sugar levels after adjusting for age, body mass index, and marital status Having more than a high school degree is significantly associated with reduced levels of blood sugar for females This effect is significantly differen t for men and women Allowing for interactions between education and race and ethnicity suggest that blacks with less than a 9 th grade education have a significantly reduced p robability of developing elevated levels of blood glucose relative to other blacks and being black has the effect of reducing the probability of having elevated blood sugar overall. Hispanics with a BA or more have a significantly increased probability of developing elevated blood glucose relative to other Hispanics but being Hispanic has no significant association with elevated blood sugar levels The association between race and blood glucose varies across levels of blood glucose for both men and women For males, given that one has EBS being either b lack or Hispanic significantly increases the risk of developing diabetes. For females, given that an individual has EBS being black significantly increases the risk of developing diabetes. The race and et hnicity and blood glucose association is significantly different across stages of blood glucose for both males and females, changing from having either no effect or a weak protective effect on development of elevated blood glucose to having a significant e ffect on increasing the risk of developing diabetes given EBS

PAGE 11

11 CHAPTER 1 HEALTH DISPARITIES Introduction Differences in health outcomes by sex, race and socioeconomic status persist in American society today. One important health out come in which disp arities exist is elevated blood sugar (EBS), defined in this paper as having either impaired fasting glucose or diabetes (Borrell et. al. 2006) The significance of diabetes as a health outcome stems from its increasing incidence within the total U.S. popu lation, and its strong association with cardiovascular disease. According to the Centers for Disease Control ., the incidence of diabetes has increased 91% in the past decade. ( CDC 2008 ). The prevalence of diabetes has also been increasing. The CDC. report s that from 1980 to 2006, the crude prevalence of diabetes has increased 132%, and this increase in diabetes is similar regardless of age standardization, indicating that the increasing prevalence of diabetes is not related to the changing age structure of the United States ( CDC 2008 ). The increase in prevalence could also be related to better treatment and longer survival times; therefore incidence is a better measure to indicate that diabetes is of increasing concern. While EBS is a health concern across the entire U.S. population, it affects different segments of the population in different ways. Two key axes of differentiation in the blood glucose disparity are race and ethnicity, and education. For example, Hispanics are twice as likely to die from di abetes as are whites and those with higher levels of education are less likely to have diabetes (Healthy People 2010, Borell 2006). Given the significance of studying diabetes due to the increasing incidence of diabetes in the U.S. population and its diffe rential impact on key segments of the population, the objective of this research is to answer the following three questions: (1) What is the association between education and the development of EBS; (2) does

PAGE 12

12 this association vary by race and ethnicity and sex; and (3) does the association between race and ethnicity, sex, and elevated blood sugar vary depending on the level of blood sugar? In order to answer these questions and develop initial hypotheses, relevant health disparities literature will first be reviewed. The definition of a health disparity and a framework for general health outcomes will be provided. A key axis of health disparities is education, which is a measure of socioeconomic status thus socioeconomic disparities in health and EBS will b e reviewed Following the outline of socioeconomic status disparities in health, socioeconomic status will be integrated into the general framework of health outcomes (Figure 1 1) After the general review of socioeconomic status and health disparities, edu cation will becom e the focus as a measure of socioeconomic status The value of education as an socioeconomic status measure its measurement, its effects on EBS, the difficulties associated with studying the effect s of education on health and theories of differential effects of education on health and EBS will complete Chapter 1 After reviewing health disparities by race and ethnicity and educational attainment and formulation of hypotheses in Chapter 1 a basic background to diabetes and impaired fast ing glucose (IFG) will be provided in Chapter 2, including their definition, incidence and prevalence, and deleterious effects on the U.S. population. The empirical link between diabetes and IFG, as well as the biological difference between the two will be outlined, followed by a discussion and formulation of a model for the risk factors of EBS. Once the critical covariates of EBS are reviewed in Chapter 2 Chapter 3 lists the hypotheses formulated from the literature review of Chapters 1 and 2. Chapter 3 also discusses the data and methodology used to test the hypotheses, including b ackground to the NHANES,

PAGE 13

13 sample analyzed, operationalizations of variables, and method of analysis are reviewed. Chapter 4 summarizes the results and Chapter 5 includes a disc ussion of the results and a conclusion Conceptualizing Health D isparities A health disparity can be defined as differences that occur by gender, race or ethnicity, education or income, disability, geographic location, or sexual orientation (Healthy Peopl e 2010). However, this definition is only one among many definitions of what constitutes a disparity ( Smedley 2003, Whitehead 1991) Different definitions of what constitutes a health disparity often involve differing foci of who or what is experiencing a disparity, such as race and ethnicity or socioeconomic status Most definitions of a disparity imply a reference group from which comparisons are to be made. These comparisons can be made to the majority group, the population mean, or to the healthiest gr oup. (Adler 2008) The difference between a disparity and a difference is not always clear, although a disparity usually implies either something amen able to change or a form of injustice (Adler 2008, Herbert et al. 2008 ) In this thesis I explicitly conce ptualize the experience of racial and ethnic minorities with impaired fasting glucose and diabetes as a disparity and not just a difference It is difficult to differentiate between a disparity and a difference in the case of EBS For example, genetic pre dispositions towards development of diabetes in African Americans explain the higher incidence of hypertension in the African American population than in the general population by mortality conditions aboard slave transportation ships which occasionally approached 20% ( Curtin 1992, Klein 2001). Such high mortality rates make trans Atlantic slave trading a plausible selective mechanism for individuals wi th certain metabolic characteristics that may predispose them to metabolic diseases under other conditions. If this

PAGE 14

14 explanation is valid, s hould the biological differences in diabetes be considered a genetic difference, or a disparity rooted in unjust hist orical circumstances? The cause s of health disparities are numerous and interwoven with each other. Because and s the causes of health disparities can be found in the interplay between these factors and the socio historical context i n which an individual or racial and ethnic group exist s The U.S. is composed o f numerous racial and ethnic groups, thus it is vitally important to examine the extent and sources of racial and ethnic health disparities in the U.S. population, keep ing in mind that the goal of addressing health disparities is to improve the overall health of society. The complex factors to be taken into account when an alyzing the sources of health disparities can be organized using a general framework developed from combining models from the Institute of Medicine and Office of Technology Assessment (Goldberg 2004). This framework has th ree primary dimensions: health bef ore care, health care access, and health care d elivery. Health before c are refers to the variables that can influence health outside of the health care system. It encompasses important factors that can influence health including individual income, educatio nal attainment, environment, personal characteristics, and overall social conditions such as employment opportunities. Health before care is the focus of the analysis in this thesis within the tri partite framework for analyzing the origins of health dispa rities. Health a ccess refers to the ability of an individual to access and receive treatment for their health condition. Language barriers, lack of financial resources, and a mistrust of the health care syste m can all serve as barriers to a ffect health ca re access. Once one has gained access to the

PAGE 15

15 system, Health care d elivery is the third dimension on which health disparities may emerge and include s variability of diagnosis, treatment, and communication Socioeconomic status is a key axis along which heal th behaviors, health care access and delivery can vary. The next section discusses the role of socioeconomic status in EBS health disparities Socioeconomic Status Health Disparities in Elevated Blood Sugar Health disparities in diabetes by socioeconomic s tatus have been well documented using multiple measures of socioeconomic status, and the effects of socioeconomic status seem to vary by sex (Smith 2007, Robbins et al. 2004. Stern et al. 1984). Those in the higher levels of socioeconomic status indicators including income, occupation, and education, tend to have lower levels of diabetes. Along some socioeconomic status indicators, the gradient appears to be growing (Smith 2007). Smith finds evidence that between N ational H ealth and Nutrition Examination S urvey II (NHANES) and NHANES IV, the gradient of diagnosed diabetes between those with less than a high school degree and those with more than a high school degree went from being insignificant to a gradient of diagnosed diabetes of about 4% (9.8% in the l owest category of education vs. 6.0% in the highest category). Sex difference s in the association between socioeconomic status and diagnosed diabetes have been found for occupational status but not for education (Robbins 2004) Robbins finds that higher le vels of income, greater occupational prestige, and higher levels of education were found to be associated with lower probability of having diagnosed diabetes among women, but among men only education and income were associated with lower probability of hav ing diagnosed diabetes, not occupational status. As the socioeconomic status gradient in health appears to be widening for diabetes, the proportion racial and ethnic groups with less than a high school degree is increasing The

PAGE 16

16 increasing proportions of ra cial and ethnic groups at lower levels of socioeconomic status may undermine the success in equalizing the rates of undiagnosed diabetes between racial and ethnic groups since the socioeconomic status gradient in undiagnosed diabetes is relatively constant (Smith 2007). Table 1 1 reports undiagnosed diabetes rates in NHANES 1999 2006, as well as slightly from the results of this analysis, so do Ioannou Framework Linking S ocioeconomic S tatus to E levated Blood Sugar A conceptual framework that links the socioeconomic status and diabetes s hould include both the "proximal" mechanisms such as health behaviors, access, and processes of care as previously discussed as well "distal" measures such as cultural patterns that mediate the relationship between socioeconomic status and EBS and the pro ximal mechanisms that link them. Brown et al. (2004) develop such a model however it is intended to model the relationship between socioeconomic status and health among persons with diabetes. This model, with some adjustment, can also be used to link soci oeconomic status with the development of EBS. Figure 1 1 displays the modified conceptual framework. In the model developed by Brown et al. (2004), socioeconomic status encompasses individual, household, and community characteristics that can shape the proximal mechanisms of health behaviors, access to health care, and t he process of health care. At the individual level education, employment, and occupational prestige are likely to shape the proximal mediators between socioeconomic status and health outc omes. At the household level, income and wealth likely affect the proximal mediators for both adults and children. Finally, average community income, education, and crime rates are a part of socioeconomic status that shape proximal mediators.

PAGE 17

17 Distal media tors and moderators are the effects of the characteristics of the individual, provider, community, and health care system on the primary, proximal mechanisms Distal mediators may include level of acculturation and social support at the individual level, l anguage concordance in patient doctor relations at the provider level, and environmental safety at the community level. Critical covariates to consider in any model with socioeconomic status are age, sex, and race and ethnicity. As Brown (2004) acknowledge s, an underlying assumption of their model is a lack of endogeneity, or that socioeconomic status influences health, rather than health influencing socioeconomic status. As will be discussed in the section on measuring education there are several issues t o consider when modeling health outcomes. Early health events can influence socioeconomic status attainment (Palloni 2006). Issues of endogeneity may also effect health insurance, income, and occupational status. The model in this study does not consider t he effect of early life events on socioeconomic status attainment. To reduce the risk of endogeneity, health insurance and income are excluded as possible explanatory variables between race and ethnicity and EBS, so that the model being estimated is a redu ced form model. While there are many possible measures of socioeconomic status, this analysis will focus on education. The reasons for using education as a socioeconomic status proxy in health research are outlined in the next section. Education as a Soci oeconomic Status Measure and General Health Outcomes Higher levels of education are associated with better health, across multiple indicators of health including including mortality, physical functioning, cardiovascular hea lth, and cognitive functioning ( K itagawa and Hauser 1973; Zimmer et al. 2002; Winkelby et al. 1992) An early study on the relationship using mortality as a measurement of health was done by Kitagawa and Hauser (Feinstein 1993). Using 1960 death certificates linked with census information with census information Kitagawa and Hauser measure d the association between

PAGE 18

18 levels of education and mortality They found a strong inverse relationship up until the age of 65. Over 65, the association between education and health was smaller Recent evi dence suggests that the education gradient has been increasing in recent decades, with much of the gradient driven by gains in life expectancy by the more highly educated in older age groups ( Meara and Cutler 2008). During the past two decades, the life e xpectancy of the highly educated increased by almost 3 years, compared to only half a year for the least educated. Increasing education gradients, with the important exception of a narrowing education gradient among young black men, occurred despite narrow ing gradients across sex and ethnicity (Meara and Cutler 2008). There is also evidence for sex differentials in the education gradient in mortality (Elo and Preston 1995). Elo and Preston (1995) confirm the finding that education differentials tend toward s a maximum at older ages, but that these trends must be differentiated by sex. Females aged 25 64 have experienced a recent narrowing of the education gradient in mortality and females age 65 and over have experienced a static gradient. Men, however, have experienced a broad pattern of widening education differentials in mortality since 1960 (Elo and Preston 1995). Besides mortality, health can also be operationalized as physical functioning and here too, research has shown a positive association with education. Zimmer et al. (2002) examined the physical functioning of the parent, defined as reported difficulty in every day tasks such as sitting, crouching and reaching for objects. They found that older adults who had more than a to those with lo w Higher levels of education are a lso associated with lower risk of cardiovascular di seases Using survey results from the Stanford Five City project, Winkleby (1992) demonstrates that

PAGE 19

19 those with the lowest levels of education tend to have the highest levels of cardiovascular risk factors. For example, among men with less than a high school education, 47% reported being cigarette smokers. Among men who had completed a college education 18% reported being cigarette smokers. Females were found to have a similar gradient between education and c igarette smoking, 41% of dropouts reported smoking, while 14% of college graduates reported smoking. Across all levels of cardiovascular risk factors, men had higher levels of risk than women did. Not every measure of health has as clear cut an association with education as mortality and physical functioning. Due to the nature of education itself, associations with cognitive functioning in late life and level of education may be spurious but i s also possible that associations between education and cognitive and infectious diseases throughout the life course, quality of health care, occupationa l or environmental exposures, or differences in health practices and lifestyle behaviors (Cagney and Lauderdale 2002 ). While using education as a measure of socioeconomic status when modeling cognitive outcomes poses difficulties, using education as a measure of socioeconomic status to model blood glucose outcomes is more appropriate. Measuring Education Aspects of a quantity of education, credential s and selectivity of education received (Ross and Mirowsky 1999). The quantity of education is the number of years of education achieved implying that each year has equal importance In con trast, a focus on credentials argues that the value of an education comes from holding certain degrees. S electivity of education refers to the prestige of the institution attended. The selectivity model could include aspects of both quantity and credential s: the institution offers a higher

PAGE 20

20 more highly valued by society, granting greater access to resources such as occupations and income. The selectivity model is more easily applied to college degrees. Measurement of recognized as prestigious, but there is no national ranking of the secondary schools attended by the vast major ity of the population. Increasing the quantity of education should increase the stock of "human capital" one has, according to human capital theory. School is seen as a place where students learn both specific skills such as mathematics, but also general problem solving skills and the ability to negotiate with others in the pursuit of a goal. Personality traits such as "self directedness" are also encouraged. Specific knowledge, general skills, and personal growth are thus substantive parts of one's educat ion that have a real impact on an individual, and the more exposure to that kind of environment, the greater "human capital" one has. In turn, human capital "ultimately s hapes health and well being" especially through the development of an internal locus o f control (Ross and Mirowsky 1999) Healthy behaviors are more likely to be practiced by somebody if they believe a good part of their fate rests in their own hands. The credential view of education argues that the substance of an education is minimal and that the true value comes from possessing a degree. This approach does not predict a linear association between years of schooling and various health outcomes. Rather, the possession of a degree is the best indicator for good health. Furthermore, the effec ts of the degree would be mediated by occupational status assuming the positive association between good quality of jobs and higher levels of health (Ross and Mirowsky 1999). The final model of education is the selectivity model, which combines as pect s of both quantity and credential views of education. This model predicts that the quality of an

PAGE 21

21 educational institution has health effects beyond the predicted effects of either quantity of education or credentials. These increasing health returns cou ld be due to either self selecting processes (the best students go to the best schools), or that the higher quality of the institution (either substantively or credentially) leads to better job placement, resulting in higher levels of health Research sugg ests that quantity of education has the largest impact on physical functio ning and perceptions of health (Ross and Mirowsky 1999) Ross and Mirowsky also find that the credential model has no significant effect. C ontrolling for a healthy life style, the se lectivity of college attended becomes insignificant, suggesting that the beneficial economic pathways theorized for going to a high quality school are not the actual pathways through which education operates. Instead, selective schools somehow promote more healthful lifestyles or select for healthier individuals than less selective schools, although this intra school difference is relatively unimportant compared to the effects of continued education at any institution of higher education Pathways Between E ducation and Health Several pathways have been offered to explain the positive associati on between education and health (Ross and Wu 1995, Adams 2002). These pathways include work and economic conditions, social and psychological res ources, a healthy lifes tyle, and health care utilization skills ("productive efficiency") (Ross and Wu 1995 Adams 2002). Work and economic effects assume those who have a higher education are more likely to have a job, with higher income and greater oppor tunities for self fulf illment. In 1991, 87% of college graduate students were employed compared to 77% of those who had a high school degree only, and 56% of people with eight years of education or less. Furthermore, those with greater educational attainment stay unemployed for lesser periods of time (Ross and Wu 1995

PAGE 22

22 Moen 1999 ) Those who do not have the skills to enter the workforce face economic hardship from lack of income, which in turn might affect health through the daily stresses of living life on the edge and limited a ccess to health care. Daily stresses may play a role in the in the health advantage more highly educated individuals enjoy, including lower levels of blood sugar (Surwit 1992, Goetsch 1990). Goetsch (1990) found stress to have a hyperglycemic effect in bot h laboratory and natural settings, with the greatest level of blood glucose range occurring on high stress days. Social and psychological resources are a s econd way in which education might improve health. Education develops both a sense of personal contro l and social support network. The highly educated will confront them with both increased knowledge and increased attention to changing unhealthy behaviors. Ironically, those who most need the tools of knowledge and the sense of efficacy to confront health issues are those least likely to have them. Social support systems ar e stronger for the college educated (Ross and Mirowsky 1999), and those with better social support systems have better hea lth outcomes, such as lower rates of mortality. Men with few social connections experience a morality rate 2.3 times higher than those with better social support (House, Landis, and Umberson 1988) A third link between education and health is the tendency for the more highly educated to practice healthier behaviors: they practice greater allocative efficiency educatio moderately (Ross and Wu 1995, Grossman 1997). Higher educational attainment is also associated with regular exercise, lower body weight, and knowledge about blood pressure. Black s and Hispanics

PAGE 23

23 had different risk profiles than whites even after adjusting for educational attainment (Shea et al. 1991). Those with greater levels of education may also have greater levels of productive efficiency. Productive efficiency occurs when a m ore highly educated patient is better able to recognize and express their symptoms to the doctor. In this way, the more highly educated are able to "get more" out of their inputs into health care processes (Grossman 1997). Support has been found for many o f the theorized mechanisms (Ross and Wu 1995 Adams 2002). However, endogenous processes may select healthier individuals for higher levels of educational attainment. Education and Endogeneity Ross and Wu (1995), while developing excellent models of possib le pathways through which education can affect health outcomes, do little to address issues of reverse causation. Both higher levels of education and better health may result from pre existing factors. Issues of endogeneity also arise for allocative and pr association with health. Individuals who are already more efficient accumulators of human capital may already be healthier, or may have personality characteristic that predispose them to defer gratification ( Grossman 1997, Palloni 2006). It could also be the case that parental background affects both health and educational attainment (Elo and Preston 1996). Despite these difficulties in measuring education's effect on health, evidence of a causal relationship between education and health has been found, with especially pronounced effects among women (Adams 2002). A similar problem occurs when studying the effects of health insurance on health. Those with health insurance may have been able to acquire health ins urance because of better levels of health, making health endogenous to the acquisition of health insurance. Acquisition of health

PAGE 24

24 insurance may also indicate individuals who place a higher initial value on health than those who s good health and health insurance may both be a result of a higher valuation of health. One possible method of dealing with endogeneity in health insurance when examining its effect on health is to examine the effect of the acquisition of Medicare on hea lth. (McWilliams et al. 2007). Income is another measure of socioeconomic status that is susceptible to issues of endogeneity. Those with lower incomes may have worse health, but those with worse health likely have lower incomes. Sudden health transition s have a significant effect on reducing income levels, an important causal pathway through which health is endogenous to income (Smith 2005). In contrast to income education is a fairly consistent status within the usual life course ordering of events. E ducational attainment is generally static after an individual has left school, making it better than other indicators such as income and occupation in efforts to avoid endogeneity (McWilliams 2002, Smith 2005). Framework Linking Education to EBS Many of th e pathways through which education affects general health outcomes are likely to affect blood sugar levels. Those with a higher level of education experience lower rates of obesity higher incomes, and great er levels of healthy behavior (Monteiro et al. 20 01, Ross and Wu 1995, Shea 1991). Given obesity's status as a risk factor for diabetes, it is expected that increased education will reduce the probability of diabetes, with BMI as an important mediator. Higher levels of education are also associated with higher incomes, which in turn are associated with lower rates of diabetes (Rabi et al. 2006). Healthy behaviors, including exercise are also associated with lower rates of diabetes, thus as educational attainment increases, it would be expected that rates of diabetes might decrease, with healthy behaviors as moderators ( Burns 2007). Given these clear associations between higher education and factors that reduce the risk

PAGE 25

25 for diabetes, it is expected that higher levels of education will be negatively associa ted with risk of EBS. However, this association may vary across different groups of the population for reasons outlined below. Non linear Effects of Education The effect of education may vary by level of education itself -i.e. its effects are non linear (Cutler and Lleras Muney 2007) That is, there may be a "heterogeneous ef fect for each year of Muney 2007:6) A l inear association between education and health has been found for some measures of health, including mortality, co lorectal screenings, and use of smoke detectors, while for other measures th ere is a non linear association (Cutler and Llerras Muney 2007). Smoking and obesity both tend to have non linear associations with education An increased effect per each year of additional schooling is only seen in those who are more highly educated. After 10 years of schooling however, health levels become linearly assoc iated with education levels (Cutler and Lleras Muney 2007 Theories of Differential Effects The association bet ween education and is stronger among whites and Hispanics than for b lacks. This finding can be interpreted as health ac ross different groups of people (Borrel 2006) There are several reasons why th e association between education and EBS might vary. Genetic and Early Life Effects G enetic pre dispositions may account for race differences in prevale nce of diabetes. If one group i would have a lesser effect. For example, a population selected for a thrifty gene, once no longer needing the thrifty gene for survival, may be at increased risk of developing risk factors for diabetes, such as

PAGE 26

26 higher BMI. The thrifty gene hypothesis for Type 2 Diabetes states that certain populations developed especially efficient metabolic mechanisms that are overwhelmed by modern progress in food supplies and are therefore predisposed to obesity and Type 2 Diabetes (Neele 1968, Joffe and Zimmet 1998). T he concept has found support in the animal world as well as in human populations particularly susceptible to diabetes. leading to cell dysfunction, also known as the Barker hypothesis (Joffe and Zimmet 1998).The Barker hypothesis conceptualizes poor fetal and early post natal nutrition as "imposing mechanisms of nutritional thrift upon the growing individual"(Hales and Barker 1992). In a fetal environment with constrai ned growth, the fetus adopts survival mechanisms which elevate the risk of diabetes in later life. Possible evidence of the Barker hypothesis includes the example of the Naruruan islanders, who experienced increased rates of diabetes when malnutrition was followed by affluence with the introduction of phosphate mining (Hales and Barker 1992). If the Barker hypothesis is true, those populations which tend to experience less healthy fetal environments may develop this syndrome in larger number, which could mo derate the relationship between diabetes and protective factors, such as education. Other early life events may also play a large role in shaping both future socioeconomic status attainment and health outcomes. While some researchers argue that education s tructures the pathways that influence health, it is also likely that early health events help to structure socioeconomic status attainment (Ross and Wu 1995, Palloni 2006). Selection and Personality Differences Personality differences might also account fo r a proportion of health disparities. behavior that define an individual's personal style and influence his or her interaction with the

PAGE 27

27 environment attenuate the relationship between education and all cause mortality in men by 34% in a study of 1989 French cohort (Navi, Kimiaki, Marmot et al. 2008) Not surprisingly, those possessing the battery of pe nti personal same personality differences that account for a significant amount of health disparities might also explain why education might have differing value for people. Tho se personal traits that have been found to promote good health, such as placing high value on self regulation of behavior and autonomy may help those who possess them harder, or staying calm under pressu re filled situations such as test taking If personality traits tend to clust er around racial ethnic groups it is possible that they might account for some of the variation in the strengths of association between education and health. Research on personal ity trait clustering, done in response to the use of personality tests in hiring to determine if certain demographic groups were more likely to be denied employment based on test results, indicates small but significant differences of mean scores on pers onality scales (Ones and Anderson 2002). if certain demographic groups were more likely to be denied employment based on test results Evidence has been found of small, but significant differences of mean scores on personality scales. Differences in scores 2002) Cultural Capital Differences Those who favor the cultural capital explanation for race and ethnic differenc es point to the different treatment of students and parents that may lead to these differences (Lareau and Horvat 1999, Morries 2005, 2007) If the value of education stems from the accumulation of skills taught in the educational setting, different treatm ent may imply differing accumulation of skills,

PAGE 28

28 altering the value of education. Evidence that the quantity of education is what matters, as opposed to selectivity of education or credentials bestowed by education, support the idea that education is a proc ess of skill accumulation C ultural capital explanations for varying effects of education argue that actors occupying the multitude of social locations have, and are perceived to have, certain kinds of cultural capital, and are treated accordingly. Those w ho are perceived to have the right cultural background will be treated more favorably, and will be more likely to accumulate skills taught in school. Advocates of the cultural capital framework point out that one measure of success in higher education is g rades, and grade distributions fall along patterns of social location, with Blacks often underperforming compared to their Hispanic, Asian, and White peers (Massey 2003). The currency of cultural capital that may vary from student to student includes a "w ide variety of resources including such things as verbal facility, general cultural awareness, aesthetic preferences, information about the school system, and educational credentials" (Swartz 1997). Other scholars have found evidence for different educatio nal experiences based on race and ethnicity Black girls are treated differently than their peers based on perceptions of black femininity within the context of hegemonic white femininity (Morries 2007). Stress Others argue that stressors within the social environment are an important source of health disparities between classes of people (Surwit 1992, Williams 1997) Stress in the social environment can occur when the psychological demands of the environment are perceived to be greater than the perception of control over the environment. If certain groups of people experience differing levels of stress in the educational environment or in their current lives, the health value of education might be lessened.

PAGE 29

29 Table 1 1 Undiagnosed Diabetes By Educational Attainment and Race, NHANES 1999 2006 Weighted Data Percentage Overall: 2.6% Reference** 2.8% Education Less Than HS 4.6% HS Degree 3.0% More Than HS 1.8% Race Black 2.6% White 2.6% Hispanic 2.2% Undiagnosed diabetes defined as individuals who reported having not been told by a doctor they have diabetes but who had blood glucose levels >= 7.0 mmol/L ** Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fast ing Glucose in Adults in the U.S. Population Diabetes Care Vol. 29:1263 1268.

PAGE 30

30 Figure 1 1 Conceptual Framework for the Association between Socioeconomic Status and Elevated Blood Sugar, Adapted from Brown (2004)

PAGE 31

31 C HAPTER 2 BACKGR OUND TO DIABETES AND IMPAIRED FASTING GLUCOSE Diabetes Type 2 diabetes (7.0 mmol/l), is associated with numerous health problems, including significantly increased risk of mortality often due to the cardiovascular complications that can result from having the disease. Those with diabetes have higher rates of mortality across all age, racial ethnic, and gender groups (Bertoni et al. 2002). Using ICD 9 Medicare records, Bertoni et. al .(2002) found that elders with diabetes suffer higher mortality rates, with a standardized mortality ratio of 1.41 compared to that of non diabetic elders. Furthermore, national declines in the death rate due to coronary heart disease have not been seen t o the same extent in diabetic population (Gu, Cowei, Harris 1998). Diabetes is also associated with significantly increased functional and wo rk disability, as well as poor mental health. N ational Health and Nutrition Examination Survey data indicate that d iabetes is associated with increased probability of disability in functional activities, including slower walking speed, decreased balance, and falling, and that the increased probability of disability is likely to reduce the quality of life (Gregg et al. 2000). Those with diabetes report more days of poor physical health (such as physical illness or injury) than matched respondents (8.3 days compared to 3.0 days), and report more days of poor mental health (such as stress or depression) than matched respon dents (2.8 days compared to 1.8 days) (Valdmanis et al. 2001 ) The prevalence of diabetes is increasing. From 1990 to 2000, the prevalence of self reported diagnosed diabetes in the U.S. increa sed 49%, from a 4.9 percent to 7.3 percent of the U.S. popula tion (Mokdad et al. 2001) Expanding the time frame to 1980 2000, the number of people with physician diagnosed diabetes in the United States increased more than two fold,

PAGE 32

32 from 5.8 million to 13.3 million (Kouz netsova 2007). Accompanying this increasing prevalence is a dramatically increased incidence rate. In 33 U.S. states that had data for both periods, the incidence rate incre ased 90% -from 4.8 per thousand in 1995 -1997 to 9.1 in 2005 -2007. Increasing prevalence of a disease associated with increa sed mortality, increased functional disability, and many other detrimental health effects has significant policy implications because of its human and financial costs Cost of illness estimates typically report either total cost to society or excess cost o f illness ( Ettaro et al. 2004). Cost of illness estimates for diabetes of both types are substantial, and total cost to society is growing with an estimated total cost to society of diabetes of 100 billion dollars in 1995 (Ettaro 2004). The excess medi cal cost for an individual with di abetes is estimated to be 2 5 t imes greater than for the non diabetic. The differential varies across groups, with those in the younger age groups typically experiencing the largest differential in excess cost. Contributi ng to the excess cost are the higher h ospitalization rates for diabetics which are approximately 2.5 time s higher than for non diabetics (Brown et. al. 1999) The higher rate contributes to the estimated $3.8 billion i n costs for in patient diabetic care from complications in 2001. Diabetes related hospital visits cost the Medica re program $1.3 billion, with estimated preventable costs of $366 million if proper primary care had been provided prior to complications such as diabetes, arising from diabetes ( Economic and Health Cost s of Diabetes: HCUP Highlight 1 2005). To compound the difficulties of higher health care costs, diabetics are more likely to experience unemployment and are more likely to have incomes less than $20,000 than non diabetics. 71% of diabetics have annual incomes less than $20,000, c ompared to 59% of non diabetics M inority status individuals are more likely have multiple hospitalizations than dia betic non minority individuals (Valdmanis 2001)

PAGE 33

33 Impaired Fasting Glucose (Pre Diabetes) Pre diabetes, also known as impaired fasting glucose is a pre diabetic state recently defined in 2004 by the American Diabetes Association as individuals with impaired fasting glucose levels between 100 mg per dl and 125mg per dl. The previous definition had included only those individuals with IFG levels greater than 110 mg per dl. It is estimated that in 2002, about 26% of the adult population has impaired fasting glucose, one of the leading risk factors for developing diabetes mellitus (Cowie 2002). U nlike diabetes, the prevalence of impaired fasting glucose has not dramatically increased. A study comparing prevalence of IFG between 1988 to 1994 and 1999 2002 using data from NHANES, the same data source u sed in this thesis, found a small increase in th e prevalence of IFG, from 26.3% to 26.9% ( Ioannou et al. 2007). Results from a second study, also using NHANES from the same year, found a slightly larger increase from 24.7% to 26.0% (Cowie 2002). While the results of these two studies are similar, they diverge more than would be expected given the same data It is not known why the estimates vary by 1 2%. Recall that rates of diagnosed diabetes have risen significantly, from 5.1% to 6.5% during the same time frame as has been reported for IFG, even as un diagnosed diabetes remained stable. Impaired fasting glucose itself is also associated with increased risk of cardiovascular disease, and those with pre diabetes may also be at increased risk of mortality (Califf et al. 2008, Peterson and McQuire 2005, Sne halatha et al. 2003). Therefore efforts to reduce the incidence of CVD can target both diabetes and pre diabetes. Link Between I mpaired Fasting Glucose and Diabetes A major reason to set reduction of IFG preval ence as an important goal is impaired fasting glucose 's strong association with development of diabetes. Reported rates of progression from IFG to diabetes have varied in the literature, with some research suggesting the percent of

PAGE 34

34 individuals progressing from IFG to diabetes to be as high as 33% over a 6 year period, to as low as 9.1% over a 12 year period, with non white study populations generally reported to have higher percentages of progression from IFG to diabetes. (Nichols et al. 2007) The recent reduction in the IFG threshold, from 110 mg/dl t o 100 mg/dl also has an effect on the expected rate of change One U.S. based study reported that, over a 9 year period, 24.3% of individuals who satisfied the old definition of IFG progressed to diabetes, whereas under the new definition of impaired fast ing glucose only 8.1% of individuals progressed from IFG to diabetes (Nichols et al. 2007 ) Thus progression from IFG to diabetes is expected to occur at a slower rate under the new definition. Other determinants of progression from IFG to diabetes include higher levels of blood glucose known hypertension, a nd high levels of triglycerides (Rasmussen et al. 2008) The empirical link between IFG and diabetes and the measurement of blood glucose along a continuous scale for both states of glycemia is the maj or reason for analyzing the data using a continuation ratio logistic regression model (CRM) This model, as explained in greater detail in Chapter 4, allows the researcher to tes t for significant differences in effects of covariates between stages of model s. Some covariates may be more important for increasing risk of IFG than diabetes, or some covariates may not be significant until one already has IFG. While fasting plasma glucose levels follow a continuous spectrum that allow for analysis via the CRM, i t is important to recognize that the two conditions are separate. Diabetes is characterized by the failure of beta cells to produce the required amount of insulin in the context of insulin resistance, whereas IFG is characterized by the ability of the beta cells to accommodate the required increase in insulin production in the context of insulin resistance (Shabha 2004).

PAGE 35

35 A Model and Critical Covariates for Risk of Diabetes This study focuses on both IFG and diabetes and examines whether some of the primary risk factors associated with IFG and diabetes are different The CRM model, for accurate comparisons, requires the inclusion of all covariates at each stage of the outcome level. The risk factors for diabetes will determine what covariates to include in t he final CRM model, acknowledging that the risk factors for IFG may be different. Figure 2 1 outlines a model of risk factors for the development of diabetes. P rimary risk factors for the development of diabetes include a family history of the disease and demographics, behavioral psychological and clinical factors (Brown 2002). Age and Sex Figure 2 2 constructed using data from the N ational H ealth I nterview S urvey and provided by the C enters for D isease C ontrol illustrates the variation in diabetes preva lence between age groups. Diabetes prevalence increases with age for all races and sexes up until 75+ years, at which point prevalence declines. Variation between age groups is much greater than it is for sex groups, across all racial categories, and is th us important to control for it. possible non linear association with risk of EBS, age is coded into 3 dummy variables: 24 35 years, 36 45 years, and 46+ years. Controlling for age also helps to account for the race and education di fferences that may occur between age groups O verall prevalence of diabetes in most populations is equivalent across the sexes. Race and Ethnicity Blacks and Hispanics are both mor e likely to have diabetes than w hites (Brancati et. al. 1996) The increas ing proportion of the Hispanic population in the U.S. population and their elevated risk of mortality compared to w hites makes it is especially important to address these health disparities in national efforts at reducing incidence and prevalence of diabet es. There are

PAGE 36

36 significant differences in prevalence of both pre diabetes and diabetes between racial ethnic groups over the age of 20, the ages analyzed here Data from the National Institute of Health demonstrate the disparity of diabetes betwe en racial ethnic groups: Among n on Hispanic whites, 8.7% aged 20 and over have diabetes. Among Non Hispanic blacks however, 13.3% aged 20 and over have diabetes. Among the Hispanic population, 9.5% aged 20 and over have diabetes. While Asian Americans aren't include d in the final analysis of this thesis due to limited sample size, they are 1.5 times as likely to have diagnosed diabetes as non Hispanic whites (National Institute of Diabetes and Digestive and Kidney Diseases 2008) Obesity Obesity is a primary risk fac tor in the development of diabetes (Shai, Im and Jiang, R. et al. 2006 ) Using data from the 2000 Behavioral Risk Factor Surveillance System, Mokdad et al. derived estimates of diabetes prevalence between the obese and non obese across numerous characteris tics. In each case, the obese group had a higher prevalence of diabetes than did the non obese group. Prevalence of obesity has been increasing, and some have called both obesity Table 2 1 reports the percentage of obese persons by sex and by age groups. Obesity is also strongly related to educational attainment, with the least educated having the gre atest likelihood of being obese (Borrell 2006) Since obesity is related to both education and diabetes, it may operate as a mediating variable in t he relationship between them. Despite has been found to explain some, but not all of the racial and ethnic differences in diabetes (Brancati 1996). With regard to impaired fasting glucose B.M.I. has been found to predictive for both blacks and whites. (Klein et. al. 2004) Using data from the National Heart, Lung and Blood Institute, Klein (2004) found that baseline B.M.I was predictive of impaired fasting glucose for

PAGE 37

37 black girls and rate of B.M .I. increase was predictive of i mpaired fasting g lucose for white girls. Whether the effect of B.M.I. remain constant through the progression of EBS remains unknown, and therefore wi ll be examined in this analysis Marital Status Marital status is also associated with diabetes C hoi and Shi (2001) found that "[w] omen who were single and 35 to 64 years old had a higher prevalence of diabetes than women of the same age who were married" Marital status is also associated with better health outcomes across other indicators of health (Shoenborn 2004) Using N HIS data pooled from the years 1999 origi n, education, income, or nativity) or health indic a tor (fair or poor health, limitations in activities, low back pain, headaches, serious psychological distress, smoking, or leisure time physical inactivity), married adults were generally found to be healt hier than adults in other marital status categories A notable exception in their study was obesity -married men were more likely to be obese than those in other martial groups. One possible explanation for all of these findings is that marriage selects for better health i.e. healthier individuals are more likely to get married. The relationship between education and marital status differs between men and women. For men, each increasing year of education is associated with an increasing likelihood o f marriage, and this association has grown stronger from 1980 2000 (Rose 2003). For women, there exists a "marriage penalty" after 12 16 years of education. Initially, women and men can expect a positive association between education and likelihood of marr iage. However, for women, after 12 16 years of education the association becomes negative This "marriage penalty" for increasing levels of education lessened from 1980 2000. Thus the strength of education's effect on likelihood of marriage seems to be gro wing for men and women alike.

PAGE 38

38 Since the more highly educated (up to 12 16 years for women) are more likely to be married, and marital status is associated with diabetes, it is important to control as another possible mediating variable in the association b etween education and diabetes.

PAGE 39

39 Figure 2 1 Conceptual Framework for Risk Factors for Diabetes, Adapted from Brown (2002).

PAGE 40

40 Figure 2 2 U.S. Diabetes Percentage by Age and Sex. Adapted from Centers for Disease Control 2006 White Male White Female BlackMale Black Female Hispanic Male Hispanic Female 0 44 1 1 2 3 2 2 45 64 10 8 17 16 13 16 65 74 20 16 27 28 30 27 75+ 17 13 22 29 23 22 0 5 10 15 20 25 30 35 Rate per 100 U.S. Diabetes Rate by Age and Sex

PAGE 41

41 Table 2 1 Obesity and Diabetes Prevalence Among U.S. Adults, by Sex and Age BRFFSS, 2000* Obesity Diabetes %(SE) %(SE) Total 19.8(.17) 7.3(.12) Sex Male 20.2(.26) 6.5(.18) Female 19.4(.21) 8.2(.15) Age 18 29 13.5(.33) 1.9(.13) 30 39 20.2(.36) 3.8(.18) 40 49 22.9(.41) 5.8(.27) 50 59 25.6(.47) 10.9(.37) 60 69 22.9(.50) 14.5(.44) >=70 15.5(.41) 14.9(.42) *Adapted from Mokdad 2008.

PAGE 42

42 C HAPTER 3 DATA AND METHODOLOGY Objectives and Research Questions The objective of this thesi s is to answer the following three questions: (1) What is the association between educa tion and the development of EBS; (2) does this association vary by race and ethnicity and sex; and (3) does the association between race and ethnicity and elevated blood sugar vary depen ding on the level of blood glucose ? Hypotheses S even hypotheses follow from the research questions and past research. First, it is hypothesized that there is a negative correlation between level of education and probability of having EBS: as level of education increases, the probability of having EBS decreases (H1). This hypothesis is based on the research that has consistently found an association between higher levels of education and better health. Second, it is hypothesized that the a ssociation between education and EBS varies by race and ethnicity. The effect of education on probability of having EBS will be weaker among Blacks (H2). This hypothesis is based on the research that suggests that predisposes blacks to EBS and that blacks may not possess the same cultural capital as whites, limiting the effect of education It is hypothesized that education has similar effects between Hispanics and whites (H3). This hypothesis is based on the previous research that found similar effects of education for whites and Hispanics but not for blacks. It is hypothesized that t he effect of education will not differ by sex (H4). This hypothesis is based on research that suggests the effect of socioeconomic status varies b y sex for occupational status, but not for level of education. Finally, it is hypothesized that th e effect of race and ethnicity and education varies by level of blood glucose This hypothesis is based on the fact that diabetes and IFG are two different, a lthough closely linked, disease states. Past research and descriptive

PAGE 43

43 statistics o f the sample analyzed suggest that being black will be associated with lower rates of IFG but higher rates of diabetes (H5). It is expected that being Hispanic will not have a different association with either IFG or diabetes (H6). It is expected that education will have a stronger association with EBS than with diabetes (H7). Data Background The NHANES (National Health and Nutrition Examination Survey has been the primary t ool of the National Center for Health Statistic s of collecting information on the health of the U.S. popula tion since 1960. NHANES data have been released in two year cycles since 1999 to allow the survey to adjust more quickly to the needs of researchers studying a diversity of health issues within the US population. The NHANES is especially useful to researchers because the survey collects laboratory measures as well as self reported information on health statuses and other relevant dimensions. Pooled dat a from the 1999 2006 NHANES are analyzed in this study. Sample and Selection Procedures The target population of the survey is the civi lian, non institutionalized US population, with low income and minority oversamples The survey population is selected thr ough a stratified Primary Sampling Units (PSUs), which are counties or small groups of contiguous counties; 2) segments within PSUs (a block or group of bloc ks containing a cluster of households); 3) households within segments; and 4) one or more participants within each household. A total of d during a 12 month time period (National Center for Health Statistics 2009). Once a household has been selected, it is notified via mail and a NHANES representative makes direct contact with the household to determine if it contains eligible participants. All eligible interviewees are al so asked to participate in the medical examination c omponent of

PAGE 44

44 N HANES. Each respondent is randomly assigned to either a morning or afternoon examination session. The morning participants were tested for their fasting levels of various blood chemicals such as glucose. Because the survey sample population varied at each level of the NHANES (questionnaire, medical e xamination c omponent, and fasting subsample), it is important to use the proper weighting variables when doing analyses NHANES is based on a clustered sampling design and respondents have different probabilitie s of selection. Failing to adjust for the sampling design through the use of weights may bias estimates and overstate significance levels (National Center for Health Statistics 2009). Blood sugar levels in the NHANES were assessed using fasting plasma glu cose, two hour glucose tolerance test, and serum insulin in particip ants aged 12 years and older. The present analysis uses the fasting plasma glucose measure This measurement comes from the smallest sub sample group in the NHANES, and following NHANES d ocumentation, the weighting for the smallest subsample is used Thus the fasting subsample weight variable is appropriate for the analysis. Since data were pooled across four survey cycles, the weight variable had to be adjusted to account for the differin g survey years. The NHANES provides a 4 year sample weight to be used with analysis for both 1999 2000 and 2001 2002 and provides a 2 year weight specific to either 2003 2004 or 2005 2006. Combined survey cycles should be representative of the population at the midpoint of the combined survey period and the sum of rescaled weights should match the survey population at the midpoint of each period. The 4 and 2 year weights used in this analysis were made directly comparable by assigning half the 4 year samp le weight for respondents in 1999 2002, and of the sa mple weight for respondents 2004 2006 (National Center for Health Statistics 2009).

PAGE 45

45 The target sa mple for analysis consists of US born persons aged 2 4 and over although descriptive tables 4 1 through 4 5 report the prevalence of IFG and diabetes in the U.S. population over the age of 20 in order to make a more precise comparison with previously published statistics. Limiting the sample to those born in the US in the regression analysis helps to pre vent measuring education received outside the US, and controls for the better health of the immigrant population also known as the effect (Kennedy et al. 2006). Variables Elevated Blood Glucose Level For the regression analysis the de pendent variable elevated blood sugar is defined as having fasting blood glucose levels greater than or equal to 100 mg/DL reporting by the respondent that he or she has been told by a doctor the individual has diabetes, or currently taking medication to reduce blood sugar levels This cut off point was chosen because it is the new standard definition of having impaired fasting glucose. Elevated blood glucose is coded dichotomously, with 1= having EBS and 0 as not having EBS. Diagnosed Diabetes While the regression analysis uses the broadest definition of diabetes, several t ables report descriptive statistics using more narrow definitions, such as diagnosed only or undiagnosed diabetes. Tables that report diagnosed diabetes define diabetes as a self report diagnosis of diabetes, without making use of the blood glucose analysis. Tables that report undiagnosed diabetes report individuals as undiagnosed if they report blood sugar levels greater than 125 mg/dL but do not report having been told by a doctor they have diabetes.

PAGE 46

46 Independent Variables Education The education variable used in this analysis was measured as an ordinal variable by handing a respondent a card, reading the categories to them as necessary, and recording the ce. The respondent could choose from first through 12 th grade, high school graduate, GED or equivalent, some college, AA degree, technical degree, BA, MA, or PhD. f ive categories: less than 9th grade, less than high school, high school degree, s ome college, and BA or g reater. Although data for blood glucose and education is available for all respondents aged 12 and over, analysis is limited to those age 24 and over because this is the standard minimum age at which someone is likely to earn a degree beyond that of the BA. Race The race variable used in this analysis was constructed from two variables that are respondent self reports on questions of race and ethnicity From these questions, the main race variable was constructed containing the following racial ethnic categories: Mexican American, Other Hispanic, Non Hispanic White, Non Hispanic Black, and Other. These categories are collapsed for the final analysis into Hispanic, Non Hispanic White, and Non Hispanic Blacks. Age Age was a response range (for the final sample used in the analysis) between 24 and 85 years of age. Anything greater than 85 was put into the 85+ category. (N=599). Age is coded into three categories: age 24 35, age 36 45, and age 46 or older.

PAGE 47

47 Marital Status Marital status was c omputed from both the question asking the respondent their marital status, and imputing data from r esponses that made reference to marital status. Marital status is coded as a dummy variable in the analysis, with 1 being married, 0 if not married BMI BMI was c alculated using measurements from the medical examination portion of the NHANES survey This variable was coded as overweight for any individual over 24.9 kg divided by meters of height squared and normal for any individual less than 25 kg divided by meters of height squared Methods: Continuation Ratio Model The continuation ratio model is an a ppropriate model when the ordered categories follow a progression of stage s. Stages of elevated blood glucose are an appropriate outcome to m odel one must p roceed through one stage of blood glucose before entering into another. This study uses the continuation ratio model because the objectives require an analysis of the effects of selected covariates across the different stages of blood glucose. Pooled data from the 1999 2006 NHANES are analyzed using binomial logistic regression to determine racial and ethnic levels. For this analysis, there exist three different stages of blood sugar to be c ompared: normal, elevated, and diabetic stages of blood sugar. The first comparison will be made between those with normal blood sugar and those at the first "cut off" point of blood glucose levels of 100 mg/dL or more. The second comparison o will be between those who have diabetes versus those who do not among the population of individuals with EBS Finally, interaction terms of the form stage*'x', where x is an independent variable, can be used to test for

PAGE 48

48 differences of effects o f the variables included in the model across the different stages of the outcome variable. A significant interaction term indicates that the effect of the particular variable differs depending on the stage of the outcome variable. Models for males and fem ales are estimated separately because an overall Wald Test indicated that at least one coefficient was significantly different for mal es and females (F=2.62, p<.05). Thus our fourth hypothesis, that of no interaction between sex and education in the associ ation with EBS is not supported Significant sex differences will be reported in Tables 4 7 and 4 8. The first model will use EBS (having either impaired fasting glucose or diabetes ) as its outcome measure. The second regression model w ill compare those w ith diabetes defined as having a blood glucose level greater than or equal to 126 mg/dL taking medication to control blood sugar, or having been told by a doctor the individual has diabetes, with those who have EBS but do not have diabetes

PAGE 49

49 CHAPTER 4 RESULTS AND ANALYSIS Data Description Table 4 1 summarizes weighte d and un weighted estimates of i mpaired fasting glucose percentages for the U.S. population over the age of 20 across sex and race and ethnicity. While later analyses use National Health a nd Nutrition Examination data for ages 24 and older, this table uses data for ages 20 and over to facilitate comparison with previous studies This table includes a column for a "reference" parameter, derived fro m another article that used data from the 19 99 2002 NHANES for estimates of impaired fasting glucose and diabetes in the U.S. population (Cowie 2002). estimates generally differ from the present analysis by about 1%. These differences likely are the result of different weighting procedures u sed by Cowie (2002) to derive estimates from a combined sample. Data for each two year release cycle of the NHANES is presented, as well as a weighted average constructed using all eight years of data. The weighted population average of IFG across all eig ht years of data is 26.6%. analysis replica te the finding that rates of impaired fasting glucose have not risen dramatically, changing from 25.3% for 1999 2002 to 27.9% in 2003 2006, although a 2.5% increase in the U.S population suggests that well over half a million people more in 2006 have impaired fasting glucose than in 2002. Overall, men are consistently found to have higher percentages of IFG than women. The eight year average of percentage of men with IFG is 34.5%, versus 21.8% for women. Men across all three race/ethnicities also have a higher percentage of IFG than women. The eight year average percentages of white, black, and Mexican American men are 34.0%, 22.0%, and 35.3%, respectively. The eight year average percenta ges of women with IFG are 21.2%, 18.9 %, and 19.0%, respectively. With respect to race, whites and Hispanics consistently have a higher

PAGE 50

50 percentage of their populations with IFG than blacks. Whites have an eight year average of 27.3% and Hispanics have a si milar 27.7% average. The eight year average for blacks is 20.3%. This table suggests tha t there are racial and ethnic differences as well as sex differences in the U .S. population in prevalence of i mpaired fasting glucose. IFG puts the individual at grea ter risk for developing diabetes. Do these apparent differences between sex and racial and ethnic groups persist into the next stage of elevated blood sugar, diabetes? Table 4 2 presents weighted and unweighted estimates of percentages of individuals with diagnosed diabetes. Data for each two year release cycle of the NHANES is presented, as well as a weighted average constructed using all eight years of data. The weighted population average of diagnosed diabetes is 7.1%. Overall, men and women have simil ar levels of diagnosed diabetes. The eight year average of percentage of men with diagnosed diabetes is 7.1%, versus 7.2% for women. B lacks and Hispanics consistently have a higher percentage of diagnosed diabetes than Whites. Whites have an eight year av erage of 6.3%, whereas blacks have the highest eight year average, 10.9%, and Hispanics have an eight year average of 7.3%. Within racial and ethnic categories, only white men have a higher percentage of diagnosed diabetes than woman, 6.5% vs. 6.1%. For blacks and Hispanics women appear to have a higher percentage of diagnosed diabetes than their male counterparts: 9.9% vs. 11.7% among blacks and 6.2% vs. 8.4% among Hispanics The apparent difference in diagnosed diabetes b y race and ethnicity reverses the trend seen in Table 4 1, where w hites consistently had a higher percentage of individuals with IFG However as previously discussed racial and ethnic differences of undiagnosed diabetes appe ar minimal. (Please see Table 1 1 ) Thus it is not surprisi ng that in Table 4 3, which presents the percentage of persons with diabetes, including diagnosed, undiagnosed, and those taking medication, the patterns seen in Table 4 2 persist.

PAGE 51

51 Data for each two yea r release cycle of the NHANES are presented in Table 4 3 as well as a weighted average constructed using all eight years of data. The weighted population average of diabetes is 9.9%. (versus 7.1% with diagnosed diabetes). Overall, men and women have similar levels of diabetes, although men have a slightly h igher percentage of individuals with diabetes than women with the expanded definition of diabetes. B lacks and Hispanics consistently have a higher percentage of diagnosed diabetes than Whites. Whites have an eight year average of 6.3%, whereas blacks have the highest eight year average, 10.9%, and Hispanics have an eight year average of 7.3%. Within racial and ethnic categories, only white men have a higher percentage of diagnosed diabetes than woman, 10.3% vs. 7.7 %. For blacks and Hispanics women appear to have a higher percentage of diabetes than their male counterparts: 9.9% vs. 11.7% among blacks and 6.2% vs. 8.4% among Hispanics Thus there seem to be two differences between leve ls of elevated blood sugar (IF G and diabetes): both Table 4 2 and Table 4 3 reverse th e trend seen in Table 1, where w hites consistently had a higher per centage of individuals with IF G.; and there are only small sex differences for diabetes, unlike the sex differences found at the IFG level of blood sugar. Tables 4 1 through 4 3 show some evidence for different effects of race and ethnicity and sex across different levels of blood sugar. Males, especially white males, seem to have a higher percentage of individuals with IFG h owever this sex effect is not as pronounced at the diabetic stage. Race effects also appear to differ: blacks have the lowest 8 year average per centage of individuals with IF G (20.3%) but the highest 8 year average percentage of individuals with diabetes (13.2%). Tables 4 4 and 4 5 present descriptive s tatistics for another independent variable of interest, education, and the weighted percentages of individuals with both IFG and diabetes

PAGE 52

52 within 3 categories of educational attainment: less than high school, h igh school degree, and college or m ore. Estimat ed percentages of both IFG and diabetes tend to decrease as level of education increases. Overall, among those in the lowest level of education, 31% and 15.2% are estimated to have IFG and diabetes, respectively. These estimates decrease to 24.3% and 7.7% respectively, for those individuals in the highest category. At the level of IFG there are some apparent race differences: the estimated percentage of whites decreases with each increasing category of education; moving from 34.8% among those with les s than a high school education to 29.8% among those with a high school degree, to 24.6% among those with some college education or more. Among Blacks, for both men and women, there appears to be little effect of increasing education on the estimated perce ntages of individuals with IFG For Hispanics the highest estimated percentage of individuals with IFG is in the lowest category (30.6%), and the lowest estimated percentage is for individuals with a high school degree (19.5%). At the level of diabetes there are again some apparent race differences: the estimated percentage of whites decreases with each increasing category of education; moving from 14.6% among those with less than a high school education, 9.9% among those with a high school degree, and 7 .3% among those with some col lege education or more. Among b lacks there also appears to be a trend of decreasing percentages of individuals with diabetes, moving from 17.5%, to 14.2%, to 9.9 % For Hispanics the highest estimated percentage of individual s with diabetes is in the lowest category of education (14.1%), and the lowest estimated percentage is for individuals with a high school degree (8.2%). For Hispanic women, the downward trend continues on into the highest category of education, with 6.7% o f Hispanic women estimated to have diabetes. However for Hispanic men, having some college education or more seems to have

PAGE 53

53 little effect on percentage of individuals with diabetes compared to Hispanic men with a high school degree. Tables 4 1 through 4 5 tables have provided evidence for differing effects of sex, race and ethnicity, and education in their respective associations with levels of blood sugar. The next section presents formal tests to see if these apparent trends have statistical significance Table 4 6 is the final descriptive statistics table, and includes the means and proportions of variables used in the continuation ratio model analysis. For percentages of educational attainment across five levels of education for the entire NHANES sample rather than just the US born segment see Table 1, Appendix A. Results for Men Table 4 7 displays the logistic regression results for males. These regression results represent the effects of the covariates across two different stages of blood sugar. The regression coefficients of the covariates reported in Stage 1 describe the effects of their respective variables on the log odds of having either impaired fasting glucose or diabetes compared to normal levels The regression coefficients of the covariate s reported in Stage 2 describe the effects of their respective variables on the log odds of having diabetes, given that the individual has elevated blood sugar Coefficients with a significance level of less than .05 are considered significant. Four differ ent models were tested: (1) a model with just race and age, (2) a model that additionally adjusts for educational attainment, (3) a model which additionally adju sts for marital status and BMI and finally (4) a model wh ich allows interaction between race a nd ethnicity and education. The column labeled as "Stage Diff" reports the results of significant stage interactions for the covariates. This interaction signals that the effect of the covariate is significantly different from one level of blood sugar to t he next. For males, stage differences were tested using equation three because of the lack of significant education and race and ethnicity interactions in equation

PAGE 54

54 four. For females, stage differences were tested using equation four, which included the int eraction between education and race and ethnicity. Although the education and race and ethnicity interaction effects are insignificant for males they are included for the sake of model comparisons with females. The same four models were estimated for males and females at the two stages of elevated blood sugar: impaired fasting glucose and diabetes. The column labeled as "Sex Diff" report the results of significant sex interactions for each of the two stages of blood sugar. Significant sex differences corre spond to equation four, which allowed for education and race and ethnicity interactions. Males, Stage 1 : Impaired Fasting Glucose Model 1 examines the main effects of race controlling for age Being black has a significant effect on reducing the log odds of developing EBS relative to whites (b= 0.28, p<.05). Being in either of the two youngest age categories is also significantly associated with reduced probability of developing EBS. Model 2 additionally adjusts for level of education. Having a BA or more is the only level of education found to be statistically different in its effect on the log odds relative to having a high school education. Having a BA or more reduces the log odds of havin g either IFG or diabetes by 0.33 Being black continues to have a significant association with reduced log odds of developing EBS. Model 3 additionally adjust s for marital status and BMI. Being married has no significant association with EBS relative to non mar ried individuals, however having a BMI of 26 or over is ass ociated with a significant increase in t he log odds of developing EBS (b= 0.94 p<.01) Being in either of the younger age groups significantly reduces the probability of having EBS (b= 1.53, p<.01 for ages 24 35 and b= 0.86, p<.01 for ages 36 45). There are no significant effect s of race, education, or marital status in the full main effects model for males. Thus for

PAGE 55

55 males, our first hypothesis of a direct association between education and probability of having EBS is not supported. Model 4, the last mode l, allows for an interaction effect between race and ethnicity and education. These interaction terms are insignificant, indicating that our second hypothesis of an interaction between being black and education al attainment that reduces the effect of educa tion on probability of having EBS is not supported Our third hypothe sis that with EBS is similar for Hispanics and whites is supported by the lack of interaction. The interaction terms, although insignificant, are included in the f inal model for proper comparisons of males with the final model for females. Males, Stage 2 : Diabetes Four models were again tested for stage 2. The regression coefficients of the covariates reported in stage 2 describe the effects of their respective vari ables on the log odds of having diabetes, given that the individual already has EBS Model 1 tests just the effects of race and age. Being black is found to have a significant association with increased lo g odds of having diabetes (b=.56 p<.01). This is a reversal blacks association with reduced probability of having elevated blood sugar However statistical tests of stage differences are not conducted until equation 3, the full model Being Hispanic is also found to have a significant association with an increased probability o f developing diabetes, given EBS (b=.70, p<.01) Being in the youngest age category has a significant association with reduced probability of developing dia betes given one has EBS (b= 1.13 p<.01). Model 2 additionally adjusts fo r level of education. Unlike in Stage 1, no level of education is found to be statistically different in its association with the log odds of developing diabetes relative to having a high school education, nor are any of the effects of education found to b e significantly different depending on stage of blood sugar or race and ethnicity After

PAGE 56

56 adjusting for education, being in the youngest category relative to the oldest category continues to be associated with reduced probability of having diabetes giv en on e has EBS (b= 1.2 p<.01). Model 3 the full main effects model for males additionally adjusts for marital status and B.M.I. Being married has no significant association with probability of having diabetes given IFG Having a BMI in the "overweight" or "obese" categories has a significant association with increased log odds of developing diabetes relative to thos e with a "normal" B.M.I. (b= 0.52 p<.05). Being black or Hispanic continues to have a significant association with increased probability of deve loping diabetes given EBS As in stage 1, m odel 4, the final model, allows for an interaction between race and ethnicity and education. These interaction terms continue to be insignificant but are included in the final model for proper comparisons of males with the final model for females. Stage Differences for Males Statistical tests for difference of effect of covariates within the statistical model for males were conducted using equation 3, the full main effects model. Three variables were found to vary significantly by stage for males: being black, being Hispanic, and BMI. The lack of an interaction between education and stage of blood glucose does not support our seventh diabet es, given IFG In stage 1, model 3, being black had a significant association with reduced probability of developing EBS. In stage 2, model 3, being black is found to have a significant association with increased probability of developing diabetes, given EBS This finding supports the fifth hypothesis that the effect of being black will vary by level of blood glucose. The proportional odds model is limited in that it forces the effect of the independent variable to have the same effect for every level of t he outcome variable. The ability of the continuation ratio model to both

PAGE 57

57 estimate and test differences across stages indicates its superiority to the proportional odds model in this particular analysis. The continuation ratio model provides more flexibilit y in the effect of the independent variable and allows a third statistical inference about the effect of being black : that these two coefficients are significantly different from each other across the stages of blood glucose The results of the stage inter action tests are evidence that the effect of being black has a significantly different effect depending on the level of blood glucose. B eing black reduces the probability of developing EBS. However, being black has a signific antly different effect given EB S on increasing the probability of diabetes. A second variable found to have significantly different effects across stages of blood glucose is being Hispanic. In stage 1, equation 3, being Hispanic has no significant effect on estimated probability of EBS. In stage 2, equation 3, being Hispanic has a significant effect on increasing the probability of developing diabetes. Thus our sixth hypothesis of no interaction between being Hispanic and level of blood glucose is rejected. The lack of statistical differ ence in stage 1 highlights the difference between a CRM stage test of difference and a standard inference about regression coefficients: CRM tests for difference of effect by level of the dependent variable whereas standard inference tests difference from zero. Thus even though both coefficient s may be insignificantly different f r o m zero, the two coefficients may be different from each other. The third variable found to have a significantly different effect by level of blood glucose is BMI. In stage 1, mod el 3, being overweight has a significant effect on increasing the pr obability of developing EBS (b= 0.94, p<.01) In stage 2, model 3, being overweight has a significant effect on increasing the probability of diabetes given EBS BMI thus appears to have a significantly smaller effect on developing diabetes given EBS then on developing EBS for males.

PAGE 58

58 Results for Females Table 8 presents the regression results for females across the two different stages of blood sugar. The regression coefficients of the co variates reported in Stage 1 describe the effects of their respective variables on the log odds of having either impaired fasting glucose or diabetes relative to the population that does not have impaired fasting g lucose or diabetes. The regression coeffi cients of the covariates reported in Stage 2 describe their effects on the log odds of having diabetes, given that the individual has impaired fasting g lucose. Four models were tested: (1) a model with just race and age, (2) a model that additionally adjus ts for educational attainment (3) a model which additionally adjusts for marital status and B.M.I. and finally (4 ) a model which allows for an interaction between race and ethnicity and education. Females, Stage 1 : Impaired Fasting Glucose In model 1 with just race and age, being either black or Hispanic has a significant association with reduced log odds of having either IFG or diabetes relative to whites. Like males, being in either non referent age group, Age 20 35 and Age 35 45, is found to have a sign ificant association with reduced log odds of having either IFG or diabetes relati ve to being over the age of 45. Model 2 additionally adjusts for level of education Each level of education for females in stage 1 was found to have a significantly different effect on log odds of developing EBS relative to having a high school degree. Having le ss than a high school degree has a significant association with increased log odds, while having more than a high school degree had a significant effect on decreased lo g odds. Model 3 additionally adjusts for marital status and B.M.I. Being married has no significant association with EBS probability however being overweight or obes e has a significant association with increased log odds of developing EBS relative to t hos e with a overweight

PAGE 59

59 B.M.I. (b= 1 .33 p<.01). All levels of education continue to be associated with probability of developing diabetes. Model 4 allows for an interaction between race and ethnicity and education. The two younger categories continue to h ave a significant association with reduced log odds of having elevated blood sugar Having less than a high school education is associated with a higher probability of having EBS and having more than a high school education is associated with a lower proba bility of having EBS, supporting the first hypothesis. An adjusted Wald test was performed to see if the slopes for educational attainment vary by race and ethnicity The omnibus test provides strong evidence that there is an interaction effect (F=3.1 4, p< .01). The interaction between being black and having less than a 9 th grade education lends support to the second hypothesis that being black lessens the association between education and probability of having EBS. An interaction for Hispanics having a B A or more is significant (b=1.6 p<.01). Thus there is evidence that being Hispanic is associated with a reduced assoication of having a BA or more on the log odds of having EBS This does not support the third hypothesis of no interaction between education and Hispanics. Females, Stage 2 : Diabetes The same four models were tested for stage 2. Again, the regression coefficients of the covariates reported in stage 2 describe the statistical effects of their respective variables on the log odds of having diabe tes, given t hat the individual already has EBS Model 1 tests just the effects of race and age. Unlike in stage 1, being black is found to have a significant association with increased log odds of having diabetes relative to whites, given one has EBS (b=. 80, p<.01).. Being Hispanic does not have a significantly different effect on the progression from IFG to diabetes. Unlike that of having either IFG or diabetes, age is not associated with progression from IFG to diabetes

PAGE 60

60 Model 2 additionally adjusts f or level of educatio n. No level of education is found to be statistically different in its effect on the log odds relative to having a high school, nor are any of the effects of education found to be significantly different depending on stage of blood suga r or race and ethnicity Model 3 additionally adjusts for marital status and B.M.I. For females, being marri ed is associated with a decreased probability of developing diabetes given EBS (b= 0.5 2 p<.01).. Being overweight or obese is again associated w ith increased log odds of developing diabetes relative to thos e wit h a "normal" B.M.I. (b=0.74 p<.01 ). Being in the youngest categories continues to be associated with reduced log odds of having elevated blood sugar relative to those in the oldest age cat egory, given one has EBS Model 4 allows for an interaction effect between race and ethnicity and education however no interactions are found to be significant. Stage Differences for Females Four covariates were found to have statistical effects that differ ed by level of blood glucose: being black, being in the youngest age category, being married, and BMI. The seventh hypothesis of a stronger association between education and lower levels of blood glucose is not supported by t he lack of an interaction betwe en educa tion and level of blood glucose. In stage 1, model 4, the effect of being black on probability of EBS was insignificant. In stage 2, model 4, being black was still insignificant in its effect on diabetes However, the two estimates are significantl y different from each other supporting the fifth hypothesis of an interaction between being black and stage of blood glucose Being black has no statistically significant effect on probability of either EBS or diabetes, but the effect in stage 2 is signif icantly greater than in stage 1 The lack of an interaction between being Hispanic and level of blood glucose in the association with probability of EBS supports the sixth hypothesis that the

PAGE 61

61 association between EBS probability and being Hispanic does not vary with level of blood glucose In stage 1, model 4, being married has no association with EBS. In stage 2, model 4, b eing married has a significant association with a reduced the probability of developing diabetes ( b = 0.77, p<.05). In stage 1, model 4, being overweight has a significant association with an increased probability of EBS (b= 1.32, p<.01). Like males, the direction of association of BMI with EBS remains constant into stage two, increasing the probability of diabetes given EBS (b=0 .77, p<.05) Again, the effect of BMI is significantly different and weaker for females in stage 2 than in stage 1. Sex Differences Stage One All tests of sex interaction were done using equation four. Three covariates were found to have significantly different eff ects by sex for stage 1 : having more than a high school education, having a BA or more, and BMI Having more than a high school degree was not significantly associated with EBS for males, however it significantly reduced the probability of EBS for females. The same pattern was found for having a BA or more. These findings suggest that educational attainment has a stronger association for females than for males in reduced probability of EBS for whites. The third covariate found to differ by sex in stage 1 wa s BMI. For females, the effect of being overweight was stronger than for males (b eta(males)=0 .94, b eta(females)= 1.32). This suggests that, like h igh educational attainment, higher BMI has a stronger association with an increased probability of EBS for fema les than for males.

PAGE 62

62 Stage Two Two sex differences in the effects of covariates on the probability of diabetes given EBS were found in stage 2: the effect of marital status and an interaction term for Hispanics and having a BA or more. For males, being marr ied does not have a significant effect on the probability of developing diabetes given IFG. For females however, being married has a significant effect on reducing the probability of diabetes (b= .54, p<.01). The interaction term is not significantly diff erent from zero for either males of females, but appears to have a different direction of effect: having a BA or more for male Hispanics is associated with higher probability of diabetes but having a BA or more for female Hispanics is associated with a low er probability. While the term approac hes significance for females neither term is significantly different from zero Figures 4 1 through 4 4 present estimated risk of EBS and diabetes given IFG for individuals in the oldest age category, single, and over weight based on equation four in tables 4 7 and 4 8 In Figure 4 1, one of the clearest rends is that for Hispanic males the estimated risk of EBS per 1,000 decreases with each increasing level of education, except for having more than a high school degree In Figure 4 2, Hispanics do not show the same nearly stepwise benefit f r o m increasing education. Whites however have a large difference in estimated relative risk of developing diabetes. Having less than a 9 th grade education for white males is estimate d to increase the probability of developing diabetes by 23% relative to a high school degree whereas having a BA or more reduces the probability by 14%. Figure 4 3 describes the risk of having EBS for females who are overweight, single, and over age 45. developing EBS, with the lowest risk among Hispanics with a high school degree and the highest risk in the two lower categories of education and in the highest category of educa tion. In Figure

PAGE 63

63 4 shaped trend disappears for relative risk of having diabetes for Hispanic females. Hispanics with a BA or more are estimated to have a 76% lower probability of having diabetes, given EBS, than Hispanics with a high school deg ree. Having a BA or more for blacks and whites is also estimated to reduce probabilities of developing diabetes relative to having a high school degree.

PAGE 64

64 Table 4 1 Frequencies and Prevalence of IFG for Individuals Aged 20 and Over Without Diagnose d Diabetes ** in the U.S. Population, NHANES 1999 2006, Aged 20 and Over Without Diagnosed Diabetes in the U.S. Population NHANES 1999 2006,Weighted Data Combined Reference** Weighted Weighted Weighted Frequencies Freque ncies Frequencies Frequencies Frequencies Prevalence Prevalence Prevalence Prevalence 1999 2000 2001 2002 2003 2004 2005 2006 1999 2002 1999 2002 1999 2002 2003 2006 1999 2006 Total 479(25.9%) 617(28.0%) 523(26.6%) 605(30.5%) 1096(27.1%) 2 6.0% 25.3% 27.9% 26.6% Men 282(32.6%) 380(36.1%) 290(30.7%) 365(37.9%) 662(34.5%) 33.0% 31.8% 34.2% 34.5% Women 197(20.0%) 237(20.6%) 233(22.9%) 240(23.6%) 434(20.4%) 20.0% 19.4% 22.1% 21.8% White 223(26.6%) 356(30.1%) 314(29.4%) 3 11(31.1%) 579(28.7%) 27.0% 26.2% 28.5% 27.3% Men 140(34.5%) 215(38.0%) 181(35.2%) 203(40.0%) 355(36.6%) 34.0% 33.0% 35.0% 34.0% Women 83(19.21%) 141(22.9%) 133(24.0%) 108(22.0%) 224(21.4%) 21.0% 20.0% 22.4% 21.2% Black 64(19.9%) 71(18.9%) 82 (22.5%) 120(27.3%) 135(19.4%) 17.0% 17.3% 23.0% 20.3% Men 29(20.6%) 39(21.6) 36(21.0%) 62(30.1%) 68(21.1%) 19.2% 19.1% 24.7% 22.0% Women 35(19.4) 32(16.5%) 46(23.8%) 58(24.7%) 67(17.9%) 15.0% 15.8% 21.7% 18.9% Hisp. 148(28.6%) 154(31.4) 94(23 .7%) 121(30.8%) 302(29.9%) 30.0% 29.2% 26.4% 27.7% Men 88(36.8%) 106(45.5) 52(26.3%) 67(36.6%) 194(41.1%) 41.1% 41.2% 30.3% 35.3% Women 60(21.6%) 48(18.7) 42(21.1%) 54(25.7%) 108(20.2%) 18.0% 15.6% 22.0% 19.0% Impaired Fasting Glucose defin ed as individuals with glucose levels 100 125 mg/ dL ( 5. 6 to 6.9 mmol/L) **Individual who had been told by doctor they had diabetes were excluded from the sample Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fasting Glucose in Adults in th e U.S. Population" Diabetes Care Vol. 29:1263 1268.

PAGE 65

65 Table 4 2 Frequencies and Prevalence (SE) of Diagnosed Diabetes for Individuals Aged 20 and Over in the U.S. Population, NHANES 1999 2006. Weighted Data Combined Reference** Weighted Weighted Weighted Frequencies Frequencies Frequencies Frequencies Frequencies Prevalence Prevalence Prevalence Prevalence 1999 2000 2001 2002 2003 2004 2005 2006 1999 2002 1999 2002 1999 2002 2003 2006 1999 2006 Total 480(9.8%) 511(9.5%) 545(10.8%) 509(10.2%) 2045(10.1%) 6.5% 6.5%2 7.7% 7.1% Men 233(10.3%) 248(9.8%) 269(11.1%) 247(104%) 997(10.4%) 6.7% 6.7% 7.4% 7.1% Women 247(9.5%) 263(9.2%) 276(10.5%) 262(10.1%) 1048(9.8%) 6.3% 6.3% 8.1% 7.2% White 146(6.6%) 220(7.7%) 243(9.0%) 192(7.7%) 801(7.8%) 5.6% 5.6% 6.9% 6.3% Men 84(7.9%) 112(8.4%) 130(10.1%) 92(7.6%) 418(8.5%) 6.1% 6.2% 6.7% 6.5% Women 62(5.4%) 108(7.1%) 113(8.0%) 100(7.8%) 383(7.1%) 5.0% 5.1% 7.1% 6.1% Black 130(14.3%) 118(11.7%) 124(12.5%) 169(15.1%) 541(13.4%) 10.0% 10.0% 11.7% 10.9% Men 55(13.4%) 53(11.1%) 53(11.1)% 89(16.4%) 250(13.1%) 8.2% 8.2% 11.4% 9.9% Women 75(15.0%) 65(12.2%) 71(13.7%) 80(13.8%) 291(13.6%) 11.4% 11.4% 12.0% 11.7% Hisp. 151(11.8%) 127(11.4%) 146(14.8%) 113(11.3%) 537(12.3%) 6.5% 6.5% 7.9% 7.3% Men 71(12.0%) 59(11.0%) 73(15.2%) 51(10.7%) 254(12.2%) 5.4% 5.3% 7.0% 6.2% Women 80(11.6%) 68(11.9%) 73(14.5%) 62(11.7%) 282(12.3%) 7.8% 7.7% 9.0% 8.4% Physician dia gnosis only of diabetes ** Cowie et al. 2006. "Prevalence of Diabetes and Impaired Fasting Glucose in Adults in the U.S. Population" Diabetes Care Vol. 29:1263 1268.

PAGE 66

66 Table 4 3 Prevalence (SE) of Diabetes as Measured from Either Dia gnosis Medication Usage or Blood Analysis for Individuals Aged 20 and Over in the U.S. population, NHANES 1999 2006 Weighted Data 1999 2000 2001 2002 2003 2004 2005 2006 1999 2006 Total 8.5%(.01) 10.0%(.01) 10.3%(.01) 10 .6%(.01) 9.9%(<.01) Men 9.1%(.01) 11.9%(.01) 11.4%(.01) 9.6%(.01) 10.6%(.01) Women 7.9%(.01) 8.3%(.01) 9.3%(.01) 11.5%(.01) 9.3%(.01) White 7.7%(.02) 9.1%(.01) 9.2%(.01) 9.7%(.01) 9%(.01) Men 9.6%(.02) 11.8%(.02) 11.3%(.01) 8.4%(.02) 10.3 %(.01) Women 5.9%(.01) 6.5%(.01) 7.3%(.01) 10.9%(.01) 7.7%(.01) Black 9.6%(.01) 13.2%(.02) 14.2%(.02) 15%(.02) 13.2%(.01) Men 5.1%(.01) 11.5%(.02) 12.9%(.03) 15.3%(.02) 11.6%(.01) Women 12.8%(.02) 14.6%(.03) 15.2%(.03) 14.8%(.02) 14.4%(.01) Hisp. 10.3(.01) 10%(.01) 11.6%(.03) 12.4%(.01) 11.3%(.01) Men 8.8%(.03) 9.5%(.02) 11.4%(.03) 11.4%(.03) 10.2%(.01) Women 11.6%(.05) 12.6%(.02) 11.8%(.04) 13.6%(.03) 12.4%(.02) N=1,854 N=2,205 N=1,977 N=1,982 N=8,018 *Diabetes includes ind ividual self reports of physician diagnosis, medication usage, and analysis of glucose levels from blood samples with diabetic levels >=7.0 mmol per L (>=126 mg per dL)

PAGE 67

67 Table 4 4 Prevalence (SE) of U.S. Population Aged 20 Years and Older with IFG* by Educational Attainment, NHANES 1999 2006: Weighted Data Im p a ired Fasting Glucose Less Than High School High School Degree College or More Total 31%(.013) 27.8%(.015) 24.3%(.010) Men 36%(.019) 31.8%(.020) 31.1%(.01 3) Women 26.4%(.015) 24.1%(.015) 18.1%(.011) White 34.8%(.013) 29.8%(.017) 24.6%(.013) Men 39.6%(.030) 34.0%(.024) 31.7%(.015) Women 30.5%(.028) 25.9%(.018) 18%(.014) Black 21.8%(.021) 23.4%(.026) 19.1%(.015) Men 22.4%(.028) 25.0%(.036) 20 .2%(.027) Women 21.2%(.212) 21.9%(.030) 18.3%(.019) Hisp. 30.6%(.016) 19.5%(.031) 21.1%(.022) Men 38.7%(.023) 23.9%(.039) 27.6%(.031) Women 22.2(.021)% 14.6%(.037) 15.0%(.032) N=2,440 N=1,872 N=3,667 Impaired Fasting Glucose defined as individuals with glucose levels 5.6 to 6.9 mmol per L (100 125 mg per dL)

PAGE 68

68 Table 4 5 Prevalence (SE) of Diabetes in U.S. Population by Educational Attainment Less Than High School High School Degree College or More Total 15. 2%(.01) 10.5(.01) 7.7%(.01) Men 12.8%(.01) 9.9%(.01) 10.1%(.01) Women 17.3%(.02) 11.1%(.01) 5.6%(.01) White 14.6%(.02) 9.9%(.01) 7.3%(.01) Men 14.3%(.02) 9.5%(.02) 9.9%(.01) Women 15%(.02) 10.2%(.01) 4.9%(.01) Black 17.5%(.02) 14.2%(. 02) 9.9%(.01) Men 14.1%(.02) 10.6%(.02) 10.0%(.02) Women 20.8%(.03) 17.4%(.03) 9.8%(.02) Hisp. 14.1%(.01) 8.4%(.02) 8.6%(.02) Men 10.8%(.02) 8.5%(.03) 10.6%(.03) Women 17.5%(.03) 8.2%(.03) 6.7%(.03) *N=2,450 N=1,880 N=3,673 *Diabetes inc ludes individual self reports of physician diagnosis, medication usage, and analysis of glucose levels from blood samples with diabetic levels >=7.0 mmol per L (>=126 mg per dL)

PAGE 69

69 Table 4 6 Proportions and Means of U.S. Born Population( SE) Over the Age of 24. Weighted Data NHANES 1999 2006 White Black Hispanic Age(In Years) 48(.42) 43.3(.50) 40.7(1.6) Education Less Than 9th Grade 4% 5% 10% Less Than HS Degree 9% 24% 21% High School Degree 28% 24% 22% More Than Hi gh School 32% 33% 37% BA or More 27% 13% 10% Marital Status Married 62% 33% 57% BMI Over 64% 74% 74% Under 2% 2% 2% Normal 34% 24% 24%

PAGE 70

70 Table 4 7 Results From a Logistic Regression Continuati on Ratio Model for Males Greater Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1): NHANES 1999 2006 Model 1 Model 2 Model 3 Model 4 Sex diff. Stage diff. Coef. SE Coef. SE Coef. SE Coef. SE Constant 0.27 0.07** 0.33 0.09** 0.62 .15** 0.61 0.15** Race (White=Ref) Black 0.28 0.12* 0.36 0.12** 0.38 .13** 0.17 0.19 x Hispanic 0.14 0.16 0.07 0.16 0.05 0.19 0.36 0.4 x Age (>45=ref) Age24 35 1.6 0.13** 1.56 0.13** 1.54 0.13** 1. 53 .14** Age 36 45 0.84 0.12** 0.84 0.13** 0.86 .13** 0.86 .13** Edu. (HSDG=Ref) < 9th Grade 0.04 0.19 0.14 0.19 0.02 0.21 High School 0.04 0.11 0.04 0.1 0.09 0 .11 # BA or More 0.33 .13* 0.26 0.13 0.25 0.13 # Married (1=Yes) 0.12 0.11 0.11 0.11 BMI( Normal =ref) Over 0.94 .10** 0.94 ** 0.09 # x Educ*Race Black*LT9th 0.02 0.46 Black*LTHS 0.16 0.32 Black*MTHS 0.51 0.28 Black* BAOM 0.09 0.4 Hisp.*LT9th 1.12 0.75 Hisp.LTHS 0.7 0.63 Hisp.*MTHS 0.14 0.22 Hisp.*BAOM 0.22 0.71 *=p<.05; **p<.01; x=sig. diff. by stage; #=sig diff. by sex

PAGE 71

71 Table 4 8 Results from a Logistic Regression Continuation Ratio Model for Males Greater Than 24 Years Old for Having Diabetes G iven EBS (Stage 2 ): NHANES 1999 2006 Model 1 Model 2 Model 3 Model 4 Sex diff. Stage diff. Coef. SE Coef. SE Coef. SE Coef. SE Constant 1 0.09 1.1 0.17 1.01 0.22 1.01 0.23** Race (White=Ref) Black 0.56 .18** 0.52 .19* 0.54 .21* 0.56 0.40 x Hispanic 0.7 0.33* 0.65 0.32* 0.62 .25* 0.25 0.57 x Age (>45=ref) Age24 35 1.13 .35** 1.14 0.35** 1.2 0.36** 1.23 .37** Age 36 45 0.42 0.25 0.4 0.26 0.48 0.26 0.48 0.27 Edu. (HSDG=Ref) < 9th Grade 0.14 0.27 0.19 0.29 0.3 0.34 HS 0.27 0.22 0.23 0.22 0.23 0.25 BA or More 0.19 0.22 0.18 0.23 0.2 0.25 Married (1=Yes) 0.05 0.17 0.05 0.17 BMI(Normal =ref) Over 0.52 .22* 0.51 0.23* x Educ*Race Black*LT9th 0.14 0 .56 Black*LTHS 0.32 0.45 Black*MTHS 0.35 0.53 Black* BAOM 0.12 0.67 Hisp.*LT9th 0.41 0.88 Hisp.LTHS 0.76 0.73 Hisp.*MTHS 0.57 0.68 Hisp.*BAOM 0.72 0.97 # *=p<.05; **p<.01; x=s ig. diff. by stage; #=sig diff. by sex

PAGE 72

72 Table 4 9 Results From a Logistic Regression Continuation Ratio Model for Females Greater Than 24 Years Old for Having EBS Given Normal Blood Glucose (Stage 1): NHANES 1999 2006 Model 1 Model 2 Model 3 Model 4 Sex diff. Stage diff. Coef. SE Coef. SE Coef. SE Coef. SE Constant 0.43 .08** 0.28 0.09** 0.22 0.11* 0.23 0.12 Race (White=Ref) Black 0.35 .12** 0.23 .12* 0.03 0.12 0.05 0.23 x Hispanic 0.43 .14** 0.21 0.15 0 .14 0.17 0.39 0.33 Age (>45=ref) Age24 35 1.95 .18** 1.9 0.18** 1.9 .18** 1.9 .19** x Age 36 45 0.42 .08** 0.93 0.14** 0.88 .14** 0.89 0.15 Edu. (HSDG=Ref) < 9th Grade 0.41 .17* 0.47 .19* 0.7 .23** HS 0.27 .12* 0.32 .13* 0.35 0.15* # BA or More 0.56 .14** 0.51 .14** 0.57 .16** # Married (1=Yes) 0.15 0.09 0.15 0.1 x BMI(Normal =ref) Over 1.33 .13** 1.32 .13** # x Educ*Race Black*LT9th 0.94 0.47* Black*LTHS 0.22 0.34 Black*MTHS 0.01 0.31 Black* BAOM 0.13 0.37 Hisp.*LT9th 0.27 0.46 Hisp.LTHS 0.5 0.37 Hisp.*MTHS 0.53 0.46 His p.*BAOM 1.6 0.55** *=p<.05; **p<.01; x=sig. diff. by stage; #=sig diff. by sex

PAGE 73

73 Table 4 10 Results from a Logistic Regression Continuation Ratio Model fo r Fe ales Greater Than 24 Years Old for Having Diabetes G iven EBS (St age 2 ): NHANES 1999 2006 Model 1 Model 2 Model 3 Model 4 Sex diff. Stage diff. Coef. SE Coef. SE Coef. SE Coef. SE Constant 0.87 .10** 0.78 0.15 0.35 0.18 0.36 0.19 Race (White=Ref) Black 0.84 .15** 0.83 .15** 0.59 .15** 0.64 0.34 x Hispanic 0.38 0.2 0.3 0.19 0.39 0.21 0.92 0.50 Age (>45=ref) Age24 35 0.36 0.4 0.27 0.42 0.33 0.39 0.34 0.41 x Age 36 45 0.58 0.3 0.56 0.3 0.49 0.32 0.49 0.33 Edu. (HSDG=Ref) < 9th Grad e 0.16 0.34 0.16 0.35 0.29 0.40 HS 0.49 0.26 0.5 .25* 0.5 0.30 BA or More 0.59 0.34 0.55 0.34 0.36 0.37 Married (1=Yes) 0.52 0 .18** 0.54 0.18** x BMI(Normal =ref) Over 0.74 .29* 0.77 .30* x Educ*Race Black*LT9th 0.18 0.77 Black*LTHS 0.21 0.56 Black*MTHS 0.08 0.54 Black* BAOM 0.84 0.69 Hisp.*LT9th 1.3 0 0.86 Hisp.LTHS 0.13 0.68 Hisp.*M THS 0.44 0.67 Hisp.*BAOM 1.9 0 1.20 # *=p<.05; **p<.01; x=sig. diff. by stage; #=sig diff. by sex

PAGE 74

74 A B Figure 4 1 Estimated Risk of EBS by Educational Attainment and Race For Males Over Age 45 : A) Estima ted Risk and B) Estimated Relative Risk 0 200 400 600 800 1000 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 652 690 648 668 589 HISPANIC 801 758 562 618 555 BLACK 565 562 555 450 470 Estimated Risk (per 1,000) of Elevated Blood Glucose 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 1.01 1.06 1.00 1.03 0.91 HISPANIC 1.42 1.35 1.00 1.10 0.99 BLACK 1.02 1.01 1.00 0.81 0.85 Estimated Relative Risk (Compared to High School)

PAGE 75

75 A B Figure 4 2 Estimated Risk of Diabetes Given EBS by Edu cational Attainment and Race: A) Estimated Risk and B) Estimated Relative Risk 0 100 200 300 400 500 600 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 330 259 267 314 230 HISPANIC 295 490 319 510 440 BLACK 428 458 389 361 371 Estimated Risk (per 1,000) of Elevated Blood Glucose 0.00 0.50 1.00 1.50 2.00 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 1.23 0.97 1.00 1.18 0.86 HISPANIC 0.93 1.54 1.00 1.60 1.38 BLACK 1.10 1.18 1.00 0.93 0.95 Estimated Relative Risk (Compared to High School

PAGE 76

76 A B Figure 4 3 Estimated Risk of EBS by Educational Attainm ent and Race For Fem ales Over Age 45, Single, and Overweight : A) Estimated Risk and B) Estimated Relative Risk 0 100 200 300 400 500 600 700 800 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 717 650 557 470 416 HISPANIC 567 675 460 505 705 BLACK 510 611 570 485 460 Estimated Risk (per 1,000) of Diabetes Given EBS 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 1.29 1.17 1.00 0.84 0.75 HISPANIC 1.23 1.47 1.00 1.10 1.53 BLACK 0.90 1.07 1.00 0.85 0.81 Estimated Relative Risk (Compared to High School

PAGE 77

77 A B Figure 4 4 Estimated Risk of Diabetes Given EBS by Educational Attainment and Race For Fem ales Over Age 45, Single, and Overweight : A)Estimated Risk and B) Estimated Relative Risk 0 100 200 300 400 500 600 700 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 483 371 411 297 327 HISPANIC 389 565 636 406 154 BLACK 596 579 570 465 285 Estimated Risk (per 1,000) of Diabetes Given EBS 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 < 9th HSDG BAOM < 9th HSDG BAOM WHITE 1.17 0.90 1.00 0.72 0.80 HISPANIC 0.61 0.89 1.00 0.64 0.24 BLACK 1.05 1.02 1.00 0.82 0.50 Estimated Relative Risk (Compared to High School

PAGE 78

78 CHAPTER 5 DISCUSSION AND CONCL USION Discussion The objective of this thesis was to answer the following three questions: (1) What is the association between educ ation and the development of elevated blood sugar (2) does this association vary by race and ethnicity and sex, and (3) does the association between race and ethnicity, sex, and elevated blood sugar vary depending on the level of blood sugar? Seven hypotheses emerged from these three research quest ions. First, i t was hypothesized that as the level of education increased, the probability of having EBS would decrease. The results of the analysis suggest that there is indeed a positive association between the development of EBS and educational attainme nt for females but not for males Females with less than a high school degree have a significantly increased probability for developing EBS, and females with more than a high school education have a significantly decreased probability of developing EBS. Mo reover, the effect of education is found to vary significantly by sex. The effect of having more than a high school education on probability of EBS is significantly greater for females than for males. This finding supports previous research that suggests i ncreasing socioeconomic status attainment has a stronger effect for females than for males for blood glucose related outcomes (Stern et al. 1984). Only limited l ittle evidence is found in support of the second and third hypotheses of an interaction betwe en race and educational attainment. Two interactions were found to be significant for females, however interpretation of these terms is difficult. In stage 1, being black significantly reduced the probability of diabetes given less than a 9th grade educati on. Figure 4 3 aids in the interpretation of this interaction term by estimating probabilities by level of education A downward trend in the relative risk of developing diabetes as educational

PAGE 79

79 attainment increases ca n be observed for white females; howev er n o such trend is observed for black females. Thus for females, there is some evidence in support of the third hypothesis. Past research has suggested that a downward trend of diabetes with increasing education for whites exists, but not for blacks (Borr ell 2006). One possibility is that these associations happened by chance i.e. noise in the data. Only 57 black females had less than a 9th grade education across all eight years, making a chance significant finding plausible. Another possibility is that o ne of the previous reasons given for expecting variation in the effect of education plays a role for blacks. It could be the effect of genes: black females may be predisposed to development of EBS he protective ef fect of education Pre natal environments (Barker hypothesis) may be worse for females, and the mechanism their bodies adapt to scarcity may predispose them to risk if their post natal environment overwhelms their metabolic mechanisms for dealing with gluc ose. Cultural capital explanations may also play the role: it could be that black females and the schools they attend are ill equipped to aid in their accumulation of human capital that pays dividends in later hea lth outcomes. Unfortunately, th e s e data mak es it impossible to disentangle the multitude of early life and current life events, as well precluding a direct test of any of these hypotheses. The second significant interaction was among Hispanic fem ales with a BA o r more. In this case, being Hispanic and having a BA or more had a significant effect on increasing the probability of EBS. This interaction does not support the third hypothesis that the statistical effect of education would be similar for whites and Hispanics. Again, there are several possible but un testable explanations for this interaction term. It could be noise in the data: Only Hispanic 39 femal es had a BA or more across all eight years, and 47% of these highly educated Hispanics

PAGE 80

80 had EBS. Again, early life events could have an influence on this outcome: Hispanic females with a BA probably have greater access to food resources than lesser educated Hispanic females. If genetic predispositions or early then greater access to food resources may not necessarily predict better health outcomes in terms of blood glucose and blood glucose related health risks. Another explanation is that having a BA or more is a proxy for level of acculturation. Acculturatio n has been found to have varying effects on the probability of developing diabetes. Some studies suggest that among Mexican Americans, greater acculturation is associated with lower risk of diabetes, whereas other studies suggest that higher levels of accu lturation are associated with greater risk of diabetes for non Mexican Hispanics (Hazuda 1988 Kandula 2008). The association with acculturatio n may vary by ethnic origin and may also vary by time. As American society becomes increasingly overweight, accu lturation to such a lifestyle may be increasingly associated with risk of diabetes. Evidence of the reduced effect of having a BA or more on lowering the probability of EBS for Hispanics lends support to the possibility that acculturation to a US lifestyle may play a role in increasing risk of EBS. The effect of the interaction between Hispanics and higher levels of education on increasing the probability of an adverse outcome runs counter to traditional ideas of a Hispanic paradox. Typically, Hispanics hav e been found to have lower mortality rates than would be expected given the socioeconomic the effect is likely due to reverse migration (Palloni 2004) Sin ce the health outcome of interest is reverse migration explana tion insofar as the interaction term is valid. Future research should

PAGE 81

81 adjust for levels of accultu ration to see if it accounts for some of the counter intuitive effects of education on Hispanic females at the extreme end of educational attainment. R acial/ethnic differences and sex differe nces in associations with EBS also exist No evidence is found in support of the fourth hypothesis of no sex and education interactions. There is strong support for the fifth hypothesis however. Among males and females blacks are at lower risk of developing impaired fasting glucose relative to whites, even after adjustments for age, educat ion, marit al status and body mass index This seems counter intuitive given the higher risk of diabetes for blacks, but is a finding borne out in other research (Cowie 2002). With regards to the third objective w hat makes the finding of the protective effect of bein g black on IFG especially interesting is the significant reversal of the effect of being black on developing diabetes, given one has EBS Among blacks, t his significant difference of effect is found for both males and females. There is mixed evidence for the sixth hypothesis. Among males, being Hispanic has a significantly different effect depending on the level of blood glucose. No such interaction is found for females. Unlike blacks, being Hispanic is associated with an increased probability of EBS at bo th stages of blood glucose. Given EBS the statistical effect of being Hispanic on the probability of having diabetes is significantly different, and stronger than association being Hispanic has when modeling EBS alone. The seventh hypothesis stated that a s the level of blood glucose increased from EBS to diabetes given IFG, the association of education with the probability of EBS would decrease. This hypothesis was not supported as education is found to have no statistical effect on increasing the probabil ity of developing diabetes, given EBS. It is important to remember,

PAGE 82

82 diabetes, rather, only that it is not associated with increased probability of progression f rom EBS to diabetes. For males, BMI is the only other variable with a different effect across stages of EBS The difference of effect of BMI is also seen in females. The effect of BMI on increasing the probability of developing diabetes is significant for both males and females and for both stages of blood sugar. However, the effect of BMI on increasing risk of IFG is significantly greater than the effect of BMI on increasing risk of diabetes. These findings suggest that while BMI is an important risk facto r for EBS, other factors help to "push" an individ ual across the edge into diabetes One avenue of further research would be to differentiate between "overweight" and "obesity" in the CRM model to test if obesity retains an equally significant effect on ri sk of diabetes. Another possibility would be to examine race by BMI interactions to see if that extra something need for BMI to push people into being diabetic is being in a particular racial and ethnic group. These BMI findings also demonstrates one advan tage of using BMI coded ordinally: it allows one to capture the varying effects of stages of BMI on risk of EBS. For females, the effect of BMI is also found to be significantly different depending on level of blood glucose. Unique to females however, is the significant reversal of the effect of being married. Being married has no significant effect on probability of EBS but has a significant effect on reducing the probability of diabetes given EBS. control covariates such as low If one conceptualizes race as a marker for a wide range of social experiences that influence health outcomes, then even if every conceivable experience wa s controlled for and explained the relationship, a health disparity would remain so long as race and ethnicity exerted constraints upon an individual's life

PAGE 83

83 experience. This conceptualization views the health related factors that contribute to the health By adjusting for BMI, age, contr racial differences. However, one objective of the analysis is to test for different effects of the variables across different stages of elevated blood sugar, thus not including them in attempts to controllin Conclusion This thesis b egan with the conceptualization of EBS as a racial and ethnic health disparity, implying that differences in health outc omes related to blood glucose can and should be addressed EBS is an especial ly important health outcome to address in the U.S today: it is increasing in incidence and prevalence and is a risk factor for the leading cause of death in the U.S. today, cardiovascular disease. Disparities such as that associated with EBS can occur as a result of health behaviors, health care access, and health care process. This paper has conceptualized education as one possible factor that might explain some of the racial disparity in E BS. Education certainly has an e ffect within all three origins of health disparities. Those who are educated practice healthier lifestyles, have better jobs, greater levels of health insurance, and can afford better quality health care. If the effect of education had been found to be significantly different by race and e thnicity, then further research into why the effect varied would be warranted in an attempt to maximize the beneficial effect of education for all. This research only found the effect of education to vary at 2 intersections of race and ethnicity and educat ion: Hispanics with a BA or more are at an increased risk of EBS and blacks with less than a 9 th grade education are at a decreased risk. Whether these variations are due to noise, early life differences, different experiences within the

PAGE 84

84 educational settin g, or different abilities to translate educational resources into better health behaviors access and process requires further research. A strength of this study is its conceptualization of elevated blood glucose as occurring along a spectrum the categori zation of the spectrum into separate stages, and the testing for differences of effect of the covariates across the stages of blood glucose. Using a CRM, t he effect of race, education, BMI, and marital status were all found to have significantly different effects given the level of blood glucose. Of these, perhaps the finding that BMI has a significantly stronger association with the probability of EBS than on diabetes given EBS provides the most immediate recommendations for health interventions. While ob esity has been recognized a primary factor in the increasing prevalence of diabetes, it plays an even more important role in the development of EBS as a whole. Continuing efforts at reducing the increasing rates of obesity should also include the reduction of elevated blood sugar as a c entral objective.

PAGE 85

85 APPENDIX ADDITIONAL EDUCATIONAL ATTAINMENT Table A 1 Percentage of Educational Attainment Within Five Category Educational Attainment and By Race/Ethncity in the U.S. Population NHANES 1999 2006 Weig hted Data Less Than Less Than High School Some College/ BA+ 9th Grade High School Degree Training Black 5.9% 26.2% 24.1% 31.0% 12.6% Hispanic 8.4% 19.5% 25.7% 33.4% 12.6% White 3.5% 9.9% 27.7% 31.0% 27.8%

PAGE 86

86 Table A 2 Prevalence (SE) of U.S. Born Population Aged 20 years and Older With IFG* By Educational Attainment, NHANES 1999 2006 Weighted Data Less Than High School High School Degree College or More Total 31.30% 2 9.10% 24.30% Men 34.5% 33.6% 30.9% Women 28.4% 24.8% 18.5% White 34.5% 30.2% 24.8% Men 38.4% 35.9% 31.7% Women 30.8% 25.7% 18.3% Black 22.5% 24.2% 19.2% Men 23.0% 26.1% 21.0% Women 22.0% 22.5% 18.1% Hisp. 30.8% 21.6% 18.9% Me n 38.4% 25.4% 19.6% Women 25.2% 17.6% 18.2% N=1,410 N=1,603 N=3,170 Impaired Fasting Glucose defined as individuals with glucose levels 5.6 to 6.9 mmol per L (100 125 mg per dL)

PAGE 87

87 LIST OF REFERENCES Ada ms, Scott J. 2002. Educational Attainment and Health: Evidence from a Sample of Older Adults Education Economics 10(1): 97 110. Agency for Healthcare Resea rch and Quality. 2005. AHRQ Pub lication No. 05 0034, Retrieved on March 18 th 2 009 from http://www.ahrq.gov/data/hcup/highlight1/high1.htm Allison, Paul. 1999. Logistic Regression Usi ng SAS: Theory and Application Cary, N orth Carolina : SAS Institute, Inc. American Diabetes Association. 2004. Position Statements: Diagnosis and clas sification of Diabetes Care. 27(Suppl. 1):S5 S10. Ardern, Chris and Peter Katzmarzy and Demographic Variation in the Prevalence of the Metabolic Syndrome in Canada. Canadian Journal of Diabetes 31(1):34 46. Bertoni Alain, Julie Krop, Gerard Anderson and Frederick Brancati. 2002. Related Morbidity and Mortality in a Diabetes Care 25:471 475 Brancati Frederick, P K Whelton L H Kuller and M J Klag tus, race, and socioeconomic status:a population Annals of Epidemiol og y. 6:67 73. Brown, JB. G A Nichols, H S Glauber and A W Type 2 diabetes: incremental medical care costs during the first 8 years after diagnosis Diabet es Car e 22(7):1116 1124 Borrell, Luisa, Florence J D allo and Kellee White 2006. Education and Diabetes in a Racially and Ethnically Diverse Population American Journal of Public Health 96:1637 1642. Burns, N.; Finucane, F .; Hatunic, M 2007. "Earl y onset type 2 diabetes in obese white subjects is characterized by a marked defect in beta cell insulin secretion, severe insulin resistance and a lack of response to aerobic exercise training Diabetologia 50(7):1362 1364 Caliif, RM, M. Boolell S M. Ha ffner M A Bethel J. McMurray A. Duggal and R.R. Holman. 2008. glucose tolerance: rationale and design of the Nateglinide And Valsartan in Impaired Glucose Tolerance Outcomes American Heart Journal 4 : 623 32. Cagney K A and Diane S. Lauderdale J ournal s of Gerontol ogy 57 (2): 163 1 72.

PAGE 88

88 Centers for Disease Control and Prevention 1990 Mor bidity and Mortality Weekly Report : Current Trends in Regional Variation in Diabetes Mellitus Prevalence -United States, 1988 and 1989 Retrieved on February 14 th 2009 from http://www.cdc.gov/mmwr/preview/mmwrhtml/00001830.htm C enters for Disease Contr ol and Prevention. 2008 a Adjusted Percentage of Civilian, Non institutionalized Population with Diagnosed Diabetes, United States, 1980 on February 14 th 2009 from http://www.cdc.gov/diabetes/statistics/prev/national/figage.h tm -----. 2008 b Morbidity and Mortality Weekly Report : State Specific Incidence of Diabetes Among Adults --Participating States, 1995 -1997 and 2005 -2007. Retrieved on April 10 th 2009 from http://www.cdc.gov/mmwr/preview/mmwrhtml/mm5743a2.htm Choi B.C. and F. Shi 2001." Risk factors for diabetes mellitus by age and sex: results of the National Population Health Survey Diabetologia 44( 10 ) :1221 1231. Cohen, D A ., B. Finch, A. Bower and A Sastry. 2005. Collective efficacy and obesity: The potie ntal influence of social factors on health. Social Science and Medicine 62(3):769 778. Cowie, C 2006. Prevalence of diabetes and impaired fasting glucose in adults in the U. S. population: National Health a nd Nutriti on Examination Survey 1999 2002". Di abetes Care 29(6) : 1263 126 8. Cutler, David M. and Adriana Lleras Muney. 2007."Education and Health: Evaluating Theories and Evidence. NBER Working Paper No. W12352 Retrieved on April 20 th 2009 from http://ssrn.com/abstract=913315 Elo, I. and S. Preston 1996. Educational differences in mortality: United States, 1979 1985 Social Science and Medicine 42 : 47 57. Ettaro, Lorraine, Thomas Songer, Ping Zhang, and Michael Englegau 2004. "Cost of illness studies in diabetes mellitus Pharmacoeconomic s 2 2(3): 149 64. Ford, Earl, Wayne Giles and William Dietz Among US Adults Findings From the Third National Health and Nutrition Examination Survey Journal of the American Medical Association 287:356 359. F einstein, Jonathon S. 1993. The Association Between Socioeconomic Status and Health: A Review of the Literature. The Milbank Quarterly 71: 279 322. Gro ssma n, M. and R. Kaestner. 1997. Effects of education on health ." J. R. Behrman and N. Stacey. (Eds ) The Social Benefits of Education Ann Arbor, MI : University of Michigan Press.

PAGE 89

89 V L Goetsch D.J. Wiebe, L.G. Veltum, and B. Van Dorsten. Stress and blood glucose in Behaviour Research and Therapy 28(6): 531 537. Goldb erg, J., W. Hayes and J. Huntley. 2004. "Understanding Health Disparities." Health Policy Institute of Ohio Gregg Beckles Leveille Williamson and Narayan Engelgau 2 Diabetes and physical Diabetes Care 23:1272 7 Herbert, P.L., J.E. Sisk, and E.A. Howell. 2008. When does a difference become a disparity? Conceptualizing racial and ethnic disparities in health. Health Affairs 27(2): 374 382. Hermann Nabi, Mika Kivimki, Michael G M armot Jane Ferrie Marie Zins Pierre Ducimetire Silla M Consoli, and Archana Singh Does personality International Journal of Epidemiology 37:591 602. Ioannou, George N., Chris L. Bryson, an d Edward Boyko. 2007 "Prevalence and trends of insulin resistance, impaired fasting glucose, and diabetes." Journal of Diabet es and its Complications. 21:363 370. U.S. Department of Health and Human Services. Healthy People 2010 2nd ed. With Understandin g and Improving Health and Objectives for Improving Health. 2 vols. Washington, DC: U.S. Government Printing Office, November 2000. Kitagawa, Evelyn M. and Philip Hauser 1968. Education Differentials in Mortality by Cause of Death :United States, 1960. Demography 5: 318 353. Kouznetsova, Maria, Xiaoyu Huang Jing Ma Lawrence Lessner and David O. Carpenter. 2007. Increased Rate of Hospitalization for Diabetes and Residential Proximity of Hazardous Waste Sites Environmental Health Perspectives 115(1) :75 79. Mayfield J A and Deb P Whitecotton Work disability and diabetes Diabetes Care 22(7):1105 9. Meara, E.R., S. Richards, and D.M. Cutler. 2008. The Gap Gets Bigger: Changes in Mortality and Life Expectancy, by Education, 1981 2000. Health Affairs 27(2): 350 360. Mcguire, D.K. and J.L. Petersen. 200 glucose -Diabetes and Vascular Disease Research 2( 1):9 15. McWilliams, Michael, Meara Ellen and Alan M. Zaslavsky. 2007. "Health of Previousl y Uninsured Adults After Acquiring Medicare Coverage JAMA 298(24): 2886 2894.

PAGE 90

90 Mirowski, John and Catherine Ross. 2003. Education, Social Status, and Health New York New York: Aldine de Gruyter Moen, E. 1999 Education, ranking, and competition for j obs ". Journal of Labor Economics 17 : 694 723. Mokdad Ali H., Barbara A Bowman and Earl S. Ford Obesity and Diabetes in the United States Journal of the American Medical Association. 286(10):1195 1200 Monteiro, Caro ls, Conde Wolney and Barry Popkin. 2001. "Independent Effects of Income and Education on the Risk of Obesity in the Brazilian Adult Population Jou r nal of Nutrition 131 : 881 886. National Center for Health St il 12 th 2009 from http://www.cdc.gov/nchs/tutorials/Nhanes/SurveyDesign/SampleDesign/intro.htm National Institute of Diabetes and Digestive and Kidney Diseases. National Diabetes Statistics, 2007 fact sheet. Bethesda, MD: U.S. Department of Health and Hum an Services, National Institutes of Health, 2008. Nichols, Gregory A., Teresa Hillier and Jonathon Brown 2007. "Progression from Newly Acquired Impaired Fasting Glucose to Type 2 Diabetes Diabetes Care 3 : 228 233. Ones, Deniz S. and Neil Anderson 200 Scales in Selection: Some British Data. Journal of Occupational and Organizational Journal of Occupational & Organizational Psychology 75(3):255 276. Peterson, J.L, and D.K. McGuire 20 05. "Impaired glucose tolerance and impaired fasting glucose a review of diagnosis, clinical implications and management". Diabetes and Vascular Disease Research 2(1) :9 15 Palloni, Alberto and Elizabeth Arias 2004. "Paradox Lost: Explaining the Hispan ic Adult Mortality Advantage Demograph y Vol. 41(3) pp. 385 415 Palloni, Alberto. 2006. Reproducing Inequalities: Luck, Wallets, and the Enduring Effects of Childhood Health. Demography 43:587 615. Rabi, Doreen, Alun Edwards, Danielle S outher, Larence Svenson, Peter Sargious, Peter Norton, Eric Larsen, and William Ghali. economic status with diabetes prevalence and utilization of diabe BMC Health Services Research 6:124. Preston, S.H ., and I.T. Elo. 1995. Are Educational Differentials in Adult Mortality Increasing in the United States? Journal of Aging and Health 7(4): 476 496.

PAGE 91

91 Rao, Shobha S ., Phillip Disraeli and Tamara McGregror and Impaired Fas ting Glucose American Family Physician. Retrieved on November 20, 2008 from http://www.aafp.org/afp/20040415/1961.html Robbins, J M ., V. Vaccarino, H. Zhang, and S.V. Kasl 2005. "Socioeconomic Status and Diagnosed Diabetes Incidence Diabetes Researc h and Clinical Practice 68(3) : 230 236. Does Education Really Disadvantage Women in the Marriage Market ? University of Washington Institute for Economic Research Working Paper No. UWEC 2003 15. Retrieved April 1 2009 from SSRN: http://ssrn.com/abstract=423360 or DOI: 10.2139/ssrn.423360 Rosenbloom, A L ., J.R. Joe, R.S. Young, and W.E. Winter. diabetes in youth Diabetes Care 22(2):345 35 Ross Catherine E. and Chia Lang Wu 1995. The Links Between Education and Health American Sociological Review 60 (5): 719 745. Ross Catherine E., and Chia Lang Journal of Health and Social Behavior 37 (1): 104 120. Ross, Catherine E., and John Mirowsky. 1999. "Refining the Association between Education and Health: The Effects of Quantity, Credential, and Selectivity Demography 36 : 4 : 445 460. Shai Im R. Jiang, and J.E. women: a 20 year follow up study. Diabetes Care 29:1585 90 Shea, S., A. Stein, C. Basch, P. Lantigua,, C. Maylahn, D. Strogatz and L. Novick. 1991 Independent associations of educational attainment and ethnicity with behavioral risk factors for cardiovascular disease American Journal of Epidemiology 134 :567 582 Shoenborn, C A. 2004. "Marital status and health: United States, 1999 2002 Advanced D ata 351: 1 32. Smedley, Brian D., Adrieene Stith, and Alan R. Nelson. 2003. Committee on Unde rstanding and Eliminating Racial and Ethnic Disparities in Health Care Washington D.C. : The National Academies Press Smith, James. 2007. "Diabetes and the Rise of the SES Health Gradient". NBER Working Paper No. W12 905 Retrieved April 1 2009 from http: / /ssrn.com/abstract=963738 Smith, James. 2005 "Unraveling the socioeconomic status : Health Connection" Population and Development Review 30:108 132.

PAGE 92

92 Snehalatha, C., A. Ramachandran, K. Satyavani, S. Sivasankari and V. Vijay. 2003 "Clustering of car diovascular risk factors in impaired fasting glucose and impaired glucose tolerance International Journal of Diabetes in Developing Countrie s 23:59 61. Stunkard A J. and Thorkild I.A. Sorensen Obesity and Socioeconomic Status: A Complex Relatio n The New England Journal of Medicine 329:1036 1037 Valdmanis, D W Smith and M R with diabetes American Journal of Public Health 91(1):129 130. Whitehead M. The concepts and principles of equity and health. International Journal of Health Services 22(3):429 45. Winkelby, M.A., D.E. Jatulis, E. Frank and S.P. Fortmann. 1992. "Socioeconomic status and health: how education, income, and occupation contribute to risk factors for cardio vascular disease." American Journal of Public Health 82(6): 816 820. Zimmer, Zachary, Albert Hermalin, and Hui Sheng Ling 2002."Whose Education Counts? The Added Impact of Adult Child Education on Physical Functioning of Older Taiwanese" The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 57:S23 S32

PAGE 93

93 BIOGRAPHICAL SKETCH Gregory Pavela earned his Bachelor of A rts from the University of Virginia with majors in sociology and history. He is currently enrolled as a graduate s tudent at the University of Florida and received his Master of Arts in sociology from the University of Florida in 2009. He plans to continue his research on diabetes and health disparities.

PAGE 94

94 EDUCATION DIFFERENCES IN ELEVATED BLOOD SUGAR: DO THEY VARY BY RACE, ETHNICITY AND SEX? Gregory Pavela (540) 645 0579 Sociology and Criminology & Law John Henreta Master of Arts July 2009 This thesis develops our understanding of the risk factors associated with impaired fasting glucose (pre diabetes) and diabetes. Diabetes is a risk factor for cardiovascular disease, the leading cause of death in the U.S., and as such it is important to understand how known risk factors such as race and ethnicity, obesity, and education interact with each other a nd with the level o f disease to increase the risk of development of diabetes. The goal of such an understanding is to reduce the overall prevalence of the disease and to reduce racial and ethnic disparities in diabetes.