The price of unhealthy food relative to healthy food and its association with diet quality,

diabetes, and insulin resistance in a multi-ethnic population

A Thesis

Submitted to the Faculty

of

Drexel University

by

David Michael Kern, MS

in partial fulfillment of the

Requirements for the degree

of

Doctor of Philosophy

December 2016

ii

© Copyright 2016

David M. Kern. All Rights Reserved.

iii

Dedications

This dissertation is dedicated to my cousin Joe “Ozer” Matteo, whose life reminds me to

take nothing for granted.

Carpe diem.

iv

Acknowledgements

Completing this dissertation and degree wouldn’t have been possible without the

love and support of my wife, Tabby. We accomplished a ton these past few years and I’m

looking forward to going back to spending lazy weekends together. Unfortunately, we

can no longer use my dissertation as an excuse to get out of plans.

I also want to thank my advisor, Amy Auchincloss, for her invaluable support and

mentoring from day one. Amy knew when I needed to be pushed to finish things up, and

when to give me time when thing were hectic at work and in my personal life. I’ve never

met anyone who works as hard as Amy; I don’t know where she finds the time to do

everything she does – teaching, mentoring, research, writing grants, reading this

dissertation, etc. – but it’s inspiring. Her input and revisions of my work helped me turn

this dissertation into something I’m incredibly proud of. The other members of my

dissertation committee also helped shape this research – Lucy Robinson for her statistical

insight, Ana Diez Roux for her unrivaled expertise in social and neighborhood

epidemiology, Genny Pham-Kanter for her blend of public health and economic

knowledge, and Mark Stehr for his economic expertise and for spending a great deal of

time explaining various economic concepts which provided this work with a perspective

that complemented the epidemiologic focus and resulted in a well-rounded study.

Lastly, I’d like to thank my family (there are too many of you to list here, you

know who you are) for believing in me and pushing me to always do my best. Love you

all.

v

Funding and legal disclaimers

Data acquisition and compilation was partially supported by U.S. Department of

Health and Human Services. National Institutes of Health (NIH), P60 MD002249-05

(National Institute of Minority Health and Health Disparities) and R01 HL071759

(National Heart, Lung, and Blood Institute [NHLBI]). Funding for the MESA parent

study came from NIH NHLBI contracts: HHSN268201500003I, N01-HC-95159 through

95169, UL1-TR-000040 and UL1-TR-001079.

Mention of trade names, commercial practices, or organizations does not imply

endorsement by the authors, the institutions where the authors work, nor by the funding

entities. The authors take sole responsibility for all data analyses, interpretation, and

views expressed in this paper. Any errors in the manuscript are the sole responsibility of

the authors, not of Information Resources Inc. (IRI, who supplied the pricing dataset).

vi

TABLE OF CONTENTS

CHAPTER 1: INTRODUCTION .................................................................................................................... 1

1.1 INTRODUCTION ................................................................................................................... 1

1.2 DIABETES ............................................................................................................................. 2

1.3 DIET QUALITY ...................................................................................................................... 5

1.4 ACCESS TO HEALTHY FOODS ............................................................................................... 9

1.5 FOOD PRICES ..................................................................................................................... 10

1.6 PRIOR WORK EXAMINING FOOD PRICES AND HEALTH OUTCOMES .................................. 15

1.7 SPECIFIC AIMS AND INNOVATION ..................................................................................... 18

CHAPTER 2: THE VARIATION OF HEALTHY AND UNHEALTHY FOOD PRICES ACROSS NEIGHBORHOODS AND THEIR ASSOCIATION WITH NEIGHBORHOOD SOCIOECONOMIC STATUS AND PROPORTION BLACK/HISPANIC .................................................................................. 20

2.1 INTRODUCTION ................................................................................................................. 22

2.2 METHODS ......................................................................................................................... 23

2.3 RESULTS ............................................................................................................................ 30

2.4 DISCUSSION ...................................................................................................................... 33

2.5 CONCLUSIONS ................................................................................................................... 37

CHAPTER 3: THE PRICE OF HEALTHY FOOD RELATIVE TO UNHEALTHY FOOD AND ITS ASSOCIATION WITH DIET QUALITY: THE MULTI-ETHNIC STUDY OF ATHEROSCLEROSIS ............................... 45

3.1 INTRODUCTION ................................................................................................................. 47

3.2 METHODS ......................................................................................................................... 48

3.3 RESULTS ............................................................................................................................ 55

3.4 DISCUSSION ...................................................................................................................... 58

3.5 CONCLUSIONS ................................................................................................................... 61

CHAPTER 4: THE PRICE OF HEALTHY FOOD RELATIVE TO UNHEALTHY FOOD AND ITS ASSOCIATION WITH TYPE 2 DIABETES AND INSULIN RESISTANCE: THE MULTI-ETHNIC STUDY OF ATHEROSCLEROSIS ...................................................................................................... 68

4.1 INTRODUCTION ................................................................................................................. 70

vii

4.2 METHODS ......................................................................................................................... 71

4.3 RESULTS ............................................................................................................................ 79

4.4 DISCUSSION ...................................................................................................................... 80

4.5 CONCLUSIONS ................................................................................................................... 84

CHAPTER 5: DISCUSSION ........................................................................................................................ 90

5.1 DISCUSSION ...................................................................................................................... 90

5.2 CONTEXT OF CURRENT RESEARCH WITH PREVIOUS WORK .............................................. 93

5.3 LIMITATIONS ..................................................................................................................... 97

5.4 CONCLUSIONS ................................................................................................................. 100

LIST OF REFERENCES ............................................................................................................................. 103

APPENDIX 1: FOR CHAPTER 2 ............................................................................................................... 123

APPENDIX 1.1 CHOICE OF PRODUCTS IN THE IRI DATASET .................................................................... 123

APPENDIX 1.2 DETAILS ON HEALTHY AND UNHEALTHY PRODUCTS INCLUDED IN THE PRICE CALCULATIONS ..... 126

APPENDIX 1.3 PRINCIPAL COMPONENT ANALYSIS ............................................................................... 133

APPENDIX 1.4 DISTRIBUTION OF THE HEALTHY-TO-UNHEALTHY PRICE RATIO ........................................... 135

APPENDIX 1.5 VALIDATION OF THE IRI PRICE DATASET AGAINST C2ER .................................................. 136

APPENDIX 1.6 ADDITIONAL SUPPLEMENTAL TABLES FOR CHAPTER 2 ...................................................... 138

APPENDIX 2: FOR CHAPTER 3 ............................................................................................................... 146

APPENDIX 2.1 SUPPLEMENTAL TABLES FOR CHAPTER 3 ....................................................................... 146

APPENDIX 3: FOR CHAPTER 4 ............................................................................................................... 150

APPENDIX 3.1 SUPPLEMENTAL TABLES FOR CHAPTER 4 ....................................................................... 150

APPENDIX 3.2 HOMA-IR VALIDITY ................................................................................................ 154

VITA ................................................................................................................................. 155

viii

LIST OF TABLES

TABLE 2-1 NEIGHBORHOOD DEMOGRAPHICS AND FOOD PRICES FOR STUDY SAMPLE ................... 39

TABLE 2-2 PRICE OF FOODS BY DEMOGRAPHIC CHARACTERISTICS .................................................. 41

TABLE 2-3 VARIANCE DECOMPOSITION BY GEOGRAPHIC AREA BEFORE AND AFTER ADJUSTING FOR REGION, URBANICITY, AND TOILET PAPER PRICE ............................................................. 42

TABLE 2-4 HIERARCHICAL MODEL ESTIMATES REGRESSING PRICE OUTCOMES ON NEIGHBORHOOD SOCIOECONOMIC STATUS (SES) AND THE PROPORTION OF INDIVIDUALS WHO ARE HISPANIC OR BLACK, NESTED WITHIN COUNTY AND STATE, N = 1953 ............................ 43

TABLE 3-1 CHARACTERISTICS OF THE INDIVIDUALS INCLUDED IN THE ANALYSIS BY TERTILES OF HEALTHY-TO-UNHEALTHY FOOD PRICE RATIO (N=2,642) ................................................ 62

TABLE 3-2 PROPORTION OF PARTICIPANTS WITH A HIGH QUALITY DIET BY TERTILE OF THE HEALTHY-TO-UNHEALTHY PRICE RATIO AND THE AVERAGE SERVING PRICE OF HEALTHY AND UNHEALTHY FOODS (N=2,765) ................................................................................. 64

TABLE 3-3 ODDS RATIOS OF HAVING A HIGH QUALITY DIET ASSOCIATED WITH THE PRICE RATIO, AND WITH THE PRICES OF HEALTHY FOODS AND UNHEALTHY FOODS AFTER SEQUENTIAL ADJUSTMENT FOR CONFOUNDERS WITHIN THE FULL POPULATION (N=2,765) AND AFTER REMOVING NEW YORK/BRONX (N=2,225) ................................... 65

TABLE 3-4 ODDS RATIOS OF HAVING A HIGH QUALITY DIET ASSOCIATED WITH THE PRICE RATIO, AND WITH THE PRICES OF HEALTHY FOODS AND UNHEALTHY FOODS, STRATIFIED BY INCOME-WEALTH AND BY EDUCATION (N=2,765) ........................................................... 66

TABLE 3-5 INSTRUMENTAL VARIABLE ANALYSIS RESULTS USING TOILET PAPER AS THE INSTRUMENT AND TWO-STAGE RESIDUAL INCLUSION MODELS FOR THE HEALTHY-TO-UNHEALTHY PRICE RATIO, HEALTHY FOOD PRICE, AND UNHEALTHY FOOD PRICE (N=2,765) A ........................................................................................................................ 67

TABLE 4-1 MESA PARTICIPANT CHARACTERISTICS BY HEALTHY-TO-UNHEALTHY PRICE RATIO AT THE TIME OF EXAM 5 (N=3,408) .............................................................................................. 85

TABLE 4-2 PROPORTION OF PARTICIPANTS WITH TYPE 2 DIABETES, INCIDENT TYPE 2 DIABETES, AND INSULIN RESISTANCE LEVELS BY TERTILE OF THE HEALTHY-TO-UNHEALTHY PRICE RATIO AND THE AVERAGE SERVING PRICE OF HEALTHY AND UNHEALTHY FOODS ......... 87

TABLE 4-3 MULTIVARIABLE MODELS FOR EACH OF THE THREE OUTCOMES OF INTEREST: PREVALENCE OF TYPE 2 DIABETES AT EXAM 5, INCIDENCE OF TYPE 2 DIABETES FROM EXAM 4 TO EXAM 5, AND LEVEL OF INSULIN RESISTANCE AT EXAM 5 A .......................... 88

SUPPLEMENTAL TABLE 1 ROTATED COMPONENT MATRIX FROM THE PRINCIPAL COMPONENT ANALYSIS OF HEALTHY FOODS ILLUSTRATING THE COMPONENT LOADING SCORES OF EACH ITEM FOR DAIRY (MILK, YOGURT AND COTTAGE CHEESE) AND FRUITS AND VEGETABLES (ORANGE JUICE AND FROZEN VEGETABLES) ............................................. 134

ix

SUPPLEMENTAL TABLE 2 ROTATED COMPONENT MATRIX FROM THE PRINCIPAL COMPONENT ANALYSIS OF UNHEALTHY FOODS ILLUSTRATING THE COMPONENT LOADING SCORES OF EACH ITEM FOR SODA, SWEETS (CHOCOLATE AND COOKIES) AND SALTY SNACKS ....... 134

SUPPLEMENTAL TABLE 3 SPEARMAN RANK CORRELATIONS FOR PRODUCTS FOUND IN THE IRI DATASET AND C2ER ........................................................................................................................ 136

SUPPLEMENTAL TABLE 4 PRODUCT SELECTION CRITERIA FOR EACH OF THE FOOD CATEGORIES USED IN THIS STUDY ..................................................................................................................... 138

SUPPLEMENTAL TABLE 5 NUTRITION INFORMATION FOR EACH OF THE HEALTHY AND UNHEALTHY FOOD CATEGORIES INCLUDED IN THIS STUDY ................................................................ 139

SUPPLEMENTAL TABLE 6 SERVING SIZES AND WEIGHTS FOR COMPOSITE PRICE CALCULATIONS USING TWO DIFFERENT WEIGHT CALCULATIONS – BASED ON NATIONAL CONSUMPTION AVERAGES AND USING EQUAL WEIGHTS WITHIN EACH FOOD CLASS ........................... 140

SUPPLEMENTAL TABLE 7 FOOD PRICES ACCORDING TO WEIGHTS BASED ON NATIONAL CONSUMPTION AVERAGES AND EQUAL WEIGHTS ................................................................................... 141

SUPPLEMENTAL TABLE 8 PRICE OF EACH PRODUCT CATEGORY BY DEMOGRAPHIC CHARACTERISTICS142

SUPPLEMENTAL TABLE 9 SENSITIVITY ANALYSIS OF THE HIERARCHICAL MODELS EXAMINING PRICE OUTCOMES ON NEIGHBORHOOD SOCIOECONOMIC STATUS (SES) AND THE PROPORTION OF INDIVIDUALS WHO ARE HISPANIC OR BLACK USING A 2-MILE BUFFER 144

SUPPLEMENTAL TABLE 10 CHARACTERISTICS OF INCLUDED AND EXCLUDED INDIVIDUALS IN THE ANALYSIS OF DIET QUALITY ............................................................................................ 146

SUPPLEMENTAL TABLE 11 RESULTS OF SENSITIVITY ANALYSES USING EQUAL WEIGHTS FOR THE PRICE OUTCOMES, USING A 5-MILE RADIUS TO CAPTURE PRICES FOR ALL INDIVIDUALS, AND USING A 1-MILE RADIUS FOR THOSE LIVING IN NEW YORK CITY AND A 3-MILE RADIUS FOR ALL OTHERS ............................................................................................................. 148

SUPPLEMENTAL TABLE 12 PATIENT CHARACTERISTICS OF THOSE WHO WERE INCLUDED IN THE STUDY (N=3,408) AND THOSE WHO WERE EXCLUDED (N=1,308) ............................................. 150

SUPPLEMENTAL TABLE 13 ANALYSIS OF INTERACTIONS BETWEEN SES AND RACE WITH THE PRICE RATIO FOR EACH OF THE THREE OUTCOMES OF INTEREST A ......................................... 152

x

LIST OF FIGURES

FIGURE 1 DISTRIBUTION OF THE HEALTH-TO-UNHEALTHY PRICE RATIO FOR THE 1953 STORES IN THE IRI DATABASE. .......................................................................................................... 135

xi

ABSTRACT

The price of unhealthy food relative to healthy food and its association with diet quality, diabetes, and insulin resistance in a multi-ethnic population

David Michael Kern

OBJECTIVE: This dissertation first evaluates food price variation within and between

neighborhoods in order to improve our understanding of access to healthy foods and

potential economic incentives and barriers to consuming a higher quality diet. The study

then spatially links individuals to their nearby supermarkets to study the association

between food price and dietary quality, insulin resistance (IR), and diabetes.

METHODS: Prices of healthy foods (dairy, fruits, and vegetables) and unhealthy foods

(soda, sweets, and salty snacks) were obtained from 1953 supermarkets across the US

during 2009-2012 from the Information Resources Inc. (IRI) database. In Aim 1, prices

of healthy and unhealthy foods, and the relative price of healthy foods compared with

unhealthy foods (healthy-to-unhealthy price ratio) were linked to census block group

socio-demographics in order to analyze associations between food prices with

neighborhood SES and proportion Black/Hispanic. Linear hierarchical regression models

were used to explore geospatial variation and adjust for confounders. The second and

third aims of this study linked average price of healthy foods, unhealthy foods and their

ratio to participants in The Multi-Ethnic Study of Atherosclerosis (MESA). For Aim 2,

individuals who completed MESA exam 5 (2010-2012) and the food frequency

questionnaire were included (N=2765). A Healthy Eating Index (HEI-2010) was

calculated for each individual according to their FFQ. Logistic regression estimated

xii

adjusted odds ratios of a high quality diet (top quantile of HEI-2010) associated with each

price exposure. Sensitivity analyses used an instrumental variable approach in which the

price of brand name toilet paper served as an instrument for food prices. For Aim 3,

individuals from MESA who completed exam 5 and exam 4 (administered five years

prior, diabetes incidence analysis only) were included. Type 2 diabetes status was

confirmed at each exam and IR was measured according to the homeostasis model

assessment index of IR. Adjusted logistic, modified Poisson and linear regression models

were used to model diabetes prevalence, incidence and IR, respectively.

RESULTS: Overall, the price of healthy foods was nearly twice as high as the price of

unhealthy foods ($0.590 vs. $0.298 per serving; healthy-to-unhealthy price ratio of 1.99).

This trend was consistent across all neighborhood characteristics. After adjusting for

covariates, no association was found between food prices (healthy, unhealthy, or the

healthy-to-unhealthy ratio) and neighborhood SES, while small positive and negative

associations were detected with the proportion Black/Hispanic, respectively.

A larger healthy-to-unhealthy price ratio was associated with lower odds of having a high

quality diet (Odds Ratio [OR]=0.76 per SD increase in the ratio, 95% CI=[0.64 to 0.91]).

Instrumental variable analyses largely confirmed these findings although confidence

intervals were wider and the result was no longer statistically significant (OR=0.82 [0.57

to 1.19]). A higher ratio of healthy-to-unhealthy food had a positive association with IR

(4.8% increase in IR for each standard deviation increase in price ratio (estimate=0.048,

95% CI=[0.00 to 0.10]) after adjusting for confounders. No association with diabetes

xiii

incidence (relative risk=1.11, 95% CI=[0.85 to 1.44]) or prevalence (OR=0.95, 95%

CI=[0.81 to 1.11]) was observed.

CONCLUSIONS: The price of healthy food was twice as expensive as unhealthy food

per serving on average. A higher price of healthy food relative to unhealthy foods appears

to be negatively associated with a high quality diet and with insulin resistance. There did

not appear to be an association with diabetes prevalence or 5-year incidence. This study

provides new insight into the relationship between food prices with diet quality, IR and

diabetes. Policies to address the large price differences between healthy and unhealthy

foods may help improve diet quality and downstream health effects in the U.S.

1

CHAPTER 1: INTRODUCTION

1.1 INTRODUCTION

Affordability may play an important role in determining whether individuals truly

have access to healthy foods. The price of food influences the purchasing decisions of

consumers (Andreyeva et al., 2010), and there is mounting evidence that healthy food is

significantly more expensive than unhealthy alternatives (Rao et al., 2013). This

discrepancy in price may be influencing consumers to purchase a larger quantity of

unhealthy foods than is recommended by dietary guidelines, especially those of lower

socioeconomic status whom are more price sensitive.

Diet plays an important role in one’s health, as individuals with a healthy diet are

less likely to be obese (Guo et al., 2004) or develop diabetes (Chiuve et al., 2012), along

with an array of other chronic conditions. Thus, the availability of healthy foods may be

critical in preventing downstream, debilitating health complications. Currently, 29

million individuals in the United States are living with diabetes – a condition that comes

with a major risk of severe complications and death (Kochanek et al., 2016). Reducing

the incidence of this disease can have enormous effects on our healthcare system,

including a reduction in mortality, morbidity of other chronic conditions, and economic

burden.

Studies have found that county and metropolitan level food prices are associated

with BMI and obesity; however, prior work has not examined neighborhood level prices

and their association with diet quality and diabetes. Because there is large variability in

neighborhood characteristics within a single metropolitan area, it is important to account

for this when examining the association between food price and health. This dissertation

2

will address the gaps of prior work by examining food prices local to each individual in

the study, include a variety of healthy and unhealthy foods, and evaluate the association

of prices with various endpoints along a clinical pathway (diet quality, insulin resistance,

and diabetes).

1.2 DIABETES

1.2.1 Epidemiology and trends of diabetes

The CDC estimates that nearly 2 million Americans, typically middle-age and

older adults, are newly diagnosed with diabetes every year and more than 29 million

individuals in the United States are currently living with diabetes as of 2012, accounting

for 9.3% of the population(Centers for Disease Control and Prevention, 2014). Of those

with diabetes more than a quarter (8.1 million) are undiagnosed, and therefore untreated

by a physician. The prevalence of diabetes is growing at an alarming rate with a 2.6%

increase in the prevalence of diagnosed diabetes since 2000(Centers for Disease Control

and Prevention, 2016).

The prevalence of diagnosed diabetes occurs more often and earlier for African-

Americans (13.2% prevalence, median age of diagnosis 49 years) and Hispanics (12.8%,

49 years), compared with Whites (7.6%, 55 years) (Centers for Disease Control and

Prevention, 2014, Centers for Disease Control and Prevention, 2013); and the incidence

of diabetes is increasing at a faster rate within Black and Hispanic individuals than it is

for Whites(Geiss et al., 2014), leading to even further divergences in the disease burden

between races. There is also a clear association between education level and diabetes

prevalence. Those with less than a high school education have much higher rates of

3

diagnosed diabetes (12.9%) compared to those with a high school diploma (9.5%) and

those with some college or more (6.7%) (Centers for Disease Control and Prevention,

2015), and the rate of the increase in prevalence is higher in those with a high school

degree or less compared with those who have more than a high school education (Geiss et

al., 2014).

1.2.2 Downstream effects of diabetes

Reducing the incidence of diabetes is important because of the complications that

result from the disease. The risk of cardiovascular disease, including stroke, is the most

prominent complication of diabetes and accounts for the largest proportion of morbidity,

mortality, and costs due to diabetes (Go and et al, 2014, American Diabetes Association,

2013), and nearly 65% of people with diabetes in the United States die from

cardiovascular disease (Donahoe et al., 2007). Other serious complications of diabetes

include neuropathy (Boulton et al., 2005), nephropathy (Mogensen et al., 1983), foot

complications especially due to peripheral arterial disease which can result in amputation

(Jude et al., 2001), eye complications (diabetic retinopathy)(Wilkinson et al., 2003) and

dyslipidemia(American Diabetes Association, 2013, Centers for Disease Control and

Prevention, 2014).

Due to such complications, diabetes was the seventh leading cause of death in the

United States in 2010 with a death rate of 20.8 per 100,000 persons (Murphy et al.,

2013). Overall, due to both direct and indirect costs resulting from morbidity and

mortality, diabetes was estimated to result in $245 billion in costs in 2012, and patients

4

with diabetes incurring more than double the average costs of patients without the disease

(Centers for Disease Control and Prevention, 2014).

Reducing the rate of diabetes through prevention of disease would have a

significant impact on morbidity, mortality, and economic healthcare burden in our

country.

1.2.3 Diabetes risk factors

Risk factors of diabetes include race/ethnicity (African American, Latino, Native

American, Asian American, Pacific Islander), familial history of diabetes, hypertension,

prediabetes, low HDL and/or high triglycerides levels, low socioeconomic status and a

history of cardiovascular disease (American Diabetes Association, 2013, Robbins et al.,

2001); however, the largest risk factor of diabetes is being overweight or obese. Large

cohort studies have found that those who are obese are at a much greater risk of

developing diabetes(Abdullah et al., 2010, Hu et al., 2001), and that an individual’s

waist/hip ratio is an important risk factor of diabetes (Vazquez et al., 2007, Janiszewski et

al., 2007). Because the disease is mainly driven by obesity, type 2 diabetes is thought to

be largely preventable through lifestyle modifications, especially diet and exercise

(Diabetes Prevention Program Research Group, 2002, Tuomilehto et al., 2001), and life

style interventions that emphasize the importance of a healthy diet and exercise have

shown to significantly reduce the risk of developing diabetes (Diabetes Prevention

Program Research Group, 2002, Tuomilehto et al., 2001). The importance of a healthy

diet on reducing the risk of diabetes is more than just preventing weight gain, as part of

5

the association between the healthfulness of a diet and diabetes is independent of weight

(de Koning et al., 2011a, Hu et al., 2001).

1.2.4 Insulin resistance

A precursor to diabetes is a condition known as insulin resistance. For an

individual with insulin resistance, muscle, fat, and liver cells fail to respond properly to

the presence of insulin resulting in the inability to properly absorb blood glucose

(Reaven, 1995). To compensate for the inability of the cells to efficiently absorb glucose,

the beta cells in the pancreas produce more insulin; however, over time the body is

unable to produce enough insulin to aid absorption of the glucose resulting in increased

blood sugar levels (Reaven, 1995). The increasing levels of glucose in the blood stream

eventually results in pre-diabetes and diabetes. Preventing insulin resistance, as well as

identifying and treating the condition after it occurs, can help prevent the onset of type 2

diabetes. Many of the same risk factors of diabetes are also risk factors of insulin

resistance, with the major cause thought to be excess intra-abdominal fat due to lipids

accumulating within insulin responsive tissues (Samuel and Shulman, 2012, Kahn and

Flier, 2000). As with diabetes, diet plays an important role in the development of insulin

resistance and so understanding factors associated with poor diet quality is an important

step in preventing the condition.

1.3 DIET QUALITY

1.3.1 Characteristics of the American diet

The quality of the average diet in the US is well below recommended dietary

guidelines due to low fruit and vegetable consumption and high intake of added sugars,

6

saturated fats, and sodium (United States Department of Agriculture and United States

Department of Health and Human Services, 2015). Eating habits in the United States

have changed significantly over the last three decades, with an increased frequency of

snacking on desserts, salty snacks, and sweetened beverages, among other highly

processed foods with low nutrient value (Piernas and Popkin, 2009). Since 2000 in the

US, total fat, especially those from salad and cooking oils, have risen dramatically, and

there has been a significant downward trend in the number of fruits and vegetables

consumed (United States Census Bureau, 2012). Additionally, the consumption of milk, a

healthier alternative to soda and other sweetened beverages, has been on the decline for a

number of years (Stewart et al., 2013, Krebs, 2013).

1.3.2 Measure of overall diet quality

A number of dietary guidelines exist, each with their own definition of what

constitutes a healthy diet. Such examples include the Mediterranean Diet Scale

(Trichopoulou et al., 2003, Trichopoulou et al., 1995), the Healthy Food Index (Osler et

al., 2002), the Dietary Approaches to Stop Hypertension (DASH) (Appel et al., 1997), the

Healthy Eating Index (HEI) (Kennedy et al., 1995, Guenther et al., 2013) and the

alternative HEI (Chiuve et al., 2012), as well as empirically derived dietary measures

(Nettleton et al., 2006, Nettleton et al., 2008b, Fung et al., 2001).

The HEI was built to reflect the USDA’s recommendations for a healthy diet and

is in turn used by the USDA to monitor and assess diet quality in the US (Kennedy et al.,

1995), with the most recent version based on the 2010 USDA dietary guidelines (United

States Department of Agriculture and United States Department of Health and Human

7

Services, 2010, Guenther et al., 2013). The 2010 USDA dietary guidelines were designed

with specific regard to the growing obesity epidemic (United States Department of

Agriculture and United States Department of Health and Human Services, 2010), a

condition that puts individuals at high risk of diabetes and cardiovascular disease, among

many other chronic medical conditions (Grundy, 2002, Kopelman, 2000, Nelson et al.,

1988, Sowers, 1998, Sturm, 2002, Wilson et al., 2002). Thus, adherence to a diet that

would have a HEI score is expected to be associated with a decreased risk of diabetes and

the wide array of resulting chronic diseases (Guo et al., 2004).

Because the HEI is focused on the type of diet that is associated with a wide range

of chronic diseases rather than any specific disease area, and because it is endorsed by the

USDA as measuring the prototypical healthy diet, it was used in this research to measure

overall diet quality.

1.3.3 Diet quality and health

The importance of a healthy diet, and what constitutes a healthy diet, has been

studied extensively, and diet quality has been linked to a number of health outcomes. A

healthy diet, as measured by the Healthy Eating Index (HEI), has been found to be

inversely associated with obesity (Guo et al., 2004), waist circumference (Tande et al.,

2010), diabetes (Chiuve et al., 2012), cardiovascular disease (Chiuve et al., 2012,

McCullough et al., 2000), stroke (Chiuve et al., 2012), cancer (Arem et al., 2013, Li et

al., 2014) and mortality (Rathod et al., 2012, Kappeler et al., 2013, Harmon et al., 2015,

Reedy et al., 2014). Unfortunately, the average consumer purchases foods that result in a

diet that falls well short of the recommendations in the USDA dietary guidelines (Volpe

8

and Okrent, 2012, United States Department of Agriculture, 2015). This is due in part to

purchasing too few fruits and vegetables and too many fats, sugars, and sweets. With a

large proportion of the American population failing to achieve a healthy diet, it is

important to understand the causes of the worsening diet quality in this country and

develop interventions to reverse this trend in order to reduce the prevalence of obesity,

diabetes, and all other health complications resulting from poor diets.

1.3.4 Diet quality and SES

Diet quality is associated race and socioeconomic status (Beydoun and Wang,

2008). African Americans – independent of income – have been shown to have an overall

poorer diet compared with all other races, as measured by the Healthy Eating Index, as

are those with a lower family income – independent of race – compared with individuals

with a higher income (Forshee and Storey, 2006). The association between SES and diet

quality has appeared to have gotten stronger over time with a widening gap in diet quality

between those of high and low SES (Wang et al., 2014). In addition to the overall diet,

lower individual and neighborhood SES (Ball et al., 2006, De Irala-Estevez et al., 2000,

Dubowitz et al., 2008) and race (Dubowitz et al., 2008) have been found to be associated

with decreased consumption of fruits and vegetables. Unsurprisingly, these risk factors of

poor diet are consistent with the risks of diabetes; i.e., those that are of lower

socioeconomic status and racial minorities are more likely to have poor diets and

diabetes.

9

1.4 ACCESS TO HEALTHY FOODS

1.4.1 Physical access to healthy foods

Throughout the United States there are individuals who do not have access to

foods that constitute a healthy diet. A large portion of food access research has focused

on food deserts and the lack of physical access to fresh fruits and vegetables and

supermarkets (Cummins and Macintyre, 2002, Walker et al., 2010). The importance of

increasing physical access to healthy foods in food deserts, such as through the

construction of supermarkets, stems on the hypothesis that improving physical access

allows more individuals the ability to purchase nutritious foods which were previously

unavailable. However, the impact of such interventions has been mixed. While early

work showed a correlation between supermarkets and obesity and diet quality (Morland

et al., 2006, Powell et al., 2007b, Rose and Richards, 2004, Black and Macinko, 2008), a

handful of more recent studies, including longitudinal and interventional designs, have

found no or minimal effect of supermarket access on diet quality or risk of obesity

(Boone-Heinonen et al., 2011, Cummins et al., 2014, Elbel et al., 2015, Ford and

Dzewaltowski, 2010). Thus, when thinking about access to healthy food, it is important

to not only consider physical proximity, but also consider other factors such as the

affordability of those foods.

1.4.2 Economic access and affordability of healthy foods

Economic access – or affordability – may play an important role in determining

whether healthy foods are truly accessible to individuals in a neighborhood. If healthy

food is physically available in the neighborhood, but many are unable to afford it, then

10

access is still poor; this situation is referred to as a food mirage (Breyer and Voss-

Andreae, 2013). The inability for individuals to afford sufficient food has resulted in

food insecurity for more than 14% of households in the US, and those most likely to be

food insecure are African Americans and those of low SES (Coleman-Jensen et al.). For

individuals whose main concern regarding access to food is to avoid hunger, and for

whom nutritional concerns are secondary, price is a major consideration. If unhealthy

food is more affordable than healthy food – even if slightly – it would be likely to follow

that food purchases would tend to be more unhealthy and the resulting diet would be

poorer (Bernstein et al., 2010, Lopez et al., 2009, Rehm et al., 2011, Ryden and Hagfors,

2011, Schroder et al., 2006), and the risk for chronic health problems would increase

(Guo et al., 2004, Chiuve et al., 2012, Li et al., 2014).

Differential availability across sociodemographic factors may also contribute to

disparities in intake, especially for individuals of lower SES who are more sensitive to

food prices (Powell et al., 2009b) due to a higher proportion of income spent on food

(Frazao et al., 2007). Black and Hispanic populations often have lower SES, thus price

differentials could be contributing to their increased risk of obesity and diabetes (Ogden

et al., 2014).

1.5 FOOD PRICES

1.5.1 Diet costs

The poor diet quality of many individuals in our country may be due in part to

purchasing too few fresh fruits and vegetables and instead purchasing more affordable

processed fats, sugars, and sweets. Low cost diets are associated with higher calorie

11

intake at the expense of fewer nutrients(Andrieu et al., 2006), while healthier diets tend to

be more expensive (Rehm et al., 2011, Ryden and Hagfors, 2011, Schroder et al., 2006,

Rehm et al., 2015). A meta-analysis found that, on average, the healthiest diets cost $1.48

more per day compared with the least healthy (Rao et al., 2013). Of the 8 measures of

overall dietary quality reported, across 6 different studies(Bernstein et al., 2010, Lopez et

al., 2009, Rehm et al., 2011, Ryden et al., 2008, Ryden and Hagfors, 2011, Schroder et

al., 2006) all but one(Ryden et al., 2008) found that the healthiest diets were significantly

more expensive than the least healthy.

Furthermore, differences in nutrient intake by individual SES can be explained

primarily by the cost of a diet (Monsivais et al., 2012, Konttinen et al., 2013), and a

systematic literature review found that individuals of lower SES tended to select food that

was of lower nutritional value and diets that were of lower-quality due to the lower price

of these foods (Darmon and Drewnowski, 2015). In fact, the cost of a healthy diet is

beyond the allotted food budget for many individuals in poverty whom already devote a

significantly greater proportion of their income to food compared with those of higher

SES (Darmon and Drewnowski, 2015, Drewnowski and Specter, 2004), and has been

estimated that up to 70% of a low-income family’s food budget would need to be devoted

to fruits and vegetables to meet recommended dietary guidelines (Cassady et al., 2007).

The higher price of healthy diets compared with more affordable unhealthy diets

is due to differences in the price of individual healthy and unhealthy foods which

comprise those diets. Food prices have been measured in a variety of ways, including the

price per calorie (Drewnowski, 2009, Andrieu et al., 2006, Drewnowski and Darmon,

2005), the price per unit of edible food weight (e.g., price per gram or ounce) (Todd et

12

al., June 2011, Lipsky, 2009), or the price per serving (Lipsky, 2009, Reed et al., 2004,

Stewart et al., February 2011, Kern et al., 2016). Depending on how price of food is

operationalized, healthier foods have been found to be more expensive than unhealthy

foods (Drewnowski, 2010, Drewnowski and Darmon, 2005, Lipsky, 2009, Andrieu et al.,

2006, Monsivais et al., 2010, Kern et al., 2016). The price of a calorie is substantially

cheaper when obtained from unhealthful, energy-dense foods, such as soda, instead of

from more healthful, less-dense foods. For example, while sugar and oil cost less than $1

per 10 MJ (2,388 kCal), milk costs roughly five times as much, ground beef nearly 10

times as much, and vegetables are $30-$100 for the same amount of energy (Drewnowski

and Darmon, 2005). However, when considering the price per serving, the cost of healthy

and healthy foods are closer, though healthy foods, specifically fruits and vegetables,

remain more expensive then unhealthy foods (Lipsky, 2009, Kern et al., 2016).

1.5.2 Price measurement

This study used the price per serving as the primary unit of analysis rather than

the price per calorie or per unit of weight. Using price per calorie has been criticized

because consumers don’t have information regarding the price per calorie immediately

available to them (Lipsky, 2009). A person would have to calculate the total number of

calories in a package and divide by the product price, and it isn’t likely that this

measurement would be responsible for the majority of purchasing decisions. And while

price per gram or ounce is useful when the foods being compared have similar forms and

serving sizes, the comparison becomes more difficult to interpret when the types of foods

(e.g., soft drinks versus produce) and/or the serving sizes differ substantially (Carlson and

Frazão, May 2012). Thus, we chose the unit of analysis that could be compared across all

13

product types, and which may be most meaningful to consumers: price per serving, which

has been used previously in similar research (Stewart et al., February 2011, Lipsky, 2009,

Reed et al., 2004).

1.5.3 Price elasticity of foods

To measure the effect of pricing differences on buying patterns, the price

elasticity of demand (or simply “price elasticity”) of foods has been studied. Price

elasticity refers to the proportional change in demand (i.e., purchasing) when the price of

a product is changed 1%. Inelasticity is defined as a product having price elasticity <1.0

(i.e., a 1% change in price leads to a <1% change in purchases), while elasticity is when

the price elasticity is >1.0 (i.e., a 1% change in price leads to a >1% change in

purchases).

In a thorough review of the literature, Andreyeva et al summarize the price

elasticity for a variety of food types (Andreyeva et al., 2010). They found the price of soft

drinks was modestly elastic at 0.79, implying that a 1% increase in the price of soda

would result in a 0.79% decrease in purchases. The result was similar for juice (0.76) and

fruit (0.70), while more inelastic for milk (0.59) and vegetables (0.58). This is consistent

with economic and marketing research which has shown that the price of food – in

addition to taste, nutrition, convenience, and other factors – affects the purchasing

choices of individuals who must navigate environments with numerous choices and

inundated with advertising(Connors et al., 2001, Glanz et al., 1998, Andreyeva et al.,

2010, Aggarwal et al., 2016). Another study examined the effect of soda taxes on a state

wide basis and found that when the price of soft drinks increased one percentage point

14

consumption of soda decreases, and that there is an increase in whole milk consumption

(Fletcher et al., 2010b). This effect is indicative of cross product elasticity in which the

price of one product influences the demand of a potential substitute product. Because of

this cross-product elasticity, as the price of unhealthy foods relative to healthy foods

becomes larger we’d expect to see lower consumption of unhealthy foods and an increase

in consumption of healthy foods, leading to better overall diet quality.

And while it appears food costs are not strongly elastic, it has been found that

increased prices would have the largest effect in decreasing consumption in populations

that are most at risk for being overweight and/or obese (Powell and Chaloupka, 2009,

Powell et al., 2009b). Because a greater proportion of income for lower income

individuals is spent on food (Frazao et al., 2007), they are more price elastic and would

be more influenced by price differences between exchangeable products (Powell and

Chaloupka, 2009). Thus, those at highest risk of diabetes due to a poor diet and high

weight appear to be those whose diet would be most influenced by changes in food

prices.

The potential effect of food prices on consumption is important because healthier

food appears to be more expensive than unhealthy food, and the consumption of healthy

food can be influenced by both its own price and the price of unhealthy food. If there are

price differences across neighborhoods, where unhealthier food is cheaper than healthy

food compared with other areas, then the residents of that neighborhood may be put at an

increased risk of obesity, diabetes, cardiovascular disease, and many other chronic

conditions.

15

1.6 PRIOR WORK EXAMINING FOOD PRICES AND HEALTH OUTCOMES

A previous paper examining the variation in healthy and unhealthy food prices

was limited to soda and milk only (Kern et al., 2016). The current study expands upon

previous research of healthy and unhealthy beverages to include proxies for fresh fruits

and vegetables as well as foods high in sugar, fat, sodium and preservatives. Like the

previous study, the current study integrated the prices of various healthy and unhealthy

foods in supermarkets across the US and measured their association with neighborhood

SES and racial makeup. The current study goes beyond the scope of the prior research by

examining the association of food prices with diet quality, diabetes, and insulin

resistance.

Research on the effect of changes in food price on health has been growing in

response to the bevy of proposed taxes on soda and other sugar sweetened beverages.

These studies have largely been ecologic in scope (looking at regional tax levels of

unhealthy foods and the rates of obesity in those same areas) or have used economic

modeling assuming specific elasticity measurements to estimate the effect of price

changes on consumption and resulting BMI. In large part, these studies have argued that

price has a modest effect on health outcomes. This is likely due to small taxes that have

been implemented or proposed (hypothesized that larger tax differences would show a

more pronounced effect (Fletcher et al., 2010a, Brownell et al., 2009)) and that a single

product – even when it is a highly consumed product -- is unlikely to have a very large

effect (Finkelstein et al., 2014). Research has concluded soda taxes may have a very

modest effect in decreasing BMI (Powell et al., 2009a, Smith et al., 2010), but could have

16

a significant impact on reducing diabetes incidence and resulting medical costs in the

country as a whole (Wang et al., 2012).

Observational research examining actual food prices and individual level

outcomes, rather than experimental or simulation studies, have typically used food prices

at the county or metropolitan level. In these studies, the price of fruits and vegetables has

primarily been found to be weakly positively associated with BMI, while other food

prices (e.g., fast food, food at home, food in restaurants) have been found to be weakly

associated with weight outcomes (Beydoun et al., 2008, Chou et al., 2004, Han and

Powell, 2011, Sturm and Datar, 2008, Powell et al., 2007a, Powell and Chaloupka, 2009,

Powell et al., 2013). Evidence of the association between food price and levels of insulin

resistance has been found to be weak (Rummo et al., 2015), may exist only within those

in the middle income range and lower education (Meyer et al., 2014), or exists only for

specific unhealthy foods (Duffey et al., 2010). Studies examining NHANES data provide

evidence that higher prices of healthy and low glycemic foods are associated with higher

blood sugar levels in those with diabetes (Anekwe and Rahkovsky, 2014) and without

(Rashad, 2007). While one prior study examined the hypothetical effect taxes may have

on diabetes prevalence by simulating specific scenarios and assumptions (Wang et al.,

2012), to our knowledge no previous research has linked area level food prices to

diabetes.

Previous work has examined general associations between food prices and health

outcomes, but there is a lack of research examining local healthy and unhealthy food

prices with outcomes, including type 2 diabetes or levels of insulin resistance. For

example, in the study by Duffey et al (Duffey et al., 2010) the prices were metropolitan-

17

level food prices from the Council for Community and Economic Research and were not

broken down by specific neighborhood. Thus, while they found an association between

the regional price of pizza and soda with insulin resistance, it isn’t known what effect

pricing had within individual neighborhoods of a metro region. There may be large

variation in what the prices were for individuals within the same metropolitan area, and

the effect of those prices on insulin resistance (and in turn an increased risk of diabetes)

within a given neighborhood may be much different than what was observed at the

metropolitan area level. Furthermore, there may be underlying differences between

metropolitan areas other than food prices that caused the observed differences in health,

or that may have masked true effects greater than what was seen.

A pair of studies by Drewnowski et al (Drewnowski et al., 2012, Drewnowski et

al., 2014) aimed to better understand the relationship between local supermarket food

prices and their impact on obesity in Seattle and Paris, with the latter expanding the scope

to consider the role of SES in the association (Drewnowski et al., 2014). Individuals

shopping at low price supermarkets had 2.5 times the risk of obesity compared with those

who shopped at high price supermarkets after adjusting for age, gender, SES, and other

covariates. The authors’ conclusion neatly sums up the motivation for the current

research: “low income communities may be vulnerable to obesity not because the nearest

supermarket is several kilometers away but because lower-cost foods tend to be energy

dense but nutrient poor. In other words, access to food needs to be measured also in

economic terms” (Drewnowski et al., 2014). While the Drewnowski studies are an

improvement over many previous studies in terms of spatial resolution of the food

18

environment, it lacks detail about the specific foods individuals were purchasing, and

instead focuses on the price of stores as a whole.

By building off of the work of our predecessors to provide new knowledge

regarding the effect of food prices on diet, insulin resistance and diabetes, this study can

provide key information to policy makers about the impact that the price of healthy and

unhealthy foods has on dietary behavior and downstream health outcomes. Even a

modest reduction of the incidence of diabetes would have a pronounced public health

effect due to the current high rate of new cases in the United States each year.

1.7 SPECIFIC AIMS AND INNOVATION

This study utilizes a large, multi-ethnic cohort of older adults for which details

regarding their diet, insulin resistance level, and diabetes status is available. Using their

geographic home location, the study was able to link individuals to their nearby

supermarkets in order to determine their local food prices. This allowed the investigation

of three specific aims:

Aim 1: Evaluate the variation of healthy and unhealthy food prices across

neighborhoods and their association with neighborhood socioeconomic

status and the proportion Black/Hispanic

Aim 2: Evaluate the association between the price of healthy food relative to

unhealthy food with diet quality, and evaluate whether individual

socioeconomic status moderates this relationship.

Aim 3: Evaluate the association between the price of healthy food relative to

unhealthy food with insulin resistance, diabetes prevalence, and diabetes

19

incidence, and evaluate whether individual socioeconomic status modifies

these relationships.

There remains a gap in the current body of research to synthesize the three

following key pieces: (1) differentiating between healthy and unhealthy foods and

evaluating their price relative to each other; (2) assessing food prices at the neighborhood

level, rather than metropolitan level, to understand the association between food price and

characteristics of the immediate surrounding area; (3) linking local food prices to diet and

health outcomes of individuals, specifically diabetes and insulin resistance. This

dissertation aimed to fill these gaps by incorporating each of these three pieces into a

single study to analyze the association between the price of healthy and unhealthy foods

with diet, insulin resistance, and diabetes in an older population. This work will expand

our knowledge of the effect that food prices have on individuals’ diet and risk of diabetes

and may inform future policies regarding the taxation and subsidization of food costs.

20

CHAPTER 2: THE VARIATION OF HEALTHY AND UNHEALTHY FOOD PRICES ACROSS NEIGHBORHOODS AND THEIR ASSOCIATION WITH

NEIGHBORHOOD SOCIOECONOMIC STATUS AND PROPORTION BLACK/HISPANIC

ABSTRACT

OBJECTIVE: To evaluate food price variation within and between neighborhoods in

order to improve our understanding of access to healthy foods and potential economic

incentives and barriers to consuming a higher quality diet.

METHODS: Prices of healthy foods (dairy, fruits, and vegetables) and unhealthy foods

(soda, sweets, and salty snacks) were obtained from 1953 supermarkets across the US

during 2009-2012, and were linked to census block group socio-demographics. Analyses

evaluated associations between neighborhood SES and proportion Black/Hispanic and

the prices of healthy and unhealthy foods, and the relative price of healthy foods

compared with unhealthy foods (healthy-to-unhealthy price ratio). Linear hierarchical

regression models were used to explore geospatial variation and adjust for confounders.

RESULTS: Overall, the price of healthy foods was nearly twice as high as the price of

unhealthy foods ($0.590 vs. $0.298 per serving; healthy-to-unhealthy price ratio of 1.99).

This trend was consistent across all neighborhood characteristics. After adjusting for

covariates, no association was found between food prices (healthy, unhealthy, or the

healthy-to-unhealthy ratio) and neighborhood SES. Similarly, there was no association

between the proportion Black/Hispanic and healthy food price, a very small positive

association with unhealthy price, and a modest negative association with the healthy-to-

unhealthy ratio.

21

CONCLUSIONS: No major differences were seen in food prices across levels of

neighborhood SES and proportion Black/Hispanic; however, the price of healthy food

was twice as expensive as unhealthy food per serving on average.

22

2.1 INTRODUCTION

The quality of the average diet in the US is well below recommended dietary

guidelines due to low fruit and vegetable consumption and high intake of added sugars,

saturated fats, and sodium (United States Department of Agriculture and United States

Department of Health and Human Services, 2015). Dietary quality is even lower among

African Americans, Hispanics, and those with low socio-economic status (SES)

(Beydoun and Wang, 2008, Forshee and Storey, 2006).

Research has primarily studied physical access to healthier versus unhealthy foods

by examining availability of supermarkets in a neighborhood. Areas without

supermarkets are often classified as food deserts (Walker et al., 2010). However,

economic access – or affordability -- may also play an important role in determining

whether healthy foods are truly available to individuals in a neighborhood. If healthy

food is physically available in the neighborhood, but many are unable to afford it, then

access is still poor; this situation is referred to as a food mirage (Breyer and Voss-

Andreae, 2013). In addition, if unhealthy food is cheaper than healthy food, this may

contribute to disparities in dietary quality since individuals of lower SES are more

sensitive to prices and to food price in particular (Powell et al., 2009b) due to higher

proportion of income spent on food (Frazao et al., 2007). Black and Hispanic populations

often have lower SES, thus price differentials could be contributing to their increased risk

of obesity and diabetes (Ogden et al., 2014).

The difference in price between unhealthy and healthy foods has been studied

previously (Andrieu et al., 2006, Drewnowski, 2010, Drewnowski and Darmon, 2005,

23

Lipsky, 2009, Monsivais et al., 2010), finding that unhealthy foods -- such as snacks

(Lipsky, 2009) or high energy dense macronutrients (sugars, fats and oils) (Drewnowski,

2010) – were much less expensive compared with healthy foods such as fruits and

vegetables. However, very little is known about small area price variations in healthy

and unhealthy foods throughout the U.S. and whether prices vary according to

neighborhood socio-demographics. Prior U.S. work examining the relationship between

food prices and area-level demographics have used within-city food store audits and

report mixed results (Gustafson et al.). Holding constant the type of food store (and

focusing on supermarkets), food in lower vs. higher socio-economic status areas was

commonly found to be cheaper (Andreyeva et al., 2008, Jetter and Cassady, 2006,

Latham and Moffat, 2007).

The current study utilized a novel, large dataset of prices of healthy and unhealthy

foods in supermarkets across the U.S. and assessed variation by neighborhood and price

associations with neighborhood SES and proportion Black or Hispanic. Understanding

how prices vary within and between neighborhoods will improve our understanding of

access to healthy foods and potential economic incentives and barriers to consuming a

higher quality diet.

2.2 METHODS

2.2.1 Price data

Product pricing data was obtained from Information Resources Inc. (IRI), a

market research group focused on consumer packaged goods sold in large chain

supermarkets and superstores across the U.S. (Albertson’s, A&P, Food Emporium,

24

Pathmark, Super Fresh, Kmart, BJ’s, Sam’s Club, Walmart, etc.) (IRI, 2014, IRI, 2015,

Bronnenberg et al., 2008, Symphony IRI, 2015). In total, data used in this study span

January 2009 through December 2012 and came from 1953 stores located in 21 states

(including Washington DC), 193 counties, and 1849 census block groups. Data elements

were product category, item information (Universal Product Code [UPC] and package

size), number of units sold, price of the item, store identifier and store address. IRI’s

pricing database included 299 categories of packaged items – foods like cream/creamers,

mustard/ketchup, spaghetti sauce, and non-foods such as household cleaners, diapers,

tobacco. Fresh fruits and vegetables were not available. For more information regarding

the type of data obtained from IRI see Appendix 1.1.

Data were purchased for 9 available product categories to serve as proxies of

unhealthier and healthier foods. For unhealthier foods, packaged, highly processed, long-

shelf life products were selected: soda, chocolate candy, cookies, and salty snacks. For

healthier foods, refrigerated products were selected in order to roughly approximate costs

of fresh fruit and vegetable spoilage and storage/distribution: refrigerated milk, yogurt,

cottage cheese, frozen vegetables, and fresh orange juice (APPENDIX 1, Appendix 1.1).

For further details on the justification of the products included in the healthy and

unhealthy domains see Appendix 1.2. In order to purchase these data, it was necessary to

request data via UPC code. In total, 171 UPC codes were selected that had the highest

sales across multiple regions and stores. For this reason, all UPCs were top-selling

branded items (n=61 brands, e.g.: Coca Cola, Hershey, Keebler, Lay’s, Tropicana,

Breakstone, Farmland, Dannon, Bird's Eye, etc.). A non-food product category, toilet

25

paper, was also selected and used in statistical models in order to proxy the cost of doing

business in a particular supermarket location (see more information below).

2.2.2 Price variables

Prices reflect the shelf price and included store-level promotions and retailer

coupons, but did not include changes in price applied at the cash register, including taxes

and manufacturer coupons. Volume equivalent prices were calculated as the price of an

item divided by the number of ounces the item contained and multiplied by the typical

serving size of each item (according to the FDA (United States Food and Drug

Administration, 2014)) to create a price per serving size.

Prices per serving were averaged over the 4 year period (2009-2012) for each

product category (e.g., milk). Temporal aggregation was done in order to maximize the

presence of the same UPCs across regions/stores, increase sample size, and stabilize

prices. There was little variability in inflation-adjusted prices between years during this

time period (Kern et al., 2016)). The average price of all items in a product category was

weighted according to volume sold (reported in IRI dataset) to create price per serving for

each product category. Factor analyses used UPC-level price to identify product classes

for healthy and unhealthy domains (See Appendix 1.3 for factor analyses details). This

resulted in two healthy food classes: 1. milk, yogurt, and cottage cheese (AKA “dairy”)

and 2. fresh orange juice and frozen vegetables (AKA “fruits and vegetables”); and three

unhealthy food classes: 1. soda, 2. cookies and chocolate candy (AKA “sweets”), and 3.

chips, onion rings, and pretzels (AKA "salty snacks"). Product prices were then averaged

within the healthy and unhealthy domains and weighted according to national

26

consumption estimates. In general, results were similar when healthy and unhealthy

domains were weighted equally across food classes, APPENDIX 1, Supplemental Table

7.

The average price of brand name toilet paper was calculated for each store in

order to control for baseline costs specific to each store which may influence the price of

food (e.g., rent, distribution, and employee wages) and may not be captured through other

variables. Toilet paper is a good proxy for the basic cost of doing business for the

following reasons: it inhabits significant shelf space, is non-perishable, is unlikely to

suffer from location specific supply shocks, and is unlikely to experience large shifts in

demand (which could also lead price to reflect factors other than store level cost)

(Turcsik, 2014).

2.2.3 Outcome variables

Outcome variables were prices per serving of each of the following product

classes: dairy, fruits and vegetables, soda, sweets, and salty snacks; as well as overall

healthy and unhealthy price and relative price of healthy foods compared with unhealthy

foods (AKA healthy-to-unhealthy price ratio). The relative price was operationalized as

the ratio of the average price per serving of healthy food divided by the average price per

serving of unhealthy food. A ratio >1.0 indicates that the average price of a single serving

of healthy food is more expensive than the price of a serving of unhealthy food (i.e., 1.92

means the price of a serving of healthy food is 92% or 1.92 times more expensive than

the price per serving of unhealthy food on average). A ratio <1.0 indicates that a serving

of unhealthy food is more expensive than healthy food.

27

2.2.4 Census variables

Each store was assigned to the population-weighted centroid of their block group.

Block groups within 1-mile of each store were selected and census data were averaged

for those block groups in order to characterize the census composition around the stores.

All block groups intersecting the 1-mile buffer were included as part of the store’s

neighborhood, and census data were weighted according to the population of each

included block group. The 1-mile buffer was chosen to expand the supermarket’s

neighborhood beyond the block group in which it was located, which may be an

industrial area with a low population and not representative of the surrounding

neighborhood. A 1-mile buffer has been referred to as a relevant consumer market area

for a supermarket (California Center for Public Health Advocacy et al., 2008). A

sensitivity analysis used a 2-mile buffer and results were similar (APPENDIX 1,

Supplemental Table 9).

Census data were obtained from the 2007-2011 American Community Survey

(ACS) 5-year summary file. Neighborhood SES index was created using six variables

from the ACS representing wealth and income, housing value, education, and managerial

or professional occupations, and was operationalized as a single continuous measure as

described by Diez-Roux et al (Diez Roux et al., 2001) (see Table 2-2 footnote).

Neighborhood proportion of Black or non-white Hispanic (AKA “Black/Hispanic”) and

neighborhood SES were converted into percentiles and units represent 20-percentile

increments. To control for potential differences in age distribution and population across

block groups the proportion of individuals aged 20 to 39 years (standardized as a z-

28

score), and the population density (individuals per square kilometer) were included as

covariates in multivariable models.

2.2.5 Geographic variables

Geographic region (Northeast, Midwest, South, and West) and urbanicity of each

store location were included as covariates in the regression analysis. Both were included

to control for differences in infrastructure and other aspects of the built environment

unique to regions or to cities versus rural areas which could affect the shelf price of

foods. Urbanicity was based on county population size, and was operationalized as a

categorical variable with three levels: large metropolitan area of 1+ million residents,

small metropolitan area of less than 1 million residents, and micropolitan (centered on an

urban area with population 10,000 to 49,999) and non-core areas (all other areas smaller

than micropolitan) (United States Department of Agriculture and Economic Research

Service, 2013).

2.2.6 Supermarket density

Supermarket density data were obtained from the National Center for Chronic

Disease Prevention and Health Promotion (NCCDPHP) of the CDC. The NCCDPHP

compiled the number of chain and non-chain supermarkets within each census tract or

within 0.5 miles of the tract boundary (Grimm et al., 2013). Supermarket density was

included as a covariate to control for potential differences in market competition that may

affect product prices (Lamichhane et al., 2013).

29

2.2.7 Statistical analysis

Bivariate associations between socio-demographics (neighborhood SES,

Black/Hispanic, region, and urban class) and food/beverage prices were analyzed using

unadjusted normal linear models. SES and Black/Hispanic were treated as continuous

variables (after verifying approximately linear relationships with price) and region and

urbanicity were treated as categorical variables.

Hierarchical models nesting stores within counties and states were used to

account for the spatial dependence of prices between stores proximal to each other. The

first models aimed to describe the effect of geography on food prices. The model

controlled for geographic region and urbanicity as geographic fixed effects along with

store-level toilet paper price to control for baseline costs. Random intercepts for state and

county allowed mean price to vary across space, and accounted for the correlation

between prices within a given geographic area. The variance of price that was explained

at the state and county level was calculated, with the remaining error considered variation

at the store level.

Lastly, multivariable hierarchical models adjusting for additional census variables

and supermarket density were analyzed. Black/Hispanic and SES effects were first

modeled separately with no covariates. The models then controlled for age, region,

urbanicity, population density, supermarket density, toilet paper price, and

Black/Hispanic (in the SES model) and SES (in the Black/Hispanic model). The effects

of 20 percentile increases in Black/Hispanic and SES were reported along with 95%

confidence intervals and p-values.

30

All statistical analysis was performed in SAS v9.3 (Cary, NC). Spatial methods

linking stores to their neighborhoods were performed using ArcGIS v10.0.

2.3 RESULTS

2.3.1 Descriptive results

Most of the 1,953 stores were located in large metropolitan areas (82.2%, 0) and

in the south (51.7%). The price of healthy foods was nearly twice as high as the price of

unhealthy foods ($0.590 vs. $0.298 per serving [PS]) a trend that was consistent across

all measured neighborhood characteristics and resulted in a healthy-to-unhealthy price

ratio of 1.99 (Table 2-2).

For the healthy food domain, there was a slight decrease in the price of healthy

foods, averaged over product classes, as neighborhood SES decreased ($0.611 vs $0.581

PS in highest to lowest SES quintile). However, within the healthy food domain, the

product category gradient differed: fruits and vegetables decreased in price as

neighborhood SES decreased ($0.526 vs. $0.460 PS) while dairy increased in price

($0.747 vs. $0.788 PS) (APPENDIX 1, Supplemental Table 8). For the unhealthy food

domain, prices were nearly identical between low and high SES quintiles ($0.299 vs.

$0.306 PS) and there were no noteworthy variations within unhealthy food domain

product categories. Thus, prices of healthy food drove variation in the healthy-to-

unhealthy price ratio by neighborhood SES. The ratio was slightly weaker in lower SES

areas: 2.01 vs. 1.95 PS in the highest vs. lowest quintiles, respectively. The association

between neighborhood Black/Hispanic and food prices followed a similar yet inverted

pattern as the SES: areas with higher Black/Hispanic had higher dairy prices, lower fruit

31

and vegetable prices, and little difference in the price of unhealthy foods. Similarly, the

healthy-to-unhealthy price ratio mirrored what was found for SES (2.00 vs 1.97 PS for

neighborhoods with the lowest vs highest proportion Black/Hispanic). Neighborhood

SES and Black/Hispanic were highly correlated with a Spearman rank correlation

coefficient of -0.69 (p<0.001).

2.3.2 Model results

To understand the contribution of area-level characteristics to price differences, a

variance decomposition analysis was performed, first including only the random effects

of state and county (“empty model”), and then and adjusted for fixed effects of region,

urbanicity, and price of toilet paper at the store. In the empty model state-county level

attributes accounted for approximately 70% of unexplained variability in unhealthy prices

and 50% of unexplained variability in healthy prices (Table 2-3). After fixed effect

adjustment for region and urbanicity, unexplained price variation at the state-county level

was slightly lower for unhealthy prices (dropped from 67% to 61%) and a lot lower for

healthy price (dropped from 48% to 26%). The fixed effects primarily accounted for

unexplained variability at the state-level (rather than the county-level). These results

conform to prior work that found fairly low spatial variability in unhealthy prices but

moderate variability in healthy prices (Kern et al., 2016). Variability in the healthy-to-

unhealthy price ratio was similar to the unhealthy food price variability after adjusting for

the fixed effects. Further adjustment for area-level SES and Black/Hispanic had minimal

effect on the decomposition results (data not shown).

32

Table 2-4 shows the estimated associations of neighborhood SES and proportion

Black/Hispanic with each of the price outcomes, including composite healthy and

unhealthy food prices and their ratio. In general, model results were similar to the SES

gradient observed in descriptive results.1 After adjustment for geographic variables,

neighborhood SES was not associated with the composite index for healthy food prices

(due to opposing SES gradients for fruits/vegetables and dairy), the composite index for

unhealthy food price (no association with soda price, a small positive association with the

price of sweets, and a small negative association with salty snacks), or the healthy-to-

unhealthy price ratio.

After adjustment, neighborhood Black/Hispanic was positively associated with

composite index for unhealthy food price (positive associations with price of sweets and

salty snacks, but no association with soda). Black/Hispanic was not associated with

healthy food prices (no associations for fruits/vegetables or dairy). This resulted in a

small negative association with the healthy-to-unhealthy price ratio, indicating that the

price differential was slightly attenuated with increases in Black/Hispanic (for every 20

percentile increase in proportion of Black/Hispanic, the PS price of healthy food relative

to unhealthy food was 1.3% lower [95% CI: -0.7% to -1.9%]). Results from sensitivity

analysis using a 2-mile buffer were consistent and similar in magnitude to primary results

using a 1-mile buffer (APPENDIX 1, Supplemental Table 9).

1 While higher levels of SES were associated with slightly higher prices of fruits and vegetables (approximately a $0.01 increase in price per serving for a 20 percentile increase of SES) higher SES was also associated with lower dairy prices ($0.018 decrease in serving price per 20 percentile SES increase) resulting in an association with a slightly higher composite healthy food price, though this association disappeared after adjusting for covariates.

33

2.4 DISCUSSION

Across all regions and neighborhoods, the average price per serving of healthy

food was nearly twice as high as unhealthy food. There were no strong associations

between neighborhood SES and neighborhood Black/Hispanic and composite indices for

price of healthy foods or unhealthy foods. The only notable associations were within

product categories for the healthy food domain. For example, the price of dairy appeared

to be higher in neighborhoods of lower SES and with higher proportion of

Black/Hispanic, whereas the price of fruits and vegetables was lower in those same

neighborhoods. There was little variation in price of unhealthy food across neighborhood

demographics regardless of the product examined. The healthy-to-unhealthy price ratio

was slightly lower (unhealthy food more expensive than healthy food) as the proportion

of Black/Hispanic increased (even after adjusting for neighborhood SES and other

covariates); there was no evidence of an adjusted association between and the price ratio

with neighborhood SES.

Large absolute differences in price between healthy and unhealthy foods is

consistent with previous work that found prices of soft drinks much lower than fluid milk

(Kern et al., 2016), and sugars, fats, and oils much lower than fruits, vegetables, meat and

poultry in a second study (Drewnowski, 2010), and a third study that reported higher diet

costs were associated with a higher quality diet as measured by the Healthy Eating Index-

2005 (Rehm et al., 2011). These differences are likely due to the increased costs of

refrigeration, farming, and transportation for perishables versus lower such costs for long

shelf-life packaged/processed foods. The combination of lower price and increased

availability of packaged/processed foods may be having profound effects on food

34

purchases and result in less-than-optimal diet quality across a broad spectrum of the

population (Kearney, 2010). The large price divide between healthier and unhealthy

foods is concerning; dietary guidelines emphasize the consumption of fruits, vegetables,

and low-fat dairy while limiting intake of sugar, saturated fats, and sodium (United States

Department of Agriculture and United States Department of Health and Human Services,

2015). If healthy foods cost twice the amount per serving as unhealthy foods, meeting

these dietary guidelines will be difficult for many people, especially those of lower SES.

Additionally, recent work has reported higher prevalence of soda/fast-food advertising in

neighborhoods with higher proportion Black/Hispanic and lower income (Powell et al.,

2014). The confluence of high availability, exposure to advertising, and lower price may

contribute to high consumption of unhealthy foods in these populations (Chandon and

Wansink, 2012, Drewnowski, 2009, Appelhans et al., 2012).

Economists note that the demand for a food product is influenced by its price and

the price of potential substitutes (Andreyeva et al., 2010). For every 10% increase in the

price of fruits and vegetables the quantity demanded of sugars and sweets rises of

upwards to 1%, and the quantity demanded of snacks and fats and oils increases up to

0.6% (Okrent and Alston, August 2012). Because of these cross product elasticities, it is

possible that the unhealthy and healthy food price differentials that we observed may be

responsible for a potential 10% increase in the consumption of unhealthy foods high in

sugar and fat. Given the large discrepancy in price, it would take a radically large, and

unlikely, tax on junk food (or similar intervention) to make the two prices equal.

We used a UPC level dataset and were only able to purchase a small number of

items to represent healthy and unhealthy domains. This is similar to prior research that

35

has heavily relied on index food and beverage products. For example, researchers have

used index items from the Council for Community and Economic Research (C2ER;

formerly the American Chamber of Commerce Researchers Association), a price dataset

for 34 food/beverages sold in U.S. metropolitan areas (pizza, steak, ground beef, bacon,

etc.) (Duffey et al., 2010, Powell and Bao, 2009, The Council for Community and

Economic Research (C2ER), 2015), and average national prices for more than 3,000

foods from the Center for Nutrition Policy and Promotion have been used to study the

relationship between food prices and nutritional value (Drewnowski, 2010, USDA/CNPP,

2008). The primary problem with databases used in prior work is they lack geographic

specificity thus cannot be used to link prices to neighborhood socio-demographics.

Furthermore, our study lacked individual level data on food prices and

sociodemographics, thus the findings of this study are limited to neighborhood level

associations rather than associations between individuals’ race, SES, and food prices at

nearby stores.

Results suggest that healthier and unhealthy food prices vary only slightly by

neighborhood sociodemographics and in the case of neighborhood SES, the direction was

inverse our hypothesis. These findings offer some good news: while healthier foods were

twice the price of unhealthier foods, we found no evidence that within chain supermarket

venues, healthier foods were more expensive and unhealthy foods cheaper in lower SES

and Black/Hispanic areas. However, prior studies have noted that areas with lower SES

and higher proportion minority are more likely to have smaller grocery stores and fewer

large supermarkets and food costs are generally higher in smaller stores (Gustafson et al.,

2012, Kaufman et al., 1997, Chung and Myers, 1999). Thus, price differentials may exist

36

between lower and higher SES communities due to area-level differences in types of food

stores. Nevertheless, it is a strength of this study that we focused on supermarkets as they

are the major venue for all U.S. retail food sales (63% of all retail food sales from 2009 to

2012 (United States Department of Agriculture, 2014a)) and a very rich supermarket

pricing dataset was available. In sum, major advantages of the dataset used in the current

study are: 1. Food/beverage prices were geographically disaggregated to store

locations/neighborhoods; and 2. The dataset offered good generalizability to large chain

supermarkets in urbanized areas across multiple areas of the U.S. for multiple years

(rather than prices for a single city as some others have done (Drewnowski et al., 2014,

Andreyeva et al., 2008)).

This study relied on price per serving as the primary unit of analysis. Other

measures of price that have been used in previous research include the price per calorie

(Drewnowski, 2009, Andrieu et al., 2006) and price of edible food weight (e.g., price per

gram or ounce) (Todd et al., June 2011). Using price per calorie has been criticized

(Burns et al., 2010, Lipsky, 2009) due to a statistical artifact the measurement creates.

And while price per gram or ounce is useful when the foods being compared have

comparable forms and serving sizes, the comparison becomes more difficult to interpret

when the types of foods (e.g., soft drinks versus produce) and/or the serving sizes differ

substantially (Carlson and Frazão, May 2012). Thus, we chose the unit of analysis that

could be compared across all product types, and which may be most meaningful to

consumers: price per serving, which has been used previously in similar research

(Stewart et al., February 2011, Lipsky, 2009, Reed et al., 2004).

37

Ideally we would have included fresh fruits and vegetables but they were

unavailable in the source dataset. It is known that price of orange juice is highly

correlated with the price of fresh oranges (Morris, 2011), and the price of frozen

vegetables is largely similar to those that are fresh, though the level of correlation varies

according to the type of vegetable (Stewart et al., February 2011). Only branded products

were included (see Methods for the rationale). Using brands ensured comparability of

products with and across regions. Brands dominate market share for most of the

unhealthy products (e.g., soda (Beverage World), salty snacks (Grocery Headquarters),

chocolate candy (Grocery Headquarters)) and some (orange juice (Beverage Industry

Magazine))) but not all (dairy (Grocery Headquarters) and frozen vegetables (Store

Brands)) of the healthier products. Additionally, we chose perishable foods to be

representative of healthy foods in order to reflect the price of fresh produce and dairy

whose costs are largely due to transport and perishability. However, not all healthy foods

are necessarily perishable and thus the cost of perishable foods may not be representative

of all healthy foods, although processed fruits and vegetables, such as those that are

canned, are not consistently more or less expensive than fresh produce (Stewart et al.,

February 2011). Ideally we would have also included prices for other non-perishable

healthy foods such as legumes and whole grains, however these data were not available.

Because of this, it is unclear how the results may have differed had the entire universe of

healthy and non-healthy foods been included.

2.5 CONCLUSIONS

This study is one of the first to examine the relationship between healthy and

unhealthy foods and their relationship within and between neighborhoods in regions

38

across the U.S. While no major differences were seen in supermarket food prices across

levels of neighborhood SES and Black/Hispanic, overall the price of healthy food was

twice as expensive as unhealthy food. Such large differences in the affordability of

healthy food compared with unhealthy substitutes may be resulting in less than optimal

diet across all population groups and in particular individuals of lower SES who are more

sensitive to price differences (Powell et al., 2009b). Further research is needed to

determine the extent to which price influences overall diet quality and potential

downstream health effects such as obesity, diabetes, and cardiovascular disease.

39

TABLES AND FIGURES - 2

Table 2-1 Neighborhood demographics and food prices for study sample

Mean/N SD/%

Number of stores 1,953

Neighborhood demographics

Proportion Black/Hispanic (mean, SD) 0.37 0.25

Urban classification

Large metro (pop. ≥1 million) 1,606 82.2

Small metro (pop. <1 million) 296 15.2

Rural (pop <50k) 51 2.6

Region

Northeast 207 10.6

Midwest 173 8.9

South 1,009 51.7

West 564 28.9

Age, proportions (mean, SD)

18 years or younger 0.24 0.05

18-34 years 0.24 0.08

35-64 years 0.40 0.05

65 or older 0.12 0.05

Supermarket density (mean, SD)a 3.34 2.25

Population density (mean, SD) 1,930 3,091

Food prices per serving (mean, SD)

Healthy foods

Frozen vegetables $0.472 $0.073

Orange juice $0.506 $0.037

Fruits and vegetables (OJ and frozen veggies) $0.489 $0.051

Dairy $0.760 $0.100

Healthy food composite $0.590 $0.056

Unhealthy foods

Soda $0.217 $0.029

Chocolate $0.535 $0.050

40

Mean/N SD/%

Cookies $0.214 $0.035

Sweets (Chocolate and cookies) $0.375 $0.037

Salty snacks $0.290 $0.014

Unhealthy food composite $0.298 $0.024

Healthy vs. unhealthy food price

Healthy-to-unhealthy ratio (weight 1) 1.990 0.190 a number of supermarkets within the supermarket census tract or within ½ mile of the tract

41

Table 2-2 Price of foods by demographic characteristics

Number of stores

Healthy food price per serving a

Unhealthy food price per

serving a

Healthy-to- Unhealthy price

per serving a Mean SD Mean SD Mean SD Overall 1953 $0.590 $0.056 $0.298 $0.024 1.990 0.190

Neighborhood SES quintile b

Lowest quintile (least advantaged) 391 $0.581 $0.047 $0.299 $0.019 1.950 0.175

Second quintile 391 $0.581 $0.048 $0.295 $0.020 1.974 0.177

Middle quintile 390 $0.589 $0.053 $0.295 $0.017 2.003 0.196

Fourth quintile 391 $0.590 $0.057 $0.294 $0.016 2.014 0.204

Highest quintile (most advantaged) 390 $0.611 $0.068 $0.306 $0.039 2.010 0.190

Proportion Black/Hispanic quintile

Lowest quint. (0.8% to 14.0% bl/Hisp) 390 $0.592 $0.058 $0.297 $0.033 $2.002 $0.182

Second quint. (14.0% to 24.0%) 392 $0.591 $0.058 $0.298 $0.026 $1.990 $0.187

Middle quint. (24.0% to 37.8%) 390 $0.593 $0.060 $0.297 $0.022 $2.004 $0.203

Fourth quint. (37.9% to 58.4%) 391 $0.587 $0.054 $0.297 $0.019 $1.981 $0.194

Highest quint. (58.5% to 98.8%) 390 $0.588 $0.050 $0.299 $0.018 $1.974 $0.184

Urban classification

Large metro (pop. ≥1 million) 1606 $0.596 $0.058 $0.298 $0.026 $2.007 $0.192

Small metro (pop. <1 million) 296 $0.568 $0.042 $0.298 $0.015 $1.912 $0.163

Rural (pop <50k) 51 $0.553 $0.023 $0.291 $0.015 $1.904 $0.139

Region

Northeast 207 $0.594 $0.074 $0.314 $0.052 $1.912 $0.179

Midwest 173 $0.599 $0.026 $0.279 $0.014 $2.155 $0.106

South 1009 $0.555 $0.024 $0.295 $0.018 $1.888 $0.138

West 564 $0.650 $0.043 $0.302 $0.013 $2.152 $0.143 a Composite healthy and unhealthy food prices were weighted based on national consumption averages of each food category b Neighborhood SES was derived from log of the median household income; log of the median value of housing units; the percentage of households receiving interest, dividend, or net rental income; the percentage of adults 25 years of age or older who had completed high school; the percentage of adults 25 years of age or older who had completed college; and the percentage of employed persons 16 years of age or older in executive, managerial, or professional specialty occupations.

42

Table 2-3 Variance decomposition by geographic area before and after adjusting for region, urbanicity, and toilet paper price

Healthy-to-Unhealthy price ratio

Healthy food price

Unhealthy food price

Variance (SE) Variance (SE) Variance (SE) Empty model

Between states 0.0146 (0.0054) 0.0006 (0.0002) 0.0002 (0.0001) Between counties (in states) 0.0047 (0.0008) 0.0003 (0.0001) 0.0002 (0.0000) Within counties 0.0118 (0.0004) 0.0010 (0.0000) 0.0002 (0.0000) ICC (state level) 47.0% 31.5% 38.1% ICC (county level) 15.1% 16.8% 28.7% Residual 37.9% 51.6% 33.2%

Control for region, urbanicity, and toilet paper price

Between states 0.01 (0.0042) 0.0001 (0.0001) 0.0002 (0.0001) Between counties (in states) 0.0048 (0.0008) 0.0002 (0.0000) 0.0001 (0.0000) Within counties 0.0118 (0.0004) 0.0008 (0.0000) 0.0001 (0.0000) ICC (state level) 37.6% 9.7% 43.8% ICC (county level) 18.1% 15.9% 17.2% Residual 44.3% 74.4% 39.0%

CI, confidence interval; ICC, Intraclass correlation, the proportion of variation in price that is accounted for by differences at the state and county levels; SE, standard error

43

Table 2-4 Hierarchical model estimates regressing price outcomes on neighborhood socioeconomic status (SES) and the proportion of individuals who are Hispanic or black, nested within county and state, N = 1953

Estimate a

95% CI Model Lower Upper P-value SES MODELS: Mean difference in price per 20 percent increase in neighborhood socioeconomic index

Healthy food price Unadjusted b 0.00452 0.0034 0.0056 <0.0001 Adjusted c 0.00090 -0.0007 0.0025 0.2836 Fruits and veggies Unadjusted b 0.01386 0.0126 0.0151 <0.0001 Adjusted c 0.01096 0.0091 0.0128 <0.0001 Dairy Unadjusted b -0.01323 -0.0153 -0.0112 <0.0001 Adjusted c -0.01836 -0.0215 -0.0152 <0.0001 Unhealthy food price Unadjusted b 0.00057 0.0001 0.0011 0.0234 Adjusted c 0.00055 -0.0001 0.0012 0.1110 Soda Unadjusted b 0.00051 -0.0001 0.0011 0.0937 Adjusted c 0.00065 -0.0003 0.0016 0.1682 Sweets Unadjusted b 0.00267 0.0018 0.0035 <0.0001 Adjusted c 0.00153 0.0005 0.0026 0.0054 Salty snacks Unadjusted b -0.00216 -0.0025 -0.0018 <0.0001 Adjusted c -0.00125 -0.0018 -0.0007 <0.0001 Healthy-to-Unhealthy price ratio

Unadjusted b 0.01146 0.0076 0.0153 <0.0001 Adjusted c 0.00037 -0.0059 0.0066 0.9088

BLACK/HISPANIC MODELS: Mean difference in price per 20 percent increase in neighborhood proportion of Black/Hispanic

Healthy price Unadjusted b -0.00319 -0.0043 -0.0020 <0.0001 Adjusted c 0.00007 -0.0016 0.0017 0.9377 Fruits and veggies Unadjusted b -0.01069 -0.0120 -0.0094 <0.0001 Adjusted c -0.00089 -0.0028 0.0010 0.3484 Dairy Unadjusted b 0.01105 0.0089 0.0132 <0.0001 Adjusted c 0.00223 -0.0009 0.0054 0.1687 Unhealthy price Unadjusted b 0.00058 0.0001 0.0011 0.0224 Adjusted c 0.00212 0.0014 0.0028 <0.0001 Soda Unadjusted b 0.00022 -0.0004 0.0008 0.4764 Adjusted c 0.00077 -0.0002 0.0017 0.1063 Sweets Unadjusted b -0.00053 -0.0014 0.0003 0.2244 Adjusted c 0.00360 0.0025 0.0047 <0.0001 Salty snacks Unadjusted b 0.00220 0.0018 0.0026 <0.0001 Adjusted c 0.00122 0.0007 0.0018 <0.0001 Healthy-to-Unhealthy price ratio

Unadjusted b -0.01486 -0.0187 -0.0110 <0.0001 Adjusted c -0.01274 -0.0190 -0.0065 <0.0001

a per 20 percentile change in neighborhood SES or Black/Hispanic b Models did not include covariates but were adjusted for county and state via model nesting

44

c Includes covariates: Age, region, urbanicity, population density, supermarket density, toilet paper price, and either race (in the SES models) or neighborhood SES (in the race models).

45

CHAPTER 3: THE PRICE OF HEALTHY FOOD RELATIVE TO UNHEALTHY FOOD AND ITS ASSOCIATION WITH DIET QUALITY: THE MULTI-

ETHNIC STUDY OF ATHEROSCLEROSIS

ABSTRACT

BACKGROUND: The price of food influences the purchasing decisions of individuals.

This study examined if the price of healthy food relative to unhealthy food is associated

with dietary quality.

METHODS: Cross-sectional data came from The Multi-Ethnic Study of Atherosclerosis

(exam 5, 2010-2012 N=2765) and supermarket food/beverage prices (Information

Resources Inc. n=794 supermarkets). For each individual, average price of healthy foods,

unhealthy foods and their ratio was computed for supermarkets within 3-miles. Logistic

regression estimated odds ratios of a high quality diet (top quantile of Healthy Eating

Index 2010) associated with healthy-to-unhealthy price ratio, adjusted for demographics,

smoking, geographic region, neighborhood socioeconomic index, supermarket and

population density, and cost of living. Sensitivity analyses used an instrumental variable

approach.

RESULTS: Healthy foods cost nearly twice as much as unhealthy foods on average

(mean±SD healthy-to-unhealthy ratio=1.97±0.14). A larger healthy-to-unhealthy price

ratio was associated with lower odds of a high quality diet (OR=0.76 per SD increase in

the ratio, 95% CI=[0.64 to 0.91]). Instrumental variable analyses largely confirmed these

findings although confidence intervals were wider and the result was no longer

statistically significant (OR=0.82 [0.57 to 1.19]).

46

CONCLUSIONS: A higher price of healthy food relative to unhealthy foods appears to

be negatively associated with a high quality diet. Policies to address the large price

differences between healthy and unhealthy foods may help improve diet quality in the

U.S.

47

3.1 INTRODUCTION

An individual’s healthy food environment encompasses more than just physical

access to fresh fruits, vegetables and other nutritious foods. Economic access and the

affordability of healthy foods in the neighborhood is also a key component of one’s food

environment. The price of food – in addition to taste, nutrition, convenience, and other

factors – affects the purchasing choices of individuals who must navigate environments

with numerous choices and saturated by advertising(Connors et al., 2001, Andreyeva et

al., 2010, Aggarwal et al., 2016). Healthy foods have been found to be more expensive

than unhealthy foods, whether measured per calorie (Drewnowski, 2010, Andrieu et al.,

2006) or per serving (Lipsky, 2009, Reed et al., 2004, Stewart et al., February 2011, Kern

et al., 2016). For this reason, low cost diets are associated with higher calorie intake at the

expense of fewer nutrients (Andrieu et al., 2006), while healthier diets tend to be more

expensive (Rehm et al., 2011, Ryden and Hagfors, 2011, Schroder et al., 2006).

A healthy diet, as measured by the Healthy Eating Index (HEI), is inversely

associated with obesity (Guo et al., 2004), waist circumference (Tande et al., 2010),

diabetes (Chiuve et al., 2012), cardiovascular disease (Chiuve et al., 2012, McCullough et

al., 2000), stroke (Chiuve et al., 2012), cancer (Arem et al., 2013, Li et al., 2014) and

mortality (Rathod et al., 2012, Kappeler et al., 2013, Harmon et al., 2015, Reedy et al.,

2014). The average consumer purchases foods that result in a diet that falls well short of

the recommendations in the USDA dietary guidelines (Volpe and Okrent, 2012). This

may be due in part to purchasing too few fresh fruits and vegetables and instead

purchasing more affordable processed fats, sugars, and sweets.

48

Lower SES (Ball et al., 2006, De Irala-Estevez et al., 2000, Dubowitz et al., 2008)

has been associated with decreased consumption of fruits and vegetables, and prior work

has found that these SES-differences can be explained primarily by the cost of a diet

(Monsivais et al., 2012, Konttinen et al., 2013). It is reasonable to expect that large

differences in price between healthy and unhealthy foods would lead to differences in

purchasing patterns and resulting diets, and that those differences would be more

prominent for individuals of lower SES (Powell and Chaloupka, 2009, Powell et al.,

2009b). An experimental study found that food taxation and subsidy policies have a

differential effect by income level (Darmon et al., 2016), while observational research

found that the effect of metropolitan area food prices on diet quality was not modified by

income level (Beydoun et al., 2008). However, there is a lack of research examining how

local neighborhood food prices may affect diet quality differently across levels of SES.

This cross-sectional study spatially linked 2,765 participants from the Multi-

Ethnic Study of Atherosclerosis (MESA) to nearby supermarkets to examine the

association between neighborhood food price and diet quality.

3.2 METHODS

3.2.1 MESA data

MESA is a population-based longitudinal cohort study of ethnically diverse adults

aged 45-84 years (Bild et al., 2002). Individuals were recruited from six sites across the

United States: Bronx/Upper Manhattan, NY; Baltimore City and Baltimore County,

Maryland; Forsyth County, North Carolina; Chicago, Illinois; St. Paul, Minnesota; and

49

Los Angeles County, California. MESA included a baseline examination (2000-2002)

and four follow-up exams; 4716 respondents participated in exam 5 (2010-2012).2

Food pricing was only available concurrent to exam 5, thus this study only

included participants who completed exam 5, administered during April 2010 through

January 2012. MESA person-level data included in this study were: diet (see details

below), age (continuous), sex, race/ethnicity (Caucasian, Chinese-American, African-

American, Hispanic), smoking status (never, former, current), marital status, body mass

index (BMI), physical activity, education level (high school diploma, GED, or less; some

college, technical or associat0es degree; bachelor’s degree or higher), and income/wealth

index (an ordinal measure ranging from 0 to 8 based on the combination of income level

and ownership of four assets: car, home, land, and investments).

3.2.2 Food Frequency Questionnaire (FFQ)

The FFQ used in the MESA was a modified-Block-style, 128-item questionnaire

adapted from the questionnaire designed for the Insulin Resistance and Atherosclerosis

Study (IRAS) (Mayer-Davis et al., 1999) and has been described elsewhere (Nettleton et

al., 2006).3 Responses to the FFQ were used to calculate an HEI score for each

individual.

2 MESA sampling frame of the cohort differed by site, but all had a strategy to recruit individuals to meet a priori defined distributions across strata defined by gender, ethnicity, and age group rather than to represent the demographic distribution of the source communities. Some field centers used stratified random sampling according to age and gender, others used recruitment based on geographic boundaries rather than demographics, while others used random digit dialing across the entire sampling frame. Sampling frames of individuals and households were obtained through a variety of sources across sites, including a Department of Motor Vehicles, employment records, commercial mailing services, and a county assessor’s office. 3 Modifications to the FFQ included additional items to reflect the racial composition of the MESA cohort. For each of the food items on the FFQ, respondents chose their consumption frequency (rare or never, 1 per month, 2-3 per month, 1 per week, 2 per week, 3-4 per week, 5-6 per week, 1 per day, and 2+ per day); their frequency of consumption was then weighted by a multiplier according to their reported typical serving size (× 0.5, × 1.0, and × 1.5 for small, medium, and large, respectively). IRAS was validated against 24-hour dietary recalls (Mayer-Davis et al., 1999) and the MESA

50

3.2.3 Outcome: Healthy Eating Index-2010 (HEI)

HEI-2010 was used to assess dietary quality. The HEI reflects 2010 federal

dietary guidelines, has been used to monitor and assess diet quality in the U.S.(Kennedy

et al., 1995, United States Department of Agriculture and United States Department of

Health and Human Services, 2010, Guenther et al., 2013), and has (1) adequate content

validity (Guenther et al., 2013) (2) sufficient construct validity, and (3) acceptable

reliability (Guenther et al., 2014). The HEI-2010 includes twelve components (whole

fruit, total vegetables, sodium, etc.), each of which contribute a minimum of 0 to a

maximum of 5, 10, or 20 points, resulting in a range of 0 to 100 for the total score and

higher scores indicate a healthier diet(Guenther et al., 2013).

Preliminary analyses revealed a roughly normal distribution of HEI scores. We

ranked the index into quintiles and operationalized “high quality diet” as the top quintile

(>80th percentile) of all scores in the sample. The top quintile has been used to define a

healthy diet in prior work (Arem et al., 2013, Harmon et al., 2015, Reedy et al., 2014)

and ranking the dietary values acknowledges the low precision inherent in dietary self-

reports.

3.2.4 Price data

Data on food prices were obtained from Information Resources Inc. (IRI), a

market research group that monitors prices of 299 consumer packaged goods sold in large

chain supermarkets and superstores across the US (IRI, 2014, IRI, 2015, Bronnenberg et

diet data correlated as expected with HDL-cholesterol and TAG concentrations (Nettleton et al., 2009b) and cardiometabolic conditions (Nettleton et al., 2008a, Gao et al., 2008, Abiemo et al., 2013, Nettleton et al., 2008b, Levitan et al., 2016).

51

al., 2008, Symphony IRI, 2015). From the 109 supermarket chains in our study area, 8

parent companies (representing 27 supermarket chains) did not agree to release their

stores’ data. Ultimately data were included from 794 stores (82 chains) located in 11

states (including Washington DC), 72 counties, and 757 census block groups. Data years

were 2009-12.

Nine food/beverage product categories were selected to serve as proxies for either

healthier or unhealthier foods. Because data for fresh fruits and vegetables were not

available, refrigerated products were selected as a proxy for fresh produce under the

assumption that they had similar spoilage and storage/distribution costs as fresh produce.

Healthier food was represented by dairy (refrigerated milk, yogurt, cottage cheese) and

fruits and vegetables (frozen vegetables, and fresh orange juice). Unhealthier food was

represented by packaged, highly processed, long-shelf life products: soda, sweets

(chocolate candy, cookies) and salty snacks. Detailed methods are reported elsewhere

(Kern et al.).

Prices in the database did not include taxes and manufacturer coupons, but instead

reflected the shelf price and included store-level promotions. The primary exposure of

interest was the price of healthy foods relative to unhealthy foods, which was

operationalized as the ratio of the average price per serving of healthy food divided by

the average price per serving of unhealthy food and referred to as the healthy-to-

unhealthy price ratio. A ratio >1.0 indicates serving of healthy food was more expensive

than a serving of unhealthy food. The price of healthy foods and unhealthy foods were

also modeled separately as secondary exposures of interest. Each price exposure was

converted to a z-score: representing an increase of 14% in the relative price of healthy

52

food compared with unhealthy food for the healthy-to-unhealthy ratio, and an increase of

$0.04 and $0.03 in the average price per serving of healthy and unhealthy food prices,

respectively.

The average price of brand name toilet paper in stores within 3 miles of each

MESA participant was also calculated and used as an instrument for unhealthy food price

and the price ratio in the sensitivity analyses described in the ‘Statistical Methods’

section.

The number of IRI supermarkets within 3-miles of each MESA participant’s

address at exam 5 was created and referred to as the supermarket density.

3.2.5 Census data

US Census data came from the American Community Survey 2007-2011.

Geographic regions (Northeast, Midwest, South, West) and population density 4 were

assigned to each participant. A neighborhood block group SES index was created using

six variables representing wealth and income, such as household income, housing value,

investment income, education level, and managerial occupations, and was operationalized

as a single continuous measure, as described elsewhere (Diez Roux et al., 2001).

4 Population density per square mile was obtained using the 2010 census. Total population within a 1 mile buffer was calculated based on block-level census population and each block was weighted by the percent of the block area that fell within the participant buffer. The total population within that block was then multiplied by this weight and the weighted populations were summed together for the total population within the buffer. The total population was divided by total buffer area in square miles.

53

3.2.6 Cost of living data and supermarket density

The cost of living index 2010 was obtained from the Council for Community and

Economic Research for each metropolitan area (Council for Community and Economic

Research, 2016).

3.2.7 Data linking

Addresses of MESA participants and supermarket addresses from the pricing

dataset were used to link individuals to the average food/beverage prices at supermarkets

within a 3-mile buffer (radius) of each MESA participant residence using ArcGIS 10.0.

Median number of supermarkets per MESA participant in the analytic sample was 5

(25th-75th percentile 3-6). A 3-mile radius was used for consistency with other research

examining neighborhood food environments, and is in line with prior research estimating

the average distance individuals travel to their primary supermarket (The Capitol Forum,

2014, Drewnowski et al., 2012, Fuller et al., 2013, Michimi and Wimberly, 2010).

Sensitivity analyses using equal weights rather than those based on national consumption

averages, using supermarkets within 5-miles instead of 3-miles, and using a 1-mile buffer

for those living in New York City produced similar results which can be found in

APPENDIX 2 Supplemental Table 11.

3.2.8 Statistical methods

Multivariable logistic regression was used to model the relative odds of having a

high quality diet for every standard deviation increase in the healthy-to-unhealthy price

ratio. Model 1 adjusted for geographic region, age (continuous), and gender. Model 2

added income/wealth index, education, race/ethnicity, and smoking status, and model 3

54

added neighborhood level SES and supermarket density. Other variables (marital status

and physical activity) were considered as covariates in the model, but were not included

due to a lack of association with the outcome and not having any impact on the effect

estimate of price in the multivariable models.

We tested the interaction of price with education level and income/wealth tertile

by including appropriate interaction terms (separate models examining education and

income/wealth) and conducting stratified analyses by tertile of income/wealth (low,

medium, and high) and education level (high school degree or less, some college, or

bachelor’s degree or more). Our hypothesis was that the effect of price on diet would be

stronger for lower levels of SES.

In addition to the above multivariable models, sensitivity analyses used an

instrumental variable approach in order to remove potential unmeasured confounders that

would affect food prices and diet such as the local food culture and the types of foods

typically available in the neighborhood. The use of instrumental variables (IV) has been

increasing in epidemiologic research as a means to estimate causal effects (Greenland,

2000, Hernan and Robins, 2006, Martens et al., 2006). Toilet paper price was chosen as

the instrument for the price of unhealthy foods, and in turn the healthy-to-unhealthy price

ratio, because it (1) has a strong association with food prices in the same stores

(particularly unhealthy long-shelf life foods), (2) has no anticipated causal association

with participants’ diet quality other than through its correlation with food price, and (3) is

in principle not associated with unmeasured confounders mentioned above. These

characteristics satisfied the three major assumptions of an instrument: it is (1) strongly

associated with the exposure; (2) any effect it has on the outcome is fully mediated by the

55

exposure; and (3) it shares no unmeasured common causes with the outcome (Rassen et

al., 2009).

A two-stage residual inclusion (2SRI) analysis was used to perform the

instrumental variable analysis. A 2SRI model was used because they are statistically

consistent in non-linear models, such as the logit model used in this study, while the two-

stage predictor substitution model commonly used in linear models is not (Terza et al.,

2008). In the first stage, linear regression was used to regress the price ratio on toilet

paper price and covariates (age, gender, region, income/wealth index, education, race,

smoking status, neighborhood SES, supermarket density, population density, and cost of

living). The residuals from the stage 1 regression were then included as a covariate in the

second stage. The second stage model used a logit model to regress the indicator for a

high quality diet on the price ratio (the main exposure of interest), the residuals obtained

from stage 1, and all covariates used in stage 1. Bootstrapping of 10,000 samples was

used to calculate 95% credible intervals for the stage 2 estimates.

3.3 RESULTS

Among 4716 MESA participants who contributed data to exam 5, 1047 were

excluded due to not having dietary data. Further exclusions were: 835 due to not being

within 3 miles of a supermarket in our dataset (of which 674 were from the St. Paul

MESA site where supermarket data were unavailable) and 68 who did not have full

covariate data. The final analytic sample was 2765. Sample characteristics for included

vs. excluded are shown in APPENDIX 2 Supplemental Table 10. On average, the

excluded sample was similar to those who were included on most characteristics. There

56

were differences in regional and racial characteristics due to the exclusion of those living

in the St. Paul area, and small differences in education levels as a smaller proportion of

college graduates were in the excluded sample.

Demographics and other characteristics of the MESA participants are reported in

Table 3-1. Overall the serving price of healthy food was nearly twice as expensive as

unhealthy food (mean±SD of healthy-to-unhealthy price ratio = 1.97±0.14, and on an

absolute scale: $0.60±$0.04 vs. $0.31±$0.03 per serving for healthy and unhealthy food,

respectively, Table 3-2). While the ratio was analyzed as a continuous z-score in all

analytic models, the tertiles of healthy-to-unhealthy price ratio are used in the table for

descriptive purposes. Large difference across the tertiles was only apparent for region

(highest ratio in west and lowest in south) and for race/ethnicity (Chinese resided in

highest ratio areas and Black in lowest ratio areas, Table 3-1). Prior to adjusting for

covariates, the proportion of individuals with a high quality diet was 24% for those with

the highest healthy-to-unhealthy ratio compared with 18% of other individuals (Table 3-

2).

3.3.1 Primary analysis

Table 3-3 displays sequential adjustment for covariates in each of the three

models. Results of the final model indicate a modest inverse association between the

healthy-to-unhealthy ratio and high quality diet. That is, for every one standard deviation

increase (14% higher) in the price of healthy food to unhealthy food, the odds of having a

high quality diet decreased by 24% (OR=0.76, 95% CI=[0. 64 to 0.91]). No association

was found between healthy food price alone with diet (OR=1.04), but a higher price (per

57

standard deviation increase, approximately $0.03) of unhealthy food was associated with

increased odds of having a high quality diet (OR=1.16, 95% CI=[1.02 to 1.33]).

There was a statistically significant interaction between individual income/wealth

and the price ratio on diet while adjusting for all covariates included in the final model (p

for interaction 0.001, Table 3-4). All stratified point estimates were in the expected

direction but the gradient was somewhat unexpected: the strongest and only statistically

significant point estimate was in the middle income/wealth tertile (OR=0.61). The next

largest effect was found within the lowest income/wealth tertile (OR=0.79), while those

in the highest tertile had the weakest effect (OR=0.87), as expected. The interaction was

not statistically significant between education and the price ratio (p for interaction 0.095),

nevertheless, stratified estimates showed the weakest association in least educated and the

strongest in most educated (OR 0.85 for high school or less and OR 0.77 for Bachelor’s

or more), counter to expectation.

3.3.2 Instrumental Variable Analyses

Results of the 2SRI IV analysis are shown in Table 3-5. In stage 1, the instrument,

toilet paper price, was positively associated with unhealthy food price (spearman

correlation coefficient, r=0.56) and negatively associated with the healthy-to-unhealthy

price ratio (r= 0.41). As expected the correlation was higher with unhealthy food prices

(r=0.56) than with healthy food prices (r=0.29). The F-statistic was 1113 and guidelines

specify the minimum value should be 10 (Stock et al., 2002). In stage 2, the association

between healthy-to-unhealthy price ratio and high quality diet was similar in magnitude

to the results obtained in the primary analysis; however, the results was not statistically

58

significant due to wide confidence intervals (OR=0.82, 95% bootstrapped CI=[0.57 to

1.19]).

3.4 DISCUSSION

Higher prices of healthy foods relative to unhealthy foods were found to be

associated with lower odds of having a high quality diet: per 14% higher ratio of healthy

food to unhealthy food, there was 24% lower odds of having a high quality diet. Despite

healthy foods costing more than unhealthy foods, there was no association between diet

and prices of healthy foods alone, but there was a positive association with unhealthy

food price alone (per $0.03 higher unhealthy food price per serving there was a 16%

higher odds of having a healthy diet). To our knowledge, no prior study examining

healthy and unhealthy food prices has linked neighborhood food prices (rather than

aggregate level prices) to a cohort of individuals throughout the U.S. to examine the

association with diet.

Consumers report that price is a main driver in food purchase decision making

(Connors et al., 2001, Glanz et al., 1998, International Food Information Council

Foundation, 2015) and our study illustrates how this may impact diet. While much work

has been done to improve the quality of diets by understanding and improving the lack of

physical access to supermarkets in food deserts (Walker et al., 2010), additional paths

should be considered. Improving the affordability of healthy foods relative to their

unhealthy substitutes may be an effective option. Examples of options proposed for

improving the relative affordability of healthy foods include taxation of unhealthy foods,

subsidizing healthy food, and combinations of these practices. One study suggests taxing

59

unhealthy foods has the benefit of raising revenue and could be an effective strategy to

influence dietary changes (Thow et al., 2014). Another study suggests that subsidizing

healthy food has the benefit of directly making healthy food more affordable and appears

to increase the consumption of those foods, though with a smaller effect as the unhealthy

food taxes (Powell et al., 2013) and may require government funding. It has been

suggested that combining strategies to influence consumption patterns would be much

more effective than any single method alone (Epstein et al., 2012, Thow et al., 2014) and

may satisfy consumers generally opposed to ‘sin’ taxes by providing them simultaneous

savings on their grocery bill due to cheaper healthy foods (O’Hara and Musicu, 2015).

We expected differences in the price exposure to have more of an effect within

those of lower SES; however, our results suggested it was instead those in the middle

range of the income/wealth index, and those with the highest level of education for which

there was a stronger effect. This result may be explained by the following: healthy food

may still be too expensive for the lowest SES group even at its most affordable, as the

lowest observed healthy-to-unhealthy price ratio for any individual was 1.55. Given that

low-income families need to devote approximately half of their food budget to meet the

dietary guidelines for fruit and vegetable intake (Cassady et al., 2007), the relative price

of healthy food may have to be even lower than what was observed in this study before a

strong effect is seen within the lowest SES individuals. These results are consistent with

findings from a recent study which found food price policies had much larger effects in

middle-income individuals compared with those of low-income (Darmon et al., 2016).

Because of this, large subsidies for healthy foods paired with high taxes of unhealthy

foods may be required that move the relative price of healthy food lower than the levels

60

measured in this study to have an effect within low SES populations (McGill et al.,

2015).

Instrumental variable analyses confirmed odd ratios found in the main analyses

although confidence intervals were wider. The IV analysis adds some face validity to the

standard regression results; however, as with all models, IV methods have assumptions

(Rassen et al., 2009), some of which are not verifiable – notably the third assumption

prohibiting any association between the IV and unmeasured cofounders (Martens et al.,

2006). Confidence intervals from the IV analysis were wider than those obtained from the

primary analysis due to the two stage structure of the models and because the prediction

of the price ratio by the IV was imperfect (Rassen et al., 2009).

3.4.1 Limitations

While the validity of the FFQ used in this study has been documented (Nettleton

et al., 2009b, Mayer-Davis et al., 1999) the limitations to FFQ are well-known (Willett

and Hu, 2006). The FFQ contained an a priori food list that may miss foods consumed by

a diverse population; however, a strength of the MESA FFQ was that it included many

foods that reflect the diversity of the multi-ethnic population. By defining a high quality

diet relative to the population (top quintile) we attempted to address some of its

limitations regarding underreporting. Selection bias may have been introduced by

limiting analyses to individuals living within 3 miles of a supermarket. However, an

examination of the characteristics of those included and excluded found few differences

with the exception of region and race/ethnicity, which was due to the excluded being

largely concentrated in a single recruitment site where supermarket data were

61

unavailable. Other measurements of food price may have been considered for this

analysis, including the price per unit of weight, or price per calorie (Andrieu et al., 2006,

Drewnowski, 2009); however, we chose the price per serving as it is a unit of analysis

that can be compared across all product types, may be most meaningful to consumers,

and which has been used previously in similar research (Lipsky, 2009, Reed et al., 2004,

Stewart et al., February 2011). Lastly, there are limitations to the products included in the

price measures. Data on fresh fruits and vegetables were not available and instead a

proxy using frozen vegetables and refrigerated orange juice was used which is highly

correlated with the price of fresh ranges (Morris, 2011) and data on other healthy foods

such as whole grains and legumes was not available in this analysis. Additionally, data

were limited to branded products in order to compare the same products across the

country.

3.5 CONCLUSIONS

With a large proportion of the U.S. population failing to achieve a healthy diet it

is important to understand the root causes. This study suggests that the large difference in

price between healthy and unhealthy food may play a role. If this is true, interventions

may be considered to improve the affordability of healthy foods relative to unhealthy

alternatives.

62

TABLES AND FIGURES - 3

Table 3-1 Characteristics of the individuals included in the analysis by tertiles of healthy-to-unhealthy food price ratio (n=2,642)

All participants

Lowest ratio, smallest

differential (1.55 - 1.88)

Middle ratio (1.88 - 2.01)

Highest ratio, largest

differential (2.01 - 2.39)

N /

Mean Col % /

SD n /

Mean Row %

/ SD n /

Mean Row %

/ SD n /

Mean Row %

/ SD Number of participants (N) 2765 938 903 924 MESA recruitment site (n, %) a

Forsyth County, NC 539 19.5% 239 44.3% 299 55.5% 1 0.2% New York, NY 538 19.5% 262 48.7% 226 42.0% 50 9.3% Baltimore, MD 464 16.8% 339 73.1% 122 26.3% 3 0.6% St. Paul, MN 8 0.3% 6 75.0% 0 0.0% 2 25.0% Chicago, IL 651 23.5% 89 13.7% 221 33.9% 341 52.4% Los Angeles, CA 565 20.4% 3 0.5% 35 6.2% 527 93.3%

Region of residence (n, %)

Northeast 534 19.3% 259 48.5% 226 42.3% 49 9.2% Midwest 642 23.2% 85 13.2% 220 34.3% 337 52.5% South 1021 36.9% 593 58.1% 421 41.2% 7 0.7% West 568 20.5% 1 0.2% 36 6.3% 531 93.5%

Supermarket density (3 mile) (mean, SD) 1.19 1.42 1.51 1.88 1.22 1.30 0.84 0.74 Female (n, %) 1466 53.0% 483 32.9% 502 34.2% 481 32.8% Age (mean, SD) 70.3 9.5 70.6 9.1 70.3 9.3 69.9 9.9 Race/ethnicity (n, %)

White 1101 39.8% 422 38.3% 485 44.1% 194 17.6% Chinese American 359 13.0% 10 2.8% 56 15.6% 293 81.6% Black 834 30.2% 413 49.5% 231 27.7% 190 22.8% Hispanic 471 17.0% 93 19.7% 131 27.8% 247 52.4%

Education (n, %) High school diploma or less 777 28.1% 231 29.7% 251 32.3% 295 38.0% Some college 761 27.5% 259 34.0% 233 30.6% 269 35.3% Bachelor’s degree or more 1227 44.4% 448 36.5% 419 34.1% 360 29.3%

63

All participants

Lowest ratio, smallest

differential (1.55 - 1.88)

Middle ratio (1.88 - 2.01)

Highest ratio, largest

differential (2.01 - 2.39)

N /

Mean Col % /

SD n /

Mean Row %

/ SD n /

Mean Row %

/ SD n /

Mean Row %

/ SD Per capita household income (in $10k) 2.6 1.9 2.8 1.8 2.8 1.9 2.3 1.8 Wealth index 2.6 1.2 2.6 1.2 2.6 1.2 2.5 1.2 Income/wealth index 5.1 2.2 5.3 2.1 5.2 2.2 4.9 2.3 Marital status (n, %)

Not married or living with partner 1107 40.0% 419 37.9% 364 32.9% 324 29.3% Married/Living with partner 1658 60.0% 519 31.3% 539 32.5% 600 36.2%

BMI (mean, SD) 28.2 5.6 28.9 5.5 28.4 5.9 27.2 5.2 <25 (n, %) 855 30.9% 236 27.6% 268 31.3% 351 41.1% 25-29.9 (n, %) 1043 37.7% 347 33.3% 341 32.7% 355 34.0% ≥30 (n, %) 867 31.4% 355 40.9% 294 33.9% 218 25.1%

Smoking status (n, %) Never smoked 1281 46.3% 366 28.6% 412 32.2% 503 39.3% Former smoker 1283 46.4% 487 38.0% 433 33.7% 363 28.3% Current smoker 201 7.3% 85 42.3% 58 28.9% 58 28.9%

Physical activity, MET min per week (mean, SD) 2774 3552 3239 4237 2765 3441 2310 2749 a This is the MESA location of the participant, not necessarily their area of residence

64

Table 3-2 Proportion of participants with a high quality diet by tertile of the healthy-to-unhealthy price ratio and the average serving price of healthy and unhealthy foods (n=2,765)

All participants Lowest ratio (1.55 - 1.88)

Middle ratio (1.88 - 2.01)

Highest ratio (2.01 - 2.39)

Number of participants (N) 2765 938 903 924

Number with high quality diet (n, %) 545 19.7% 165 17.6% 161 17.8% 219 23.7%

Average food prices per serving

Healthy food price per serving a

(mean, SD, median) $0.60 $0.04 $0.60 $0.61 $0.06 $0.57 $0.59 $0.03 $0.58 $0.62 $0.03 $0.62

Unhealthy food price per serving b

(mean, SD, median) $0.31 $0.03 $0.30 $0.33 $0.04 $0.31 $0.30 $0.01 $0.30 $0.29 $0.01 $0.29

Ratio of Healthy-to-Unhealthy

(mean, SD, median) 1.97 0.14 1.93 1.83 0.05 1.85 1.95 0.05 1.93 2.14 0.09 2.13 a Healthy foods were represented by frozen vegetables, orange juice, and dairy (milk, yogurt, and cottage cheese) b Unhealthy foods were represented by soda, salty snacks (chips, pretzels, onion rings), and sweets (chocolate candy and cookies)

65

Table 3-3 Odds ratios of having a high quality diet associated with the price ratio, and with the prices of healthy foods and unhealthy foods after sequential adjustment for confounders within the full population (n=2,765) and after removing New York/Bronx (n=2,225)

Exposure of interest

Healthy-to-unhealthy ratio Healthy food price Unhealthy food price 95% CI 95% CI 95% CI

Model covariates Odds ratio Lower Upper

p-value

Odds ratio Lower Upper

p-value

Odds ratio Lower Upper

p-value

Full sample (n=2,765)

Model 1: region, age, gender 0.97 0.83 1.13 0.6978 0.98 0.88 1.10 0.7655 1.00 0.91 1.10 0.9639

Model 2: Model 1 plus income/wealth,

education level, smoking status, and race 0.86 0.73 1.01 0.0571 0.97 0.86 1.09 0.5709 1.03 0.93 1.13 0.6045

Final Model: Model 2 plus neighborhood

SESa and neighborhood supermarket density 0.76 0.64 0.91 0.0027 1.04 0.88 1.22 0.6371 1.16 1.02 1.33 0.0267 a Neighborhood SES was derived from log of the median household income; log of the median value of housing units; the percentage of households receiving interest, dividend, or net rental income; the percentage of adults 25 years of age or older who had completed high school; the percentage of adults 25 years of age or older who had completed college; and the percentage of employed persons 16 years of age or older in executive, managerial, or professional specialty occupations.

66

Table 3-4 Odds ratios of having a high quality diet associated with the price ratio, and with the prices of healthy foods and unhealthy foods, stratified by income-wealth and by education (n=2,765)

Exposure of interest Healthy-to-unhealthy ratio Healthy food price Unhealthy food price 95% CI 95% CI 95% CI OR Lower Upper p-value OR Lower Upper p-value OR Lower Upper p-value Wealth/income tertile a

Lowest (1-4), n=956 0.79 0.58 1.06 0.1149 0.99 0.76 1.28 0.9200 1.16 0.90 1.49 0.2654

Middle (5-6), n=793 0.61 0.43 0.87 0.0067 1.13 0.81 1.57 0.4746 1.32 1.02 1.70 0.0365

Highest (7-8), n=893 0.87 0.64 1.18 0.3621 1.11 0.83 1.49 0.4846 1.10 0.89 1.36 0.3825

Education level b

HS degree or less, n=777 0.85 0.58 1.24 0.3897 0.79 0.58 1.09 0.1513 0.90 0.65 1.25 0.5354

Some college, n=761 0.81 0.57 1.16 0.2552 1.11 0.78 1.59 0.5586 1.20 0.89 1.62 0.2339

Bachelor’s degree or more, n=1227 0.77 0.59 0.99 0.0412 1.15 0.91 1.47 0.2512 1.18 0.99 1.41 0.0621

a p-value for interaction between the income-wealth and the price ratio p = 0.0012; healthy price p=0.4047 ; unhealthy price p=0.0053 b p-value for interaction between the education level and the price ratio p = 0.0949; healthy price p=0.4415; unhealthy price p=0.2111

67

Table 3-5 Instrumental variable analysis results using toilet paper as the instrument and two-stage residual inclusion models for the healthy-to-unhealthy price ratio, healthy food price, and unhealthy food price (n=2,765) a

Stage 1 Stage 2 Price outcome

N N

events t-value F-

statistic

Odds Ratio

Lower CLb

Upper CLb

Healthy-to-unhealthy price ratio

2765 543 -33.36 1113 0.82 0.57 1.19

Healthy food price 2765 543 59.86 3583 1.15 0.91 1.45

Unhealthy food price 2765 543 98.59 9720 1.10 0.93 1.30

a Covariates in each stage included age, gender, geographic region, wealth/income index, race, smoking status, neighborhood SES, supermarket density, population density, and cost of living index b Confidence limits were obtained from a bootstrapped analysis using 10,000 replications

68

CHAPTER 4: THE PRICE OF HEALTHY FOOD RELATIVE TO UNHEALTHY FOOD AND ITS ASSOCIATION WITH TYPE 2 DIABETES AND INSULIN RESISTANCE: THE MULTI-ETHNIC STUDY OF ATHEROSCLEROSIS

ABSTRACT

OBJECTIVE: To examine the association between the price of healthy food relative to

unhealthy food and type 2 diabetes prevalence, incidence and insulin resistance (IR).

METHODS: Data came from the Multi-Ethnic Study of Atherosclerosis exam 5

administered 2010-2012 (exam 4, five years prior, was used only for diabetes incidence)

and supermarket food/beverage prices derived from Information Resources Inc. For each

individual, average price of healthy foods, unhealthy foods and their ratio was computed

for supermarkets within 3-miles. Type 2 diabetes status was confirmed at each exam and

IR was assessed via the homeostasis model assessment index. Logistic, modified Poisson

and linear regression models were used to model diabetes prevalence, incidence and IR,

respectively as a function of price and covariates; 2,353 to 3,408 participants were

included in analyses (depending on the outcome).

RESULTS: A higher ratio of healthy-to-unhealthy food was associated with greater IR

(4.8% increase in IR for each standard deviation higher price ratio [0.048, 95% CI 0.00 to

0.10] after adjusting for region, age, gender, race/ethnicity, family history of diabetes,

income/wealth index, education, smoking status, physical activity, and neighborhood

socioeconomic status). No association with diabetes incidence (relative risk=1.11, 95%

CI 0.85 to 1.44) or prevalence (odds ratio=0.95, 95% CI 0.81 to 1.11) was observed.

69

CONCLUSIONS: Higher prices of healthy food relative to unhealthy food were

positively associated with IR, but not with either of the diabetes outcomes. This study

provides new insight into the relationship between food prices with IR and diabetes.

70

4.1 INTRODUCTION

An estimated 29 million individuals are living with diabetes, accounting for more

than 9% of the population (Centers for Disease Control and Prevention, 2014), and the

prevalence of diagnosed diabetes has grown sharply since 2000 (Centers for Disease

Control and Prevention, 2016). It is vital that we identify potential areas of intervention to

reverse this trend.

Type 2 diabetes is mainly driven by obesity, and diet quality plays an important

role in its development. Diets high in unhealthy, sugary, energy-dense and nutrient-poor

foods are associated with an increased risk of obesity and diabetes (Malik et al., 2010b,

Brunner et al., 2008), while consumption of healthier foods – such as fresh fruits and

vegetables and dairy – has shown protective effects (Mursu et al., 2014, Margolis et al.,

2011, Choi et al., 2005).

Unfortunately, foods that are energy dense and generally considered unhealthy

tend to be the most affordable (Drewnowski, 2010, Drewnowski and Darmon, 2005,

Lipsky, 2009, Andrieu et al., 2006, Monsivais et al., 2010, Kern et al., 2016). The price

of food is associated with food purchasing decisions and consumption (Connors et al.,

2001, Glanz et al., 1998, Andreyeva et al., 2010, Aggarwal et al., 2016), and thus price

differences between healthy and unhealthy foods are expected to be associated with

downstream effects of diet, including diabetes and insulin resistance.

Prior work that has examined prices of healthy and unhealthy foods and health

outcomes have focused on the relationship between food prices and body weight. Most

71

(Powell et al., 2007a, Powell and Bao, 2009, Chaloupka and Powell, 2009, Chou et al.,

2004, Finkelstein et al., 2014) but not all (Beydoun et al., 2008, Han and Powell, 2011) of

these studies have found higher healthier food prices associated with higher BMI/obesity

and/or higher unhealthier prices associated with lower BMI/obesity.

No studies have examined food price association with type 2 diabetes and only a

few studies have linked food prices directly to insulin resistance. Those studies focused

primarily on metropolitan-area fast-food prices (hamburger, pizza, fried chicken) (Duffey

et al., 2010, Rummo et al., 2015, Meyer et al., 2014). Duffey et al (2010) found

metropolitan area prices inversely associated with insulin resistance (Duffey et al., 2010),

while Rummo et al (2015) found no association (Rummo et al., 2015) and Myer et al

(2014) found associations only among less advantaged residents(Meyer et al., 2014). No

work to date has examined food prices at stores nearby residents and examined

associations with type 2 diabetes and insulin resistance.

This study spatially linked participants from a large multi-ethnic sample to nearby

supermarkets to examine associations between neighborhood food price and type 2

diabetes prevalence, type 2 diabetes incidence and insulin resistance.

4.2 METHODS

4.2.1 MESA data

This study utilized data gathered by the Multi-Ethnic Study of Atherosclerosis

(MESA). The MESA is a population-based longitudinal cohort study of ethnically diverse

adults aged 45-84 years (Bild et al., 2002). Individuals were recruited from six sites

72

across the United States: Bronx/Upper Manhattan, NY; Baltimore City and Baltimore

County, Maryland; Forsyth County, North Carolina; Chicago, Illinois; St. Paul,

Minnesota; and Los Angeles County, California. The first examination was conducted in

2000-2002 and four examinations occurred during 10 years of follow-up. Food prices

were only available concurrent to exam 5, thus this study only included participants who

completed exam 5 (April 2010-January 2012). Participant characteristics for the incident

diabetes analysis were gathered from exam 4 (which occurred five years earlier: 2005-07,

considered the “baseline” in this analysis) while exam 5 was used for all other analyses.

4.2.2 Outcomes - Diabetes and insulin resistance

Type 2 diabetes status was based on the 2003 American Diabetes Association

criteria, defined as fasting glucose ≥126 mg/dl and/or use of antidiabetic

medications(2003). Prevalent type 2 diabetes at exam 5 was defined as meeting the

diabetes criteria at any of the five examinations. Incident type 2 diabetes at exam 5 was

defined as meeting the diabetes criteria at the time of exam 5 with no evidence of

diabetes at any prior exam.

Fasting glucose was measured by rate reflectance spectrophotometry using thin-

film adaptation of the glucose oxidase method on the Vitros analyzer (Johnson &

Johnson Clinical Diagnostics, Rochester, NY). Insulin was measured in serum or EDTA,

heparin or citrate plasma on a Roche Elecsys 2010 Analyzer (Roche Diagnostics

Corporation) using a sandwich immunoassay method. Insulin resistance at exam 5 was

measured according to the homeostasis model assessment index of insulin resistance

73

(HOMA-IR). This index is well correlated with measures from the gold-standard

hyperinsulinemic clamp and is calculated as (Matthews et al., 1985):

𝐻𝐻𝐻𝐻 𝐼𝐼 =𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝐹𝑔𝑔𝑔𝑔𝐹𝑔 �𝑚𝑚𝑔𝑔𝐿 � × 𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝐹𝐹𝐹𝑔𝑔𝐹𝐹 (𝐹𝐹 𝑚𝐹𝑔𝑚𝑔𝑔𝐹𝐹𝐹𝐹 𝑝𝑔𝑚 𝑔𝐹𝐹𝑔𝑚)

22.5

A log transformation was used to reduce skewness of the distribution and created

a normal distribution of values for use in multivariable analyses. See Appendix 3.2 for

details regarding the validity of the HOMA-IR measurement used in this study.

4.2.3 Covariates

Other MESA person-level data included in this study were: diet (see details

below), age (continuous), sex, race/ethnicity (Caucasian, Chinese-American, African-

American, Hispanic), family history of diabetes, smoking status (never, former, current),

education level (high school diploma, GED, or less; some college, technical or associates

degree; bachelor’s degree or higher), income/wealth index (an ordinal measure ranging

from 0 to 8 based on the combination of income level and ownership of four assets: car,

home, land, and investments), and physical activity (measured as total hours of physical

activity per day; operationalized into tertiles for modeling).

4.2.4 Price data

Data on food prices were obtained from Information Resources Inc. (IRI) (IRI,

2014, IRI, 2015, Bronnenberg et al., 2008, Symphony IRI, 2015), a market research

group that monitors prices of 299 consumer packaged goods sold in large chain

supermarkets and superstores across the U.S.(IRI, 2014, IRI, 2015, Bronnenberg et al.,

74

2008, Symphony IRI, 2015). Twenty-nine parent companies owned 100 supermarket

companies in our study area. Among those, 8 parent companies (that owned 28

companies) did not agree to release their stores’ data and no further information was

available for those companies. Data used in this study were from 794 stores located in 11

states (including Washington DC), 72 counties, and 757 census block groups. Data years

were 2009-2012.

Nine food/beverage product categories were selected to serve as proxies for either

healthier or unhealthier foods. Because data for fresh fruits and vegetables were not

available, refrigerated products were selected in order to roughly approximate costs of

fresh fruit and vegetable spoilage and storage/distribution, and proxy fresh produce.

Healthier food was represented by dairy (refrigerated milk, yogurt, cottage cheese), fruits

and vegetables (frozen vegetables, and fresh orange juice). While the healthfulness of

orange juice is questionable, the product was chosen to represent healthy foods due to its

high correlation with the price of fresh oranges (Morris, 2011) not for its own nutritional

value. Unhealthier food was represented by packaged, highly processed, long-shelf life

products: soda, sweets (chocolate candy, cookies), and salty snacks.

The primary exposure of interest was the price of healthy foods relative to

unhealthy foods, which was operationalized as the ratio of the average price per serving

of healthy food divided by the average price per serving of unhealthy food and referred to

as the healthy-to-unhealthy price ratio. The price of healthy foods and unhealthy foods

were also modeled separately as secondary exposures of interest. Average prices of

healthy and unhealthy foods were calculated using weights for each product class based

75

on national consumption estimates (i.e., food types – e.g., fruits and vegetables – that are

consumed more received larger weights) and converted to z-scores. A one unit change in

the z-scores for the price ratio were equivalent to a 14% change in the price of healthy

food relative to unhealthy food while a one unit change in the z-scores for healthy and

unhealthy foods represented differences of $0.04 and $0.03 per serving, respectively. A

sensitivity analysis which calculated overall healthy and unhealthy food prices using

equal weights for each product class (e.g., dairy) was also performed and model results

were similar to the results using the original weighting methodology.

4.2.5 Other data sources

US Census data came from the American Community Survey (ACS) 2007-2011.

Geographic regions (Northeast, Midwest, South, West) and population density 5 were

assigned to each participant. A neighborhood block group SES index was created using

six variables representing wealth and income, housing value, education, and managerial

or professional occupations, and was operationalized as a single continuous measure as

described by Diez-Roux et al (Diez Roux et al., 2001).

Addresses of MESA participants and supermarket addresses from the pricing

dataset were used to link individuals to the average food/beverage prices at supermarkets

within a 3-mile buffer (radius) of each MESA participant residence using ArcGIS 10.0.

Median number of supermarkets per MESA participant in the analytic sample was 5

5 Population density per square mile was obtained using the 2010 census. Total population within a 1 mile buffer was calculated based on block-level census population and each block was weighted by the percent of the block area that fell within the participant buffer. The total population within that block was then multiplied by this weight and the weighted populations were summed together for the total population within the buffer. The total population was divided by total buffer area in square miles.

76

(25th-75th percentile 3-6). A 3-mile radius was used for consistency with other research

examining neighborhood food environments, and is in line with prior research estimating

the average distance individuals travel to their primary supermarket (2014, Drewnowski

et al., 2012, Fuller et al., 2013, Michimi and Wimberly, 2010).

4.2.6 Statistical models

The following sequence of models was examined for each outcome: unadjusted

for covariates (model 0), and then adjusted for potential confounders: model 1,

geographic region, age, gender, and family history of diabetes; model 2, plus

income/wealth index, education, race, smoking status and physical activity; and model 3,

plus neighborhood level SES.

The analysis of diabetes prevalence included all individuals completing exam 5

and who lived within 3 miles of a supermarket from our dataset. Individuals with type 1

diabetes, those missing data for covariates, and those missing type 2 diabetes status at

exam 5 were excluded (n=258). For diabetes prevalence, logistic regression was used to

compute odds ratios (ORs) and 95% confidence intervals (CIs).

The analysis of diabetes incidence included all individuals free of type 2 diabetes

at exam 4 and who lived within 3 miles of a supermarket at any time between exam 4 and

exam 5. Of those included for analysis, 97% lived within 3 miles of a supermarket for the

entire duration between the two exams, and just 1.3% lived at an address near a

supermarket for less than half the time. A modified Poisson regression model with robust

error variance(Zou, 2004) was used to measure the relative risk of developing diabetes

77

between exam 4 and 5 for each unit increase in the z-score of the price exposure.

Covariates were measured during exam 4 and included the same variables described

above. Risk ratios (RRs) and 95% CIs are reported.

The analysis of insulin resistance levels included all individuals free of type 2

diabetes at exam 5 and with complete data for fasting insulin and glucose levels. HOMA-

IR levels were assessed using a normal linear model. Because HOMA-IR was heavily

skewed a log transformation was used and the resulting distribution was normal, and thus

the log-transformed HOMA-IR values were modeled. Regression coefficients of the

exposure illustrate the mean change in the log of HOMA-IR (reported as the percent

change of HOMA-IR for more meaningful interpretation) for every one unit change in

the z-score of the price exposure (healthy price, unhealthy price, or the ratio). Log-linear

effect estimates and 95% CIs are reported.

There was no evidence of non-linearity in the association between prices and the

outcome variables, thus prices were analyzed as continuous exposures, operationalized as

z-scores, in all analytic models. Tertiles of healthy-to-unhealthy price ratio are used in

tables 1 and 2 for descriptive purposes.

The multiplicative interaction 6 between individual SES – measured separately by

the income/wealth index and education level – with the price ratio was examined,

adjusting for all covariates from model 3, for each outcome of interest. Our hypothesis

was that the effect of price on the outcomes would be stronger within lower levels of

6 Departure from multiplicative assumption was tested within the regression models. Departure from additivity was also tested (results not shown). Additive results were similar for the prevalence models but additivity in incidence models was not fully assessed due to model convergence problems

78

SES. To illustrate how the effect differed by level of SES, stratified analyses were

performed within each tertile of income/wealth (low, medium, and high) and education

(high school degree or less, some college, or bachelor’s degree or more). Similarly, the

analysis tested for an interaction between race with the price ratio and a stratified analysis

was conducted within each race category (White, Chinese, Black, and Hispanic).

All analyses were performed using SAS software version 9.3 (SAS Institute,

Cary, NC).

4.2.7 Sample

Of the 4,716 individuals completing exam 5 there were 3,666 living within 3

miles of a supermarket in our dataset. Participants retained for analyses of each outcome

were: 3,408 for diabetes prevalence (excluded type 1 diabetes n=7, missing data for

covariates n=181, and unknown diabetes status at exam 5 n=70); 2,829 for diabetes

incidence (excluded prevalent or unknown diabetes status at exam 4); 2,353 for HOMA-

IR (excluded diabetes n=680 and those missing glucose or insulin data n=336). More

than half of the 1,308 excluded from prevalence analyses were from the St. Paul site (a

site with many Hispanics), as supermarket data was not available for this area. This led to

a higher proportion of excluded participants being Hispanic (see APPENDIX 3

Supplemental Table 12 for comparison of excluded and included). In addition, those

excluded had slightly higher BMI and education; but otherwise excluded and included

were generally similar.

79

4.3 RESULTS

Demographics and other characteristics of the MESA participants are reported in

Table 4-1. Overall the serving price of healthy food was nearly twice as expensive as

unhealthy food (mean±SD of healthy-to-unhealthy price ratio = 1.97±0.14; $0.61±$0.04

vs. $0.31±$0.03 average price per serving for healthy and unhealthy food, respectively).

Large difference across the healthy-to-unhealthy price ratio tertiles was only apparent for

region (highest ratio in west which was primarily the Los Angeles area and lowest in

south which was primarily North Carolina and Baltimore) and for race/ethnicity (Chinese

resided in highest ratio areas and Black in lowest ratio areas) (Table 4-1). Participants

with the lowest income (primarily Hispanic and Chinese) lived in areas with the highest

ratio. There were no apparent differences in the proportion of individuals with prevalent

type 2 diabetes or levels of insulin resistance across tertiles of the price ratio (Table 4-2).

There appeared to be a slight increase in the rate of incident diabetes as the price ratio

increased (4.8%, 5.2%, and 6.4% from lowest to highest).

4.3.1 Multivariable models

Results from the analytic models adjusting for covariates are shown in Table 4-3.

Our hypotheses were that the ratio of healthy-to-unhealthy price would be positively

associated with insulin resistance and diabetes. The healthy-to-unhealthy price ratio was

modestly associated with the level of insulin resistance: a one unit increase in the price

ratio z-score was associated with a 5.2% higher HOMA-IR score (Estimate=0.051, 95%

CI=[0.00 to 0.10], adjusted for age, gender, race/ethnicity, family history of diabetes,

80

income/wealth index, education level, smoking status, and physical activity). After

additional adjustment for neighborhood SES, the estimate remained largely the same

though confidence intervals included the null (0.047, 95% CI=[-0.00 to 0.10]). For

diabetes prevalence and incidence, there was no evidence of an association with healthy-

to-unhealthy ratio (OR model 3 =0.95, 95% CI=[0.81 to 1.11] and adjusted RR=1.11,

95% CI=[0.85 to 1.44], respectively).

Separate models for healthy and unhealthy food prices yielded similar results to

those reported above: significant negative associations with insulin resistance were

detected, while no significant associations were detected with either diabetes outcome.

As expected, unhealthier food price was negatively associated with HOMA-IR; but

contrary to expectation, healthier food price was also negatively associated with HOMA-

IR.

There was no evidence of effect modification -- by income/wealth, education or

race -- of the healthy-to-unhealthy price ratio association with any of the three outcome

measures (p for interactions ranged 0.07 to 0.82). See APPENDIX 3 Supplemental Table

13 for results.

4.4 DISCUSSION

This study of healthy-to-unhealthy food price ratios at local supermarkets found

positive associations with insulin resistance, indicating that residents living in areas with

larger price differentials between healthier and unhealthy food have greater insulin

81

resistance. No association was found between healthy-to-unhealthy food prices and

prevalence or incidence of type 2 diabetes.

No previous studies have examined the price of unhealthy foods at nearby grocery

stores and insulin resistance, but the association detected between local food prices and

insulin resistance in this study is consistent with prior research linking insulin resistance

to other neighborhood factors including the availability of healthy food and physical

activity resources (Auchincloss et al., 2008), neighborhood deprivation (Andersen et al.,

2008), and neighborhood SES (Diez Roux et al., 2002). Prior work has examined fast

food price and insulin resistance in a cohort of young adults (CARDIA) and found mixed

results: one study found expected results with soda and pizza prices and insulin resistance

(Duffey et al., 2010) but results were null in another study (Rummo et al., 2015) or

significant only within relatively disadvantaged groups (middle income range, lower

education, Black (Meyer et al., 2014)).

In our study, no significant association between price and diabetes was observed.

This may not be surprising given the chronic nature of diabetes. The lack of association

with prevalence is especially unsurprising given that diabetes may have been diagnosed

in the far past and an individual’s most recent price exposure may not be reflective of the

exposures faced many years prior. However, an association with diabetes incidence was

more likely to be observed given the concurrent timing of diagnosis and the price

exposure. In fact, the effect trended in the expected direction, but was not significant due

to limited power and wide confidence intervals. And, while the price exposure was

measured synchronously with the diagnosis of diabetes, risk factors of diabetes, such as

82

diet and weight, tend to be chronic exposures, and not limited to the time period during

which the disease was diagnosed. Thus, while we measured the price of food around the

time of first diagnosis of diabetes in the incidence analysis, it would have been beneficial

to know what the long-term exposures to food prices were, but these data were not

available. There is little prior literature with which to compare to our results. Studies

examining NHANES data suggested that higher prices of healthy and low glycemic foods

were associated with higher blood sugar levels in those with diabetes (Anekwe and

Rahkovsky, 2014) and without (Rashad, 2007). Compared to our dataset, the NHANES

population was younger, less educated and had a higher proportion of White individuals.

Market region food prices from 10 distinct food groups obtained from Nielsen Homescan

consumer surveys were used in one study (Anekwe and Rahkovsky, 2014) while the

other used aggregate national food prices for four high glycemic index foods from the

Bureau of Labor and Statistics (Rashad, 2007). Thus, our study differentiates itself from

prior work by using prices found at stores within the immediate neighborhood of

individuals.

Food price is just one piece of a complex food environment, and the lack of

association or weak association between food/beverage prices ratio and health may be

due to the influence of other factors such as the neighborhood built environment

(Christine et al., 2015, Auchincloss et al., 2008, Diez Roux and Mair, 2010), food

marketing(Zimmerman, 2011) and nutrition information in retail environments (Campos

et al., 2011), among others, which were not captured in this study.

83

Intervention studies suggest that price interventions modify purchases of the

targeted foods (Waterlander et al., 2012, Harnack et al., 2016), yet nutritional quality

does not necessarily improve due to substitution with other untaxed less healthy items

(Epstein et al., 2012, Mursu et al., 2014, Ni Mhurchu et al., 2010). Combining subsidies

for healthy foods with restrictions or increased prices of unhealthy foods can partially

address negative substitutions (Harnack et al., 2016, Thow et al., 2014, Epstein et al.,

2012, Duhaney et al., 2015, Niebylski et al., 2015). For this reason, our study focused

mostly on the price ratio of healthy-to-unhealthier foods/beverages as the main exposure.

Separately modeled estimates of the adjusted association between unhealthier and

healthier prices on HOMA-IR indicated lower levels of insulin resistance for both (when

prices of unhealthier soda, sweets and salty snacks were more expensive and, counter-

intuitively, when price of healthier food was more expensive). Prior work suggests that

incentivizing purchase and use of healthier foods may have weaker impacts relative to

disincentivizing the purchase of unhealthy foods (Epstein et al., 2010, Mayne et al.,

2015). Associations between availability and price of healthier foods/beverages and

health are complex and potentially mediated or moderated by available time, education,

and skills in food preparation (Duhaney et al., 2015, Niebylski et al., 2015).

4.4.1 Limitations

Requiring individuals to live within 3 miles of a supermarket in our dataset may

have introduced selection bias. An examination of the characteristics of those included

and excluded found differences with region and race/ethnicity, which was due to those

excluded being largely concentrated in a single recruitment site where supermarket data

84

were unavailable, as well as differences in education and income levels. Data on fresh

fruits and vegetables were not available and instead a proxy was refrigerated products

(frozen vegetables and refrigerated orange juice and dairy). Lastly, data were limited to

branded products in order to include products that are available in many regions of the

country.

4.5 CONCLUSIONS

As the prevalence of type 2 diabetes increases we must understand what factors

are contributing to the development of this disease and what steps need to be taken to

reverse the trend. While a large amount has been written regarding the hypothetical effect

of subsidizing and taxing foods to influence consumption and the resulting health effects,

very little work has examined real world data to understand how changes in price effect

rates of diabetes and levels of insulin resistance. This study was the first of its kind to

examine the association between neighborhood food prices with insulin resistance and

diabetes for individuals in multiple areas across the U.S. While prior work has examined

food prices at the metropolitan level, current literature regarding the direct association

between local neighborhood food prices and diabetes is lacking, and our study provides

the only currently published data examining this relationship.

85

TABLES AND FIGURES - 4

Table 4-1 MESA participant characteristics by healthy-to-unhealthy price ratio at the time of exam 5 (N=3,408)

All

participants

Lowest healthy-to-unhealthy

ratio (1.55 - 1.88)

Middle healthy-to-unhealthy

ratio (1.88 - 2.01)

Highest healthy-to-unhealthy

ratio (2.01 - 2.39)

N /

Mean Col %

/ SD N /

Mean Row

% / SD N /

Mean Row

% / SD N /

Mean Row

% / SD Number of participants (N) 3408 1148 1141 1119 MESA location (n, %)

Forsyth County, NC 645 18.9% 295 45.7% 349 54.1% 1 0.2% New York, NY 665 19.5% 322 48.4% 273 41.1% 70 10.5% Baltimore, MD 570 16.7% 420 73.7% 147 25.8% 3 0.5% St. Paul, MN 7 0.2% 6 85.7% 0 0.0% 1 14.3% Chicago, IL 814 23.9% 102 12.5% 333 40.9% 379 46.6% Los Angeles, CA 707 20.7% 3 0.4% 39 5.5% 665 94.1%

Region of residence (n, %) Northeast 660 19.4% 318 48.2% 273 41.4% 69 10.5% Midwest 797 23.4% 95 11.9% 331 41.5% 371 46.5% South 1236 36.3% 732 59.2% 496 40.1% 8 0.6% West 715 21.0% 3 0.4% 41 5.7% 671 93.8%

Age (mean, SD) 70.2 9.4 70.4 9.2 70.4 9.4 69.8 9.8 Female (n, %) 1809 53.1% 603 33.3% 636 35.2% 570 31.5% Race/ethnicity (n, %)

White 1274 37.4% 475 37.3% 563 44.2% 236 18.5% Chinese American 481 14.1% 10 2.1% 89 18.5% 382 79.4% Black 1072 31.5% 549 51.2% 321 29.9% 202 18.8% Hispanic 581 17.0% 114 19.6% 168 28.9% 299 51.5%

Family history of diabetes (n, %) 1529 44.9% 546 35.7% 490 32.0% 493 32.2% BMI (mean, SD) 28.2 5.6 29.1 5.6 28.3 5.8 27.1 5.2

<25 (n, %) 1040 30.5% 277 26.6% 337 32.4% 426 41.0% 25-29.9 (n, %) 1283 37.6% 422 32.9% 446 34.8% 415 32.3% ≥30 (n, %) 1085 31.8% 449 41.4% 358 33.0% 278 25.6%

Education level (n, %) HS/GED or less 1022 30.0% 310 30.3% 331 32.4% 381 37.3% Some college, Technical or Associate degree 945 27.7% 328 34.7% 306 32.4% 311 32.9%

86

All

participants

Lowest healthy-to-unhealthy

ratio (1.55 - 1.88)

Middle healthy-to-unhealthy

ratio (1.88 - 2.01)

Highest healthy-to-unhealthy

ratio (2.01 - 2.39)

N /

Mean Col %

/ SD N /

Mean Row

% / SD N /

Mean Row

% / SD N /

Mean Row

% / SD Bachelor’s degree or higher 1441 42.3% 510 35.4% 504 35.0% 427 29.6%

Per capita income (in $10k) 2.6 1.9 2.7 1.8 2.8 2.0 2.2 1.8 Wealth index 2.6 1.2 2.6 1.2 2.6 1.2 2.5 1.2 Income/wealth index 5.0 2.2 5.2 2.1 5.2 2.2 4.7 2.4 Marital status (n, %)

Not married or living with partner 1395 40.9% 533 38.2% 474 34.0% 388 27.8% Married/Living with partner 2013 59.1% 615 30.6% 667 33.1% 731 36.3%

Smoking status (n, %) Never smoked 1575 46.2% 444 28.2% 507 32.2% 624 39.6% Former smoker 1585 46.5% 598 37.7% 556 35.1% 431 27.2% Current smoker 248 7.3% 106 42.7% 78 31.5% 64 25.8%

Physical activity, hours per day (mean, SD) 8.8 5.4 9.5 5.8 9.1 5.3 7.7 4.8 Daily calorie intake (mean, SD) 1643 782 1668 793 1641 765 1619 789

87

Table 4-2 Proportion of participants with type 2 diabetes, incident type 2 diabetes, and insulin resistance levels by tertile of the healthy-to-unhealthy price ratio and the average serving price of healthy and unhealthy foods

All participants Lowest ratio (1.55 - 1.88)

Middle ratio (1.88 - 2.01)

Highest ratio (2.01 - 2.39)

Number of participants (N) 3,408 1,148 1,141 1,119 Number with diabetes at exam 5 (n, %) 790 23.2% 275 24.0% 247 21.6% 268 23.9% Number with incident diabetes at exam 5 (n, %) a 154 5.4% 45 4.8% 49 5.2% 60 6.4% Insulin resistance level at exam 5 (mean, SD, median) b 2.32 0.67 2.33 2.31 0.66 2.32 2.32 0.68 2.33 2.31 0.69 2.34 Average food prices per serving

Healthy food price per serving (mean, SD, median) $0.61 $0.04 $0.60 $0.60 $0.06 $0.57 $0.63 $0.05 $0.61 $0.59 $0.03 $0.58 Unhealthy food price per serving (mean, SD, median) $0.31 $0.03 $0.30 $0.33 $0.04 $0.31 $0.32 $0.03 $0.31 $0.30 $0.01 $0.30 Ratio of Healthy-to-Unhealthy (mean, SD, median) 1.97 0.14 1.94 1.83 0.05 1.85 1.96 0.07 1.98 1.95 0.05 1.94

a percentages are within the analytic population for the incidence analysis, n=2,829 b values are from within the analytic population for the insulin resistance analysis, n=2,353; insulin resistance is measured as the log of HOMA-IR

88

Table 4-3 Multivariable models for each of the three outcomes of interest: prevalence of type 2 diabetes at exam 5, incidence of type 2 diabetes from exam 4 to exam 5, and level of insulin resistance at exam 5 a

Exposure of interest Healthy-to-unhealthy ratio Healthy food price Unhealthy food price

95% CI 95% CI 95% CI

Type 2 diabetes prevalence (n=3,408) OR Lower Upper p-value OR Lower Upper p-value OR Lower Upper p-value

Model 0: No covariates 1.003 0.927 1.085 0.9443 0.943 0.869 1.022 0.1529 0.949 0.874 1.031 0.2180 Model 1: Region, age, gender, race, family history of diabetes 0.959 0.824 1.116 0.5862 0.888 0.784 1.005 0.0589 0.948 0.855 1.051 0.3128 Model 2: SES, smoking, PA 0.968 0.831 1.129 0.6796 0.896 0.791 1.016 0.0862 0.950 0.856 1.055 0.3375 Model 3: Neighborhood SES 0.948 0.812 1.106 0.4984 0.931 0.816 1.061 0.2839 0.981 0.879 1.093 0.7233

Type 2 diabetes incidence (n=2,829) RR Lower Upper p-value RR Lower Upper p-value RR Lower Upper p-value

Model 0: No covariates 1.179 1.025 1.355 0.0208 0.985 0.855 1.134 0.8285 0.828 0.684 1.002 0.0530 Model 1: Region, age, gender, race, family history of diabetes 1.146 0.882 1.490 0.3066 0.864 0.698 1.071 0.1821 0.857 0.695 1.057 0.1485 Model 2: SES, smoking, PA 1.138 0.874 1.481 0.3367 0.868 0.700 1.077 0.1981 0.862 0.698 1.064 0.1659 Model 3: Neighborhood SES 1.107 0.850 1.441 0.4503 0.941 0.754 1.174 0.5885 0.912 0.734 1.134 0.4075

Insulin resistance (n=2,353) Estimate Lower Upper p-value Estimate Lower Upper p-value Estimate Lower Upper p-value Model 0: No covariates -0.007 -0.034 0.020 0.6128 -0.025 -0.052 0.003 0.0768 -0.014 -0.041 0.013 0.3023 Model 1: Region, age, gender, race, family history of diabetes 0.051 0.002 0.100 0.0419 -0.052 -0.091 -0.012 0.0101 -0.046 -0.079 -0.014 0.0050 Model 2: SES, smoking, PA 0.051 0.002 0.100 0.0410 -0.051 -0.090 -0.011 0.0115 -0.046 -0.078 -0.014 0.0055

89

Exposure of interest Healthy-to-unhealthy ratio Healthy food price Unhealthy food price

95% CI 95% CI 95% CI

Model 3: Neighborhood SES 0.047 -0.002 0.097 0.0623 -0.046 -0.087 -0.004 0.0307 -0.042 -0.076 -0.008 0.0143 CI, confidence interval; OR, odds ratio; RR, risk ratio; SES, socioeconomic status; PA, physical activity. a Model 0 includes only the price exposure; Model 1 covariates include region, age, gender, race/ethnicity, and family history of diabetes; Model 2 covariates include those in Model 1 plus income/wealth index, education level, smoking status, and physical activity; Model 3 covariates include those in Model 2 plus neighborhood SES

90

CHAPTER 5: DISCUSSION

5.1 DISCUSSION

The affordability of healthy food relative to unhealthy food may be a key

consideration for individuals when deciding what foods to purchase at their local

supermarket. These purchasing decisions may directly affect their diet quality and may

lead to considerable downstream health outcomes, including obesity, insulin resistance

and diabetes. This study linked individuals from the Multi-Ethnic Study of

Atherosclerosis (MESA) – a large, multi-site, longitudinal cohort study – to supermarkets

close to their home. We examined the relationship between the price of foods at these

supermarkets with outcomes collected from MESA, including the quality of diet, level of

insulin resistance, and presence of diabetes. We also examined the potential effect

modification of socioeconomic status on this relationship.

In the first aim of this study we found that across 1,953 supermarkets the average

price per serving of healthy food was nearly twice as high as unhealthy food ($0.590 for

healthy food vs. $0.298 for unhealthy food per serving). In fact, the average per serving

price of healthy food was more expensive than unhealthy food in every one of the

supermarkets examined (range of the healthy-to-unhealthy ratio: 1.48 – 2.45). The

evaluation of neighborhood SES and neighborhood Black/Hispanic showed no strong

association with food prices, including the healthy-to-unhealthy price ratio, and the

individual composite prices of healthy and unhealthy foods. Notable associations were

found when examining individual product categories within the healthy food domain.

The price of dairy, for example, was higher in neighborhoods of lower SES and

91

neighborhoods with a higher proportion of Black/Hispanic individuals, while the price of

fruits and vegetables was lower in those same neighborhoods. A similar examination of

products within the unhealthy food domain yielded little; there was little variation in

price of any unhealthy food products. The relative price of healthy food versus unhealthy

food was slightly more affordable as the proportion of Black/Hispanic increased after

adjusting for neighborhood SES and other covariates, while there was no evident

association between neighborhood SES and the price ratio in covariates adjusted models.

In the second aim of the study we found that higher relative prices of healthy

foods compared to unhealthy foods were found to be associated with decreased odds of

having a high quality diet. Although healthy foods were more expensive than unhealthy

foods, consistent with findings in the first aim of this research, there was no association

between diet and the absolute (i.e., non-relative) price of healthy food, while there was a

significant association with the price of unhealthy foods alone. It appears that the relative

price of healthy food compared with unhealthy alternatives is a more meaningful metric

of affordability rather than the absolute price of healthy food without consideration of

potential unhealthy substitutes. Thus, the examination of any food or food class should

consider the price of substitutes.

This study found an interaction between SES and the healthy-to-unhealthy price

ratio on the odds of having a high quality diet; however, the direction of the interaction

was not in line with our a priori hypothesis. Instead of the price ratio having more of an

effect within those of lower SES, the largest effects were observed in those with the

highest level of education and those in the middle range of the income/wealth index. One

explanation for this may be that because healthy food was twice as expensive as

92

unhealthy food on average and 55% more expensive at its most affordable, healthy food

may have been too expensive for those with the lowest income, regardless of what the

price ratio was. Those with the highest level of income and wealthy may have the

necessary means to purchase healthy food regardless of how high or low its price was

relative to unhealthy substitutes, and so fluctuations in price had little effect on their

purchasing habits. Thus, it could be those who fall in the middle of the SES spectrum that

are most sensitive to the relatively small variations in the price ratio observed in this

study. To have an impact on lower SES individuals it may require strong subsidies of

healthy foods and high taxes of unhealthy foods to make a significant difference in the

diet of the most vulnerable populations. In fact, subsidies may be the preferred

intervention as they would raise the real income of lower income individuals and make

them more responsive to price differences (McGill et al., 2015).

In the third aim of this study we detected a positive association between the

healthy-to-unhealthy price ratio and level of insulin resistance as measured by HOMA-

IR, indicating that individuals living in areas with larger price differentials between

healthy and unhealthy foods had a higher prevalence of insulin resistance. However, no

significant association was detected between the price ratio with either the prevalence or

incidence of diabetes. There was also no interaction detected between the price ratio with

any of the three outcomes. While slightly higher effects were found for those in the

middle income/wealth tertile, the confidence intervals were wide and showed no clear

separation from other levels of SES.

The null association with diabetes may not be surprising given the chronic nature

of the disease. Food prices for this study were for the years 2009-2012, yet in the

93

prevalence analysis diabetes may have been first diagnosed years or decades prior and

thus recent food prices may not have been relevant if they weren’t a reflection of the

prices faced prior to developing diabetes. Furthermore, for the analysis of incident

diabetes, the price of food was captured near the time of first diagnosis of diabetes;

however, data regarding the long-term food prices were not available, and thus we don’t

know what the exposure was for the majority of the time leading up to the first diagnosis

of diabetes. The causes of insulin resistance precede diabetes, and HOMA-IR levels – a

combination of current insulin and glucose levels – are more likely than diabetes to be a

direct result of recent exposures such as recent food prices and the subsequent diet.

Therefore, our ability to detect an association was greatest for insulin resistance, and it

may require a much greater observation period for food prices to detect an association

with the development of type 2 diabetes.

5.2 CONTEXT OF CURRENT RESEARCH WITH PREVIOUS WORK

The large absolute differences in price between healthy and unhealthy foods

observed in this study is consistent with previous work that found prices of soft drinks

much lower than fluid milk (Kern et al., 2016), and the price of sugars, fats, and oils

much lower than fruits, vegetables, meat and poultry (Drewnowski, 2010), that higher

diet costs were associated with a higher quality diet (Rehm et al., 2015), and overall,

healthy food tends to cost more than unhealthy food (Darmon and Drewnowski, 2015).

These differences are likely due to the increased costs of refrigeration, farming, and

transportation for perishables versus lower such costs for long shelf-life

packaged/processed foods. The combination of lower price and increased availability of

packaged/processed foods may be having profound effects on food purchases and result

94

in less-than-optimal diet quality across a broad spectrum of the population (Kearney,

2010). The large price divide between healthier and unhealthy foods is concerning;

dietary guidelines emphasize the consumption of fruits, vegetables, and low-fat dairy

while limiting intake of sugar, saturated fats, and sodium (United States Department of

Agriculture and United States Department of Health and Human Services, 2015), yet the

differential price of these foods may be a deterrent to achieving a healthy diet (Darmon

and Drewnowski, 2015).

While it is well established that fresh food is generally more expensive than

unhealthy food, our study is unique in that it linked individuals to their nearby

supermarkets to examine the association between food price with their diet quality. To

our knowledge, this is the first study to do so. Recent work has found that healthier diets

are more expensive than unhealthy diets (Rehm et al., 2015, Rao et al., 2013, Monsivais

et al., 2012), but these studies did not link local food prices to individual diet quality

across a geographically diverse cohort in the US. Instead, previous studies have

calculated diet costs according to national food price estimates (Rehm et al., 2015, Morris

et al., 2014) rather than food prices local to an individual’s neighborhood or have been

limited to a single city (Monsivais et al., 2012).

If healthy foods are twice as expensive as unhealthy foods, as found to be in this

study, meeting established dietary guidelines will be difficult for many people, especially

those of lower SES. A pair of recent studies utilizing data from a single city, the Seattle

Obesity Study (SOS), examined the role of individual SES and types of supermarkets

where individuals shopped and found that low SES consumers shopped at lower price

supermarkets and had a higher risk of obesity compared to those of higher SES, and that

95

SES was positively associated with diet quality as measured by the HEI (Drewnowski et

al., 2016, Drewnowski et al., 2014). A third study from the SOS determined that SES

was positively associated with food prices and that food price mediated the relationship

between SES and diet quality (Aggarwal et al., 2011). A systematic review of the

literature found that these findings are consistent across the majority of research, and that

discrepancies in the price between healthy and unhealthy food are likely a key component

of socioeconomic disparities in diet quality (Darmon and Drewnowski, 2015). Further

worsening the food environment for many lower SES and non-White individuals is a

higher prevalence of soda and fast-food advertising in neighborhoods with higher

proportion of lower income and Black/Hispanic persons (Powell et al., 2014). The union

of high availability, exposure to advertising, and lower price may contribute to high

consumption of unhealthy foods in these populations leading to poorer diets (Chandon

and Wansink, 2012, Drewnowski, 2009, Appelhans et al., 2012).

There is ample evidence that individuals of lower SES tend to have poorer, less

expensive diets compared with individuals of higher SES (Darmon and Drewnowski,

2015); however, it is not clear how neighborhood food prices may differentially effect the

diet of different groups of individuals according to their SES. A greater proportion of

income for lower SES individuals is spent on food (Frazao et al., 2007), and lower SES

individuals tend to be more price elastic consumers (Powell and Chaloupka, 2009), and

so it was hypothesized that the largest effects of price on diet would be observed for

individuals with the lowest levels of SES. However, that was not the case, and instead we

found largest effects within the middle income category consistently across our

outcomes, and this study provides insight to a previous gap in the literature regarding the

96

role SES may play in the causal pathway between food price and diet quality. This

finding is consistent with a recently published experimental economics study which

simulated food subsidies and taxes on diets across levels of SES, and found that the effect

would be greatest for those in the medium-income group rather than those with the lowest

incomes (Darmon et al., 2016).

No previous studies have examined the price of unhealthy foods at nearby grocery

stores and insulin resistance, but prior work has examined metropolitan area fast food

prices and insulin resistance in a cohort of young adults (CARDIA) and found mixed

results. One study found a negative association between soda and pizza prices and insulin

resistance (Duffey et al., 2010), while results were null in another study (Rummo et al.,

2015), and significant only within relatively disadvantaged groups (middle income range,

lower education, Black) in a third study (Meyer et al., 2014). As an investigation of

neighborhood level exposures, our results are consistent with prior research linking

insulin resistance to neighborhood availability of healthy food and physical activity

resources (Auchincloss et al., 2008), neighborhood deprivation (Andersen et al., 2008),

and neighborhood SES (Diez Roux et al., 2002).

There is little prior literature with which to compare to our results for the diabetes

outcomes, as to our knowledge no study has examined the direct association between

food prices and diabetes. One study examining NHANES data paired with food price data

from Nielsen Homescan consumer surveys found that higher prices of healthy foods were

associated with higher blood sugar levels in those with diabetes (Anekwe and

Rahkovsky, 2014). A second study utilizing NHANES data examined aggregate regional

food prices from the Bureau of Labor Statistics and found a non-statistically significant

97

inverse association between the price of high glycemic index foods and HbA1c level, and

a positive association of similar magnitude between low glycemic index food prices and

HbA1c (Rashad, 2007).

Instead of focusing on a specific condition such as diabetes or insulin resistance,

the majority of work examining the relationship between food prices and health outcomes

has focused on weight gain and BMI. These studies have largely concluded that price

changes would have weak, but non-negligible effect on weight outcomes (Powell and

Chaloupka, 2009, Finkelstein et al., 2014), and so it isn’t surprising that no strong

association was detected between food price and diabetes in our study, while modest

associations were detected with insulin resistance.

5.3 LIMITATIONS

There are some limitations of this study that are worth noting. The IRI data set did

not have prices for fresh fruits and vegetables and thus we were limited to using proxies

of fresh fruits and vegetables, specifically orange juice and frozen vegetables. The price

of orange juice is highly correlated with the price of fresh oranges (Morris, 2011) and the

prices of frozen vegetables are similar to those that are fresh, with some variation in the

amount of correlation across vegetable types (Stewart et al., February 2011), thus these

items appear to be acceptable proxies for their fresh counterparts.

Only branded products were included in the price calculations. This was done in

order to maximize product penetration across all stores included in the analysis to ensure

comparability of products across stores and avoid biasing the results due to the inclusion

of a large number of store- or region-specific products. Brands dominate market share for

98

most of the unhealthy products, including soda (Beverage World), salty snacks (Grocery

Headquarters), and chocolate candy (Grocery Headquarters). Branded products are

somewhat dominant for orange juice (Beverage Industry Magazine), but much less for dairy

(Grocery Headquarters) and frozen vegetables (Store Brands), and our price estimates for

healthy products may be less representative of the overall healthy food landscape than

our price estimates for unhealthy foods.

The products chosen to represent healthy foods in this study were those that were

to proxy the price of fresh, perishable foods; i.e., fruits, vegetables, and dairy. In doing so

we selected only a portion of all foods that can generally be considered healthy. Canned

fruits (packaged in water, not syrup) and vegetables may also be considered healthy but

were not represented in our data; however, processed fruits and vegetables, including

frozen and canned varieties are not consistently more or less expensive than fresh options

(Stewart et al., February 2011), and so our proxies are likely sufficient for these items.

Beyond fruits, vegetables and dairy, there are other healthy foods such as legumes and

whole grains that were not included in this analysis because they were not available in the

price dataset. It is not clear how our price estimates, and ultimately the results of our

study, would have differed had we had access to all possible healthy foods.

It is possible that limiting our analysis to individuals from MESA who lived

within 3 miles of an IRI supermarket may have introduced selection bias to our study.

However, there was little difference in the characteristics of those who were and were not

included with the exception of race/ethnicity and region due to the exclusion of one of the

MESA sites for which supermarket data was unavailable.

99

There are limitations to self-report dietary data obtained via a food frequency

questionnaire (FFQ), as was used in this study. One of the major limitations of an FFQ is

the underreporting of calories; however, the validity of the FFQ used in this study is well

documented and FFQs may be a better measure of long-term diet – rather than recent or

current diet – than other self-reported diet measures taken at a single point in time

(Willett and Hu, 2006). In an attempt to address underreporting across all participants we

defined a high quality diet relative to the population rather than according to an absolute

scale. If underreporting did exist in this study, we do not believe there is a theoretical

reason for the potential measurement error of the FFQ to differ by the price exposure, and

thus any measurement error is likely to be non-differential and would bias results towards

the null hypothesis rather than away from it. The FFQ used in this study was based on an

a priori list of foods that may not have captured every possible food item consumed by

every individual in the study; however, the FFQ used in the MESA study was modified

from a previous FFQ in order to reflect the diversity of the population by including many

foods not typically used that are specific to the cultures of the individuals included in the

cohort study.

Lastly, the analyses of diabetes prevalence and incidence in this study may have

been limited due to the development of the disease occurring over a long period of time

in response to chronic risk factors, while our data captured an exposure period of just a

few years. Ideally, we would have had data on neighborhood supermarket food prices of

each individual for the majority of their adult (and even childhood) life, but instead our

data were limited to a four year period from 2009 to 2012. If this small period of time

was reflective of one’s price environment for the majority of their life than the exposure

100

is much more relevant to the outcome of developing diabetes, as opposed to instances in

which the recent place of residence and affordability of foods is vastly different than what

has been faced for many years prior.

5.4 CONCLUSIONS

This dissertation provides meaningful insight to the relationship between the price

of healthy and unhealthy foods with diet quality, diabetes, and insulin resistance within

adults across the U.S. Research on the access to healthy food has gained popularity in

recent years, and was even a cause championed by the first lady Michele Obama as part

of her “Let’s Move!” campaign to fight childhood obesity and the government’s Healthy

Food Financing Initiative (Administration for Children and Families, 2016). There are

several key takeaways resulting from this work, and a few areas that would benefit from

more in depth research in the future.

While there was a large emphasis on the results examining the association

between the health-to-unhealthy price ratio with diet and health outcomes, it is important

to not lose sight of the fact that on average a serving of healthy food costs approximately

twice as much as a serving of unhealthy food. Though this finding isn’t novel on its own,

it adds to a growing collection of literature that has consistently told a similar story over

the last two decades. This large price differential between healthy and unhealthy food

creates a significant barrier to accessible food, and provides the foundation on which the

rest of the research contained in this dissertation stands. This finding reinforces the notion

that access to healthy food is more than just physical access, but also includes economic

access.

101

Second, this study found that the price of food, particularly the price of healthy

food relative to unhealthy food, as well as the price of unhealthy food itself, may have an

impact on diet quality. Diet quality plays an essential role in the development and

prevention of health complications, including obesity, diabetes, cardiovascular disease,

and many others, and the implications of the findings in this study could be immense.

Interventions that affect the price of healthy and/or unhealthy food, such as taxation and

subsidization, could result in reduced morbidity in the U.S. population, and in turn the

economic implications that would follow. In fact this study found a positive association

with insulin resistance, a disease largely mediated by one’s diet, though no significant

association was found with diabetes. Data for this study were limited to a few recent

years, but in order to more likely find an association with more chronic disease outcomes,

such as diabetes, a more historical measure of food prices may be necessary.

Furthermore, as this study was limited to a handful of types of healthy and unhealthy

foods, it is important that future research expand these categories, if the data allows, in

order to improve the generalizability and validity of these findings.

Lastly, this study hypothesized that associations between food price and each of

the outcomes of interest would differ according to an individual’s socioeconomic status.

We hypothesized that those of low SES would be the most affected by price differentials,

as they have the fewer means and are more price sensitive than those with higher levels

of income and education. However, it was instead individuals in the middle of the

income/wealth distribution whom were affected the most by price differences for every

outcome examined – diet quality, diabetes prevalence, diabetes incidence, and insulin

resistance. This is an important finding when considering interventions to make healthy

102

and unhealthy food more or less affordable, respectively. While the intention would be to

help those most in need, it is possible such interventions would not sufficiently help the

most vulnerable populations and instead could increase inequalities in nutrition between

socioeconomic groups (Darmon et al., 2016). Prior to implementing any policies aimed at

increasing the affordability of healthy foods, a thorough amount of research and

community outreach should be implemented to ensure that those who need help the most

would receive it. Fiscal interventions to influence the price of healthy and unhealthy

foods may be effective in improving diets and health; however, it is unlikely to be a

panacea for the obesity crisis. Instead, a multipath approach should be used to improve

access to a healthy diet – including educational programs to improve nutritional

knowledge, nutritional labeling on food packaging and menus, regulations on marketing

and advertising of junk food, in addition to improving physical and economic access.

103

LIST OF REFERENCES

ABDULLAH, A., PEETERS, A., DE COURTEN, M. & STOELWINDER, J. 2010. The

magnitude of association between overweight and obesity and the risk of diabetes: A meta-analysis of prospective cohort studies. Diabetes Research and Clinical Practice, 89, 309-319.

ABIEMO, E. E., ALONSO, A., NETTLETON, J. A., STEFFEN, L. M., BERTONI, A. G., JAIN, A. & LUTSEY, P. L. 2013. Relationships of the Mediterranean dietary pattern with insulin resistance and diabetes incidence in the Multi-Ethnic Study of Atherosclerosis (MESA). Br J Nutr, 109, 1490-7.

ADMINISTRATION FOR CHILDREN AND FAMILIES 2016. Healthy Food Financing Initiative brings jobs and nutritious food to low-income families. Available at http://www.acf.hhs.gov/media/press/2016/healthy-food-financing-initiative-brings-jobs-and-nutritious-food-to-low-income-families. Last accessed November 26, 2016.

AGGARWAL, A., MONSIVAIS, P., COOK, A. J. & DREWNOWSKI, A. 2011. Does diet cost mediate the relation between socioeconomic position and diet quality[quest]. Eur J Clin Nutr, 65, 1059-1066.

AGGARWAL, A., REHM, C. D., MONSIVAIS, P. & DREWNOWSKI, A. 2016. Importance of taste, nutrition, cost and convenience in relation to diet quality: Evidence of nutrition resilience among US adults using National Health and Nutrition Examination Survey (NHANES) 2007–2010. Preventive Medicine, 90, 184-192.

AGRICULTURE AND AGRI-FOOD CANADA 2012. International Markets Bureau, American eating trends report: Cookies and sweet biscuits. Available at: http://www5.agr.gc.ca/resources/prod/Internet-Internet/MISB-DGSIM/ATS-SEA/PDF/6272-eng.pdf.

AMERICAN DIABETES ASSOCIATION 2013. Standards of Medical Care in Diabetes - 2013. Diabetes Care, 36, S11-S66.

ANDERSEN, A. F., CARSON, C., WATT, H. C., LAWLOR, D. A., AVLUND, K. & EBRAHIM, S. 2008. Life-course socio-economic position, area deprivation and Type 2 diabetes: findings from the British Women's Heart and Health Study. Diabet Med, 25, 1462-8.

ANDREYEVA, T., BLUMENTHAL, D. M., SCHWARTZ, M. B., LONG, M. W. & BROWNELL, K. D. 2008. Availability and prices of foods across stores and neighborhoods: the case of New Haven, Connecticut. Health Aff (Millwood), 27, 1381-8.

ANDREYEVA, T., LONG, M. W. & BROWNELL, K. D. 2010. The Impact of Food Prices on Consumption: A Systematic Review of Research on the Price Elasticity of Demand for Food. American Journal of Public Health, 100, 216-222.

ANDRIEU, E., DARMON, N. & DREWNOWSKI, A. 2006. Low-cost diets: more energy, fewer nutrients. Eur J Clin Nutr, 60, 434-6.

104

ANEKWE, T. D. & RAHKOVSKY, I. 2014. The association between food prices and the blood glucose level of US adults with type 2 diabetes. Am J Public Health, 104, 678-85.

APPEL, L. J., MOORE, T. J., OBARZANEK, E., VOLLMER, W. M., SVETKEY, L. P., SACKS, F. M., BRAY, G. A., VOGT, T. M., CUTLER, J. A., WINDHAUSER, M. M., LIN, P. H. & KARANJA, N. 1997. A clinical trial of the effects of dietary patterns on blood pressure. DASH Collaborative Research Group. N Engl J Med, 336, 1117-24.

APPELHANS, B. M., MILLIRON, B.-J., WOOLF, K., JOHNSON, T. J., PAGOTO, S. L., SCHNEIDER, K. L., WHITED, M. C. & VENTRELLE, J. C. 2012. Socioeconomic Status, Energy Cost, and Nutrient Content of Supermarket Food Purchases. American Journal of Preventative Medicine, 42, 398-402.

AREM, H., REEDY, J., SAMPSON, J., JIAO, L., HOLLENBECK, A. R., RISCH, H., MAYNE, S. T. & STOLZENBERG-SOLOMON, R. Z. 2013. The Healthy Eating Index 2005 and risk for pancreatic cancer in the NIH-AARP study. J Natl Cancer Inst, 105, 1298-305.

AUCHINCLOSS, A. H., DIEZ ROUX, A. V., BROWN, D. G., ERDMANN, C. A. & BERTONI, A. G. 2008. Neighborhood resources for physical activity and healthy foods and their association with insulin resistance. Epidemiology, 19, 146-57.

BALL, K., CRAWFORD, D. & MISHRA, G. 2006. Socio-economic inequalities in women's fruit and vegetable intakes: a multilevel study of individual, social and environmental mediators. Public Health Nutrition, 9, 623-630.

BAZZANO, L. A., JOSHIPURA, K. J., LI, T. Y. & HU, F. B. 2008. Intake of Fruit, Vegetables, and Fruit Juices and Risk of Diabetes in Women. Diabetes Care, 31, 1311-1317.

BERKEY, C. S., ROCKETT, H. R., FIELD, A. E., GILLMAN, M. W. & COLDITZ, G. A. 2004. Sugar-Added Beverages and Adolescent Weight Change. Obesity Research, 12, 778-788.

BERNSTEIN, A. M., BLOOM, D. E., ROSNER, B. A., FRANZ, M. & WILLETT, W. C. 2010. Relation of food cost to healthfulness of diet among US women. Am J Clin Nutr, 92, 1197-203.

BEVERAGE INDUSTRY MAGAZINE Sales of the leading refrigerated orange juice brands of the United States in 2015 (in million U.S. dollars). Available at: http://www.statista.com/statistics/188749/top-refrigerated-orange-juice-brands-in-the-united-states/. Last accessed: May 18, 2016.

BEVERAGE WORLD. Market share of the leading soft drink companies in the United States in 2014. Available at: http://www.statista.com/statistics/237687/market-share-of-the-leading-us-soft-drink-companies/. Last accessed: May 18, 2016 [Online].

BEYDOUN, M. A., POWELL, L. M. & WANG, Y. 2008. The association of fast food, fruit and vegetable prices with dietary intakes among US adults: is there modification by family income? Soc Sci Med, 66, 2218-29.

BEYDOUN, M. A. & WANG, Y. 2008. Do nutrition knowledge and beliefs modify the association of socio-economic factors and diet quality among US adults? Preventive Medicine, 46, 145-153.

105

BILD, D. E., BLUEMKE, D. A., BURKE, G. L., DETRANO, R., DIEZ ROUX, A. V., FOLSOM, A. R., GREENLAND, P., JACOB, D. R., JR., KRONMAL, R., LIU, K., NELSON, J. C., O'LEARY, D., SAAD, M. F., SHEA, S., SZKLO, M. & TRACY, R. P. 2002. Multi-ethnic study of atherosclerosis: objectives and design. Am J Epidemiol, 156, 871-81.

BIRDS EYE. 2016. Birds Eye® Boxed Vegetables - Classic Mixed Vegetables - Nutrition Information. Available: http://www.birdseye.com/vegetable-products/steamfresh-pure-simple-blends/mixed-vegetables.

BLACK, J. L. & MACINKO, J. 2008. Neighborhoods and obesity. Nutr Rev, 66, 2-20. BOONE-HEINONEN, J., GORDON-LARSEN, P., KIEFE, C. I., SHIKANY, J. M.,

LEWIS, C. E. & POPKIN, B. M. 2011. Fast food restaurants and food stores: longitudinal associations with diet in young to middle-aged adults: the CARDIA study. Arch Intern Med, 171, 1162-70.

BOULTON, A. J. M., VINK, A. I., AREZZO, J. C., BRIL, V., FELDMAN, E. L., FREEMAN, R., MALIK, R. A., MASER, R. E., SOSENKO, J. M. & ZIEGLER, D. 2005. Diabetic Neuropathies: A statement by the American Diabetes Association. Diabetes Care, 28, 956-962.

BRAY, G. A. & POPKIN, B. M. 2014. Dietary Sugar and Body Weight: Have We Reached a Crisis in the Epidemic of Obesity and Diabetes? Health Be Damned! Pour on the Sugar. Diabetes Care, 37, 950-956.

BREYER, B. & VOSS-ANDREAE, A. 2013. Food mirages: geographic and economic barriers to healthful food access in Portland, Oregon. Health Place, 24, 131-9.

BRONNENBERG, B. J., KRUGER, M. W. & MELA, C. F. 2008. Database Paper—The IRI Marketing Data Set. Marketing Science, 27, 745-748.

BROWNELL, K. D., FARLEY, T., WILLETT, W. C., POPKIN, B. M., CHALOUPKA, F. J., THOMPSON, J. W. & LUDWIG, D. S. 2009. The public health and economic benefits of taxing sugar-sweetened beverages. N Engl J Med, 361, 1599-605.

BRUNELLO, G., DE PAOLA, M. & LABARTINO, G. 2014. More apples fewer chips? The effect of school fruit schemes on the consumption of junk food. Health Policy.

BRUNNER, E. J., MOSDOL, A., WITTE, D. R., MARTIKAINEN, P., STAFFORD, M., SHIPLEY, M. J. & MARMOT, M. G. 2008. Dietary patterns and 15-y risks of major coronary events, diabetes, and mortality. Am J Clin Nutr, 87, 1414-21.

BURNS, C., SACKS, G., RAYNER, M., BILENKIJ, G. & SWINBURN, B. 2010. Correctly calculating the cost of food. Nutr Rev, 68, 182-3; author reply 184-5.

CALIFORNIA CENTER FOR PUBLIC HEALTH ADVOCACY, POLICYLINK & THE UCLA CENTER FOR HEALTH POLICY RESEARCH 2008. Designed for Disease: The Link Between Local Food Environments and Obesity and Diabetes.

CAMPOS, S., DOXEY, J. & HAMMOND, D. 2011. Nutrition labels on pre-packaged foods: a systematic review. Public Health Nutrition, 14, 1496-1506.

CARLSON, A. & FRAZÃO, E. May 2012. Are Healthy Foods Really More Expensive? It depends on How You Measure the Price. U.S. Department of Agriculture, Economic Research Service, EIB-96.

106

CASSADY, D., JETTER, K. M. & CULP, J. 2007. Is Price a Barrier to Eating More Fruits and Vegetables for Low-Income Families? J Am Diet Assoc, 107, 1909-1915.

CENTERS FOR DISEASE CONTROL AND PREVENTION. 2013. Median Age at Diagnosis of Diabetes Among Adult Incident Cases Aged 18–79 Years, by Race/Ethnicity, United States, 1997–2011 [Online]. Available: http://www.cdc.gov/diabetes/statistics/age/fig4.htm [Accessed December 7, 2014.

CENTERS FOR DISEASE CONTROL AND PREVENTION 2014. National Diabetes Statistics Report: Estimates of Diabetes and Its Burden in the United States, 2014. Atlanta, GA: US Department of Health and Human Services.

CENTERS FOR DISEASE CONTROL AND PREVENTION 2015. Age-Adjusted Rates of Diagnosed Diabetes per 100 Civilian, Non-Institutionalized Population, by Education, United States, 1980–2014. Available at http://www.cdc.gov/diabetes/statistics/prev/national/figbyeducation.htm. Last accessed November 20, 2016.

CENTERS FOR DISEASE CONTROL AND PREVENTION 2016. Long-term Trends in Diabetes. Aprill 2016. Available from https://www.cdc.gov/diabetes/statistics/slides/long_term_trends.pdf. Last accessed November 20, 2016.

CHALOUPKA, F. J. & POWELL, L. M. 2009. Price, Availability, and Youth Obesity: Evidence From Bridging the Gap. Preventing Chronic Disease: Public Health Research, Practice, and Policy, 6, 1-6.

CHANDON, P. & WANSINK, B. 2012. Does food marketing need to make us fat? A review and solutions. Nutrition Reviews, 70, 571-593.

CHIUVE, S. E., FUNG, T. T., RIMM, E. B., HU, F. B., MCCULLOUGH, M. L., WANG, M., STAMPFER, M. J. & WILLETT, W. C. 2012. Alternative dietary indices both strongly predict risk of chronic disease. J Nutr, 142, 1009-18.

CHOI, H. K., WILLETT, W. C., STAMPFER, M. J., RIMM, E. & HU, F. B. 2005. Dairy Consumption and Risk of Type 2 Diabetes Mellitus in Men. Archives of Internal Medicine, 165, 997-1003.

CHOU, S.-Y., GROSSMAN, M. & SAFFER, H. 2004. An economic analysis of adult obesity: results from the Behavioral Risk Factor Surveillance System. Journal of Health Economics, 23, 565-87.

CHRISTINE, P. J., AUCHINCLOSS, A. H., BERTONI, A. G. & ET AL. 2015. Longitudinal associations between neighborhood physical and social environments and incident type 2 diabetes mellitus: The multi-ethnic study of atherosclerosis (mesa). JAMA Internal Medicine, 175, 1311-1320.

CHUNG, C. & MYERS, S. L. 1999. Do the Poor Pay More for Food? An Analysis of Grocery Store Availability and Food Price Disparities. Journal of Consumer Affairs, 33, 276-296.

COCA-COLA. 2014. Product Nutrition Information [Online]. Available: http://productnutrition.thecoca-colacompany.com/ [Accessed December 1, 2014.

COLEMAN-JENSEN, A., RABBITT, M. P., GREGORY, C. & SINGH, A. 2015. Household Food Security in the United States in 2014. ERR-194, U.S. Department of Agriculture, Economic Research Service, September 2015.

107

CONNORS, M., BISOGNI, C. A., SOBAL, J. & DEVINE, C. M. 2001. Managing values in personal food systems. Appetite, 36, 189-200.

COUNCIL FOR COMMUNITY AND ECONOMIC RESEARCH 2016. Cost of Living Index. Available at http://coli.org/. Last accessd November 25, 2016.

CUMMINS, S., FLINT, E. & MATTHEWS, S. A. 2014. New Neighborhood Grocery Store Increased Awareness Of Food Access But Did Not Alter Dietary Habits Or Obesity. Health Affairs, 33, 283-291.

CUMMINS, S. & MACINTYRE, S. 2002. “Food deserts”—evidence and assumption in health policy making. BMJ, 325, 436-8.

DANNON 2015. Dannon All Natural Nutrition Facts. http://www.dannon.com/dannon-all-natural/.

DARMON, N. & DREWNOWSKI, A. 2015. Contribution of food prices and diet cost to socioeconomic disparities in diet quality and health: a systematic review and analysis. Nutr Rev, 73, 643-60.

DARMON, N., LACROIX, A., MULLER, L. & RUFFIEUX, B. 2016. Food Price Policies May Improve Diet but Increase Socioeconomic Inequalities in Nutrition. World Rev Nutr Diet, 115, 36-45.

DAVIS, C. G., BLAYNEY, D. P., DONG, D., STEFANOVA, S. & JOHNSON, A. 2010. Long-Term Growth in U.S. Cheese Consumption May Slow. United States Department of Agriculture, Economic Research Service, LDP-M-193-01.

DE IRALA-ESTEVEZ, J., GROTH, M., JOHANSSON, L., OLTERSDORF, U., PRATTALA, R. & MARTINEZ-GONZALEZ, M. A. 2000. A systematic review of socio-economic differences in food habits in Europe: consumption of fruit and vegetables. European Journal of Clinical Nutrition, 54, 706-714.

DE KONING, L., CHIUVE, S. E., FUNG, T. T., WILLETT, W. C., RIMM, E. B. & HU, F. B. 2011a. Diet-quality scores and the risk of type 2 diabetes in men. Diabetes Care, 34, 1150-1156.

DE KONING, L., MALIK, V. S., RIMM, E. B., WILLETT, W. C. & HU, F. B. 2011b. Sugar-sweetened and artificially sweetened beverage consumption and risk of type 2 diabetes in men. American Journal of Clinical Nutrition, 93, 1321-1327.

DHINGRA, R., SULLIVAN, L., JACQUES, P. F., WANG, T. J., FOX, C. S., MEIGS, J. B., D'AGOSTINO, R. B., GAZIANO, J. M. & VASAN, R. S. 2007. Soft Drink Consumption and Risk of Developing Cardiometabolic Risk Factors and the Metabolic Syndrome in Middle-Aged Adults in the Community. Circulation, 116, 480-488.

DIABETES PREVENTION PROGRAM RESEARCH GROUP 2002. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin New England Journal of Medicine, 346, 393-403.

DIEZ ROUX, A. V., JACOBS, D. R. & KIEFE, C. I. 2002. Neighborhood Characteristics and Components of the Insulin Resistance Syndrome in Young Adults. The Coronary Artery Risk Development in Young Adults (CARDIA) Study, 25, 1976-1982.

DIEZ ROUX, A. V. & MAIR, C. 2010. Neighborhoods and health. Annals of the New York Academy of Sciences, 1186, 125-145.

DIEZ ROUX, A. V., STEIN MERKIN, S., ARNETT, D., CHAMBLESS, L., MASSING, M., NIETO, J., SORLIE, P., SZKLO, M., TYROLER, H. A. & WATSON, R. L.

108

2001. Neighborhood of residence and incidence of coronary heart disease. New England Journal of Medicine, 345, 99-106.

DONAHOE, S. M., STEWART, G. C., MCCABE, C. H., MOHANAVELU, S., MURPHY, S. A., CANNON, C. P. & ANTMAN, E. M. 2007. Diabetes and mortality following acute coronary syndromes. JAMA, 298, 765-75.

DOVE, E. R., HODGSON, J. M., PUDDEY, I. B., BEILIN, L. J., LEE, Y. P. & MORI, T. A. 2009. Skim milk compared with a fruit drink acutely reduces appetite and energy intake in overweight men and women. American Journal of Clinical Nutrition, 90, 70-75.

DREWNOWSKI, A. 2009. Obesity, diets, and social inequalities. Nutrition Reviews, 67, S36-S39.

DREWNOWSKI, A. 2010. The cost of US foods as related to their nutritive value. American Journal of Clinical Nutrition, 92, 1181-1188.

DREWNOWSKI, A., AGGARWAL, A., COOK, A., STEWART, O. & MOUDON, A. V. 2016. Geographic disparities in Healthy Eating Index scores (HEI–2005 and 2010) by residential property values: Findings from Seattle Obesity Study (SOS). Preventive Medicine, 83, 46-55.

DREWNOWSKI, A., AGGARWAL, A., HURVITZ, P. M., MONSIVAIS, P. & MOUDON, A. V. 2012. Obesity and Supermarket Access: Proximity or Price? American Journal of Public Health, 102, e74-e80.

DREWNOWSKI, A. & DARMON, N. 2005. Food Choices and Diet Costs: an Economic Analysis. The Journal of Nutrition, 135, 900-904.

DREWNOWSKI, A., MOUDON, A. V., JIAO, J., AGGARWAL, A., CHARREIRE, H. & CHAIX, B. 2014. Food environment and socioeconomic status influence obesity rates in Seattle and in Paris. International Journal of Obesity, 38, 306-314.

DREWNOWSKI, A. & SPECTER, S. E. 2004. Poverty and obesity: the role of energy density and energy costs. American Journal of Clinical Nutrition, 79, 6-16.

DUBOWITZ, T., HERON, M., BIRD, C. E., LURIE, N., FINCH, B. K., BASURTO-DAVILA, R., HALE, L. & ESCARCE, J. J. 2008. Neighborhood socioeconomic status and fruit and vegetable intake among whites, blacks, and Mexican Americans in the United States. American Journal of Clinical Nutrition, 87, 1883-1891.

DUFFEY, K. J., GORDON-LARSEN, P., SHIKANY, J. M., GUILKEY, D., JACOBS, D. R., JR. & POPKIN, B. M. 2010. Food price and diet and health outcomes: 20 years of the CARDIA Study. Arch Intern Med, 170, 420-6.

DUHANEY, T., CAMPBELL, N., NIEBYLSKI, M. L., KACZOROWSKI, J., TSUYUKI, R. T., WILLIS, K., MANG, E., ARANGO, M., MORRIS, D. & ASHLEY, L. 2015. Death by diet: the role of food pricing interventions as a public policy response and health advocacy opportunity. Can J Cardiol, 31, 112-6.

EL-ABBADI, N. H., DAO, M. C. & MEYDANI, S. N. 2014. Yogurt: role in healthy and active aging. Am J Clin Nutr, 99, 1263S-70S.

ELBEL, B., MORAN, A., DIXON, L. B., KISZKO, K., CANTOR, J., ABRAMS, C. & MIJANOVICH, T. 2015. Assessment of a government-subsidized supermarket in

109

a high-need area on household food availability and children's dietary intakes. Public Health Nutrition, 18, 2881-90.

EMOTO, M., NISHIZAWA, Y., MAEKAWA, K., HIURA, Y., KANDA, H., KAWAGISHI, T., SHOJI, T., OKUNO, Y. & MORII, H. 1999. Homeostasis model assessment as a clinical index of insulin resistance in type 2 diabetic patients treated with sulfonylureas. Diabetes Care, 22, 818-22.

EPSTEIN, L. H., DEARING, K. K., ROBA, L. G. & FINKELSTEIN, E. 2010. The influence of taxes and subsidies on energy purchased in an experimental purchasing study. Psychol Sci, 21, 406-14.

EPSTEIN, L. H., JANKOWIAK, N., NEDERKOORN, C., RAYNOR, H. A., FRENCH, S. A. & FINKELSTEIN, E. 2012. Experimental research on the relation between food price changes and food-purchasing patterns: a targeted review. The American Journal of Clinical Nutrition, 95, 789-809.

EXPERT COMMITTEE ON THE DIAGNOSIS AND CLASSIFICATION OF DIABETES MELLITUS 2003. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care, 26, s5-s20.

FAVELL, D. J. 1998. A comparison of the vitamin C content of fresh and frozen vegetables. Food Chemistry, 62, 59-64.

FINKELSTEIN, E. A., STROMBOTNE, K. L., ZHEN, C. & EPSTEIN, L. H. 2014. Food prices and obesity: a review. Adv Nutr, 5, 818-21.

FLETCHER, J. M., FRISVOLD, D. & TEFFT, N. 2010a. Can Soft Drink Taxes Reduce Population Weight? Contemp Econ Policy, 28, 23-35.

FLETCHER, J. M., FRISVOLD, D. E. & TEFFT, N. 2010b. The effects of soft drink taxes on child and adolescent consumption and weight outcomes. Journal of Public Economics, 94, 967-974.

FORD, P. B. & DZEWALTOWSKI, D. A. 2010. Limited supermarket availability is not associated with obesity risk among participants in the Kansas WIC Program. Obesity (Silver Spring), 18, 1944-51.

FORSHEE, R. A. & STOREY, M. L. 2006. Demographics, not beverage consumption, is associated with diet quality. International Journal of Food Sciences & Nutrition, 57, 494-511.

FRAZAO, E., ANDREWS, M., SMALLWOOD, D. & PRELL, M. 2007. Food Spending Patterns of Low-Income Households Will Increasing Purchasing Power Result in Healthier Food Choices? Can Food Stamps Do More to Improve Food Choices? An Economic Perspective, Economic Information Bulletin Number 29-4. Washington, D.C.: U.S. Department of Agriculture, Economic Research Service. .

FRITO LAY 2014. DORITOS® Nacho Cheese Flavored Tortilla Chips Nutrition Information.

FULLER, D., CUMMINS, S. & MATTHEWS, S. A. 2013. Does transportation mode modify associations between distance to food store, fruit and vegetable consumption, and BMI in low-income neighborhoods? Am J Clin Nutr, 97, 167-72.

FUMERON, F., LAMRI, A., ABI KHALIL, C., JAZIRI, R., PORCHAY-BALDERELLI, I., LANTIERI, O., VOL, S., BALKAU, B. & MARRE, M. 2011. Dairy consumption and the incidence of hyperglycemia and the metabolic syndrome:

110

results from a french prospective study, Data from the Epidemiological Study on the Insulin Resistance Syndrome (DESIR). Diabetes Care, 34, 813-7.

FUNG, T. T., RIMM, E. B., SPIEGELMAN, D., RIFAI, N., TOFLER, G. H., WILLETT, W. C. & HU, F. B. 2001. Association between dietary patterns and plasma biomarkers of obesity and cardiovascular disease risk. Am J Clin Nutr, 73, 61-7.

GAO, S. K., BERESFORD, S. A., FRANK, L. L., SCHREINER, P. J., BURKE, G. L. & FITZPATRICK, A. L. 2008. Modifications to the Healthy Eating Index and its ability to predict obesity: the Multi-Ethnic Study of Atherosclerosis. Am J Clin Nutr, 88, 64-9.

GEISS, L. S., WANG, J., CHENG, Y. J. & ET AL. 2014. PRevalence and incidence trends for diagnosed diabetes among adults aged 20 to 79 years, united states, 1980-2012. JAMA, 312, 1218-1226.

GITTELSOHN, J., WOLEVER, T. M., HARRIS, S. B., HARRIS-GIRALDO, R., HANLEY, A. J. & ZINMAN, B. 1998. Specific patterns of food consumption and preparation are associated with diabetes and obesity in a Native Canadian community. J Nutr, 128, 541-7.

GLANZ, K., BASIL, M., MAIBACH, E., GOLDBERG, J. & SNYDER, D. 1998. Why Americans eat what they do: taste, nutrition, cost, convenience, and weight control concerns as influences on food consumption. J Am Diet Assoc, 98, 1118-26.

GO, A. S. & ET AL 2014. Heart Disease and Stroke Statistics - 2014 Update: A Report from the American Heart Association. Circulation, 129, 28-295

GREENLAND, S. 2000. An introduction to instrumental variables for epidemiologists. International Journal of Epidemiology, 29, 722-729.

GRIMM, K. A., MOORE, L. V. & SCANLON, K. S. 2013. Access to healthier food retailers - United States, 2011. MMWR Surveill Summ, 62 Suppl 3, 20-6.

GROCERY HEADQUARTERS Leading vendors of potato chips in the United States in 2016, based on sales (in million U.S. dollars). Available at: http://www.statista.com/statistics/188220/leading-potato-chips-vendors-in-the-united-states-in-2011/. Last accessed: May 18, 2016.

GROCERY HEADQUARTERS Sales of the leading chocolate candy brands of the United States in 2016 (in million U.S. dollars)*. Available at: http://www.statista.com/statistics/190393/top-chocolate-candy-brands-in-the-united-states/ Last accessed: May 18, 2016.

GROCERY HEADQUARTERS Supermarket categories with the highest private label share in the United States in 2012, based on unit market share. Available at: http://www.statista.com/statistics/259966/private-label-share-in-us-supermarket-categories/ Last accessed: May 18, 2016.

GRUNDY, S. M. 2002. Obesity, metabolic syndrome, and coronary atherosclerosis. Circulation, 105, 2696-8.

GUENTHER, P. M., CASAVALE, K. O., REEDY, J., KIRKPATRICK, S. I., HIZA, H. A., KUCZYNSKI, K. J., KAHLE, L. L. & KREBS-SMITH, S. M. 2013. Update of the Healthy Eating Index: HEI-2010. J Acad Nutr Diet, 113, 569-80.

GUENTHER, P. M., KIRKPATRICK, S. I., REEDY, J., KREBS-SMITH, S. M., BUCKMAN, D. W., DODD, K. W., CASAVALE, K. O. & CARROLL, R. J. 2014. The Healthy Eating Index-2010 is a valid and reliable measure of diet

111

quality according to the 2010 Dietary Guidelines for Americans. J Nutr, 144, 399-407.

GUO, X., WARDEN, B. A., PAERATAKUL, S. & BRAY, G. A. 2004. Healthy Eating Index and obesity. Eur J Clin Nutr, 58, 1580-6.

GUSTAFSON, A., HANKINS, S. & JILCOTT, S. 2012. Measures of the consumer food store environment: a systematic review of the evidence 2000-2011. J Community Health, 37, 897-911.

HAN, E. & POWELL, L. M. 2011. Effect of food prices on the prevalence of obesity among young adults. Public Health, 125, 129-35.

HARMON, B. E., BOUSHEY, C. J., SHVETSOV, Y. B., ETTIENNE, R., REEDY, J., WILKENS, L. R., LE MARCHAND, L., HENDERSON, B. E. & KOLONEL, L. N. 2015. Associations of key diet-quality indexes with mortality in the Multiethnic Cohort: the Dietary Patterns Methods Project. Am J Clin Nutr, 101, 587-97.

HARNACK, L., OAKES, J. M., ELBEL, B., BEATTY, T., RYDELL, S. & FRENCH, S. 2016. Effects of Subsidies and Prohibitions on Nutrition in a Food Benefit Program: A Randomized Clinical Trial. JAMA Intern Med.

HE, F. J., NOWSON, C. A., LUCAS, M. & MACGREGOR, G. A. 2007. Increased consumption of fruit and vegetables is related to a reduced risk of coronary heart disease: meta-analysis of cohort studies. J Hum Hypertens, 21, 717-28.

HERNAN, M. A. & ROBINS, J. M. 2006. Instruments for causal inference: an epidemiologist's dream? Epidemiology, 17, 360-72.

HERSHEY'S. 2014. Hershey's Milk Chocolate Bar Nutrition Information [Online]. Available: http://www.hersheys.com/pure-products/details.aspx?id=3480 [Accessed November 30, 2014.

HU, F. B. 2013. Resolved: there is sufficient scientific evidence that decreasing sugar-sweetened beverage consumption will reduce the prevalence of obesity and obesity-related diseases. Obesity Reviews, 14, 606-619.

HU, F. B. & MALIK, V. S. 2010. Sugar-sweetened beverages and risk of obesity and type 2 diabetes: Epidemiologic evidence. Physiology & Behavior, 100, 47-54.

HU, F. B., MANSON, J. E., STAMPFER, M. J., COLDITZ, G. A., LIU, S., SOLOMON, C. G. & WILLETT, W. C. 2001. Diet, Lifestyle, and the Risk of Type 2 Diabetes Mellitus in Women. New England Journal of Medicine, 345, 790-797.

HUNG, H. C., JOSHIPURA, K. J., JIANG, R., HU, F. B., HUNTER, D., SMITH-WARNER, S. A., COLDITZ, G. A., ROSNER, B., SPIEGELMAN, D. & WILLETT, W. C. 2004. Fruit and vegetable intake and risk of major chronic disease. J Natl Cancer Inst, 96, 1577-84.

INTERNATIONAL COCOA ORGANIZATION The World Cocoa Economy: Past and Present. In: COMMITTEE, E. (ed.). Executive Committee EX/142/6.

INTERNATIONAL FOOD INFORMATION COUNCIL FOUNDATION 2015. Food and Health Survey. Available at http://www.foodinsight.org/sites/default/files/2015%20Food%20and%20Health%20Survey%20-%20FINAL.pdf. Last accessed November 12, 2016.

IRI. 2014. 2013 Market and Region Profiles [Online]. Available: http://www.infores.com/public/mrp/default.htm [Accessed June 17, 2015.

112

IRI. 2015. Academic Data Set [Online]. Available: http://www.iriworldwide.com/solutions/Academic-Data-Set [Accessed June 22, 2015.

JACOBSON, M. F. 2005. Liquid candy: How soft drinks are harming Americans' health. In: INTEREST, C. F. S. I. T. P. (ed.). Washington, D.C.

JANISZEWSKI, P. M., JANNSSEN, I. & ROSS, R. 2007. Does Waist Circumference Predict Diabetes and Cardiovascular Disease Beyond Commonly Evaluated Cardiometabolic Risk Factors? Diabetes Care, 30, 3105-3109.

JETTER, K. M. & CASSADY, D. L. 2006. The availability and cost of healthier food alternatives. Am J Prev Med, 30, 38-44.

JUDE, E. B., OYIBO, S. O., CHALMERS, N. & BOULTON, A. J. M. 2001. Peripheral Arterial Disease in Diabetic and Nondiabetic Patients. Diabetes Care, 24, 1433-1437.

KAHN, B. B. & FLIER, J. S. 2000. Obesity and insulin resistance. The Journal of Clinical Investigation, 106, 473-481.

KAPPELER, R., EICHHOLZER, M. & ROHRMANN, S. 2013. Meat consumption and diet quality and mortality in NHANES III. Eur J Clin Nutr, 67, 598-606.

KAUFMAN, P. R., MACDONALD, J. M., LUTZ, S. M. & SMALLWOOD, D. M. 1997. Do the Poor Pay More for Food? Item Selection and Price Differences Affect Low-Income Household Food Costs. Economic Research Service, U.S. Department of Agriculture, Agricultural Economic Report No. 759.

KEARNEY, J. 2010. Food consumption trends and drivers. Philosophical Transactions of the Royal Society B, 365, 2793-2807.

KENNEDY, E. T., OHLS, J., CARLSON, S. & FLEMING, K. 1995. The Healthy Eating Index: design and applications. Journal of the American Dietic Association, 95, 1103-1108.

KERN, D. M., AUCHINCLOSS, A. H., BALLISTER, L. & ROBINSON, L. F. 2016. Neighbourhood variation in the price of soda relative to milk and its association with neighbourhood socio-economic status and race. Public Health Nutrition, 19, 3386-3396.

KERN, D. M., AUCHINCLOSS, A. H., ROBINSON, L. F., STEHR, M. & PHAM-KANTER, G. The variation of healthy and unhealthy food prices across neighborhoods and their association with neighborhood socioeconomic status and proportion Black/Hispanic. Under Review.

KIT, B. K., FAKHOURI, T. H., PARK, S., NIELSEN, S. J. & OGDEN, C. L. 2013. Trends in sugar-sweetened beverage consumption among youth and adults in the United States: 1999-2010. Am J Clin Nutr, 98, 180-8.

KOCHANEK, K. D., MURPHY, S. L., XU, J. & TEJADA-VERA, B. 2016. Deaths: Final Data for 2014. National vital statistics reports. Hyattsville, MD: National Center for Health Statistics.

KONTTINEN, H., SARLIO-LAHTEENKORVA, S., SILVENTOINEN, K., MANNISTO, S. & HAUKKALA, A. 2013. Socio-economic disparities in the consumption of vegetables, fruit and energy-dense foods: the role of motive priorities. Public Health Nutr, 16, 873-82.

KOPELMAN, P. G. 2000. Obesity as a medical problem. Nature, 404, 635-43.

113

KREBS, A. L. 2013. Beyond the Gallon Jug: How Can Innovation Change the Path of Fluid Milk? CoBANK Knowledge Exchange, June.

KUCHLER, F., TEGENE, A. & HARRIS, J. M. 2004. Taxing Snack Foods: What to Expect for Diet and Tax Revenues. In: UNITED STATES DEPARTMENT OF AGRICULTURE, E. R. S. (ed.).

KUROWSKA, E. M., SPENCE, J. D., JORDAN, J., WETMORE, S., FREEMAN, D. J., PICHE, L. A. & SERRATORE, P. 2000. HDL-cholesterol-raising effect of orange juice in subjects with hypercholesterolemia. Am J Clin Nutr, 72, 1095-100.

LAMICHHANE, A. P., WARREN, J., PUETT, R., PORTER, D. E., BOTTAI, M., MAYER-DAVIS, E. J. & LIESE, A. D. 2013. Spatial patterning of supermarkets and fast food outlets with respect to neighborhood characteristics. Health Place, 23, 157-64.

LATHAM, J. & MOFFAT, T. 2007. Determinants of variation in food cost and availability in two socioeconomically contrasting neighbourhoods of Hamilton, Ontario, Canada. Health Place, 13, 273-87.

LEVITAN, E. B., AHMED, A., ARNETT, D. K., POLAK, J. F., HUNDLEY, W. G., BLUEMKE, D. A., HECKBERT, S. R., JACOBS, D. R., JR. & NETTLETON, J. A. 2016. Mediterranean diet score and left ventricular structure and function: the Multi-Ethnic Study of Atherosclerosis. Am J Clin Nutr, 104, 595-602.

LI, W. Q., PARK, Y., MCGLYNN, K. A., HOLLENBECK, A. R., TAYLOR, P. R., GOLDSTEIN, A. M. & FREEDMAN, N. D. 2014. Index-based dietary patterns and risk of incident hepatocellular carcinoma and mortality from chronic liver disease in a prospective study. Hepatology, 60, 588-97.

LIPSKY, L. M. 2009. Are energy-dense foods really cheaper? Reexamining the relation between food price and energy density. Am J Clin Nutr, 90, 1397-401.

LOPEZ, C. N., MARTINEZ-GONZALEZ, M. A., SANCHEZ-VILLEGAS, A., ALONSO, A., PIMENTA, A. M. & BES-RASTROLLO, M. 2009. Costs of Mediterranean and western dietary patterns in a Spanish cohort and their relationship with prospective weight change. J Epidemiol Community Health, 63, 920-7.

LUDWIG, D. S., PETERSON, K. E. & GORTMAKER, S. L. 2001. Relation between consumption of sugar-sweetened drinks and childhood obesity: a prospective, observational analysis. The Lancet, 357, 505-508.

MAGEE, E. 2014. Junk-Food Facts: Are you a junk-food junkie? Here's what you need to know. [Online]. Available: http://www.webmd.com/diet/features/junk-food-facts [Accessed November 30, 2014.

MAJOR, G. C., CHAPUT, J. P., LEDOUX, M., ST-PIERRE, S., ANDERSON, G. H., ZEMEL, M. B. & TREMBLAY, A. 2008. Recent developments in calcium-related obesity research. Obes Rev, 9, 428-45.

MALIK, V. S. & HU, F. B. 2012. Sweeteners and Risk of Obesity and Type 2 Diabetes: The Role of Sugar-Sweetened Beverages. Current Diabetes Reports, 12, 195-203.

MALIK, V. S., POPKIN, B. M., BRAY, G. A., DESPRES, J.-P., WILLETT, W. C. & HU, F. B. 2010a. Sugar-Sweetened Beverages and Risk of Metabolic Syndrome and Type 2 Diab. Diabetes Care, 33, 2477-2483.

114

MALIK, V. S., POPKIN, B. M., BRAY, G. A., DESPRES, J. P. & HU, F. B. 2010b. Sugar Sweetened Beverages, Obesity, Type 2 Diabetes and Cardiovascular Disease risk. Circulation, 121, 1356-1373.

MALIK, V. S., SCHULZE, M. B. & HU, F. B. 2006. Intake of sugar-sweetened beverages and weight gain: a systematic review. The American Journal of Clinical Nutrition, 84, 274-288.

MARGOLIS, K. L., WEI, F., DE BOER, I. H., HOWARD, B. V., LIU, S., MANSON, J. E., MOSSAVAR-RAHMANI, Y., PHILLIPS, L. S., SHIKANY, J. M. & TINKER, L. F. 2011. A diet high in low-fat dairy products lowers diabetes risk in postmenopausal women. J Nutr, 141, 1969-74.

MARTENS, E. P., PESTMAN, W. R., DE BOER, A., BELITSER, S. V. & KLUNGEL, O. H. 2006. Instrumental variables: application and limitations. Epidemiology, 17, 260-7.

MATTHEWS, D. R., HOSKER, J. P., RUDENSKI, A. S., NAYLOR, B. A., TREACHER, D. F. & TURNER, R. C. 1985. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia, 28, 412-9.

MAYER-DAVIS, E. J., VITOLINS, M. Z., CARMICHAEL, S. L., HEMPHILL, S., TSAROUCHA, G., RUSHING, J. & LEVIN, S. 1999. Validity and reproducibility of a food frequency interview in a Multi-Cultural Epidemiology Study. Ann Epidemiol, 9, 314-24.

MAYNE, S. L., AUCHINCLOSS, A. H. & MICHAEL, Y. L. 2015. Impact of policy and built environment changes on obesity-related outcomes: a systematic review of naturally occurring experiments. Obes Rev, 16, 362-75.

MCCULLOUGH, M. L., FESKANICH, D., RIMM, E. B., GIOVANNUCCI, E. L., ASCHERIO, A., VARIYAM, J. N., SPIEGELMAN, D., STAMPFER, M. J. & WILLETT, W. C. 2000. Adherence to the Dietary Guidelines for Americans and risk of major chronic disease in men. Am J Clin Nutr, 72, 1223-31.

MCGILL, R., ANWAR, E., ORTON, L., BROMLEY, H., LLOYD-WILLIAMS, F., O'FLAHERTY, M., TAYLOR-ROBINSON, D., GUZMAN-CASTILLO, M., GILLESPIE, D., MOREIRA, P., ALLEN, K., HYSENI, L., CALDER, N., PETTICREW, M., WHITE, M., WHITEHEAD, M. & CAPEWELL, S. 2015. Are interventions to promote healthy eating equally effective for all? Systematic review of socioeconomic inequalities in impact. BMC Public Health, 15, 457.

MEYER, K. A., GUILKEY, D. K., NG, S. W., DUFFEY, K. J., POPKIN, B. M., KIEFE, C. I., STEFFEN, L. M., SHIKANY, J. M. & GORDON-LARSEN, P. 2014. Sociodemographic differences in fast food price sensitivity. JAMA Intern Med, 174, 434-42.

MICHIMI, A. & WIMBERLY, M. C. 2010. Associations of supermarket accessibility with obesity and fruit and vegetable consumption in the conterminous United States. Int J Health Geogr, 9, 49.

MOGENSEN, C. E., CHRISTNSEN, C. K. & VITTINGHUS, E. 1983. The Stages in Diabetic Renal Disease: With Emphasis on the Stage of Incipient Diabetic Nephropathy. Diabetes, 32, 64-78.

115

MONSIVAIS, P., AGGARWAL, A. & DREWNOWSKI, A. 2012. Are socio-economic disparities in diet quality explained by diet cost? J Epidemiol Community Health, 66, 530-5.

MONSIVAIS, P., MCLAIN, J. & DREWNOWSKI, A. 2010. The rising disparity in the price of healthful foods: 2004-2008. Food Policy, 35, 514-520.

MORLAND, K., DIEZ ROUX, A. V. & WING, S. 2006. Supermarkets, other food stores, and obesity: the atherosclerosis risk in communities study. Am J Prev Med, 30, 333-9.

MORRIS, A. 2011. Impact of Orange Juice Market Dynamics on Fruit Prices. Presented at the International Citrus Economics Conference. October 27, 2011. Orlando, Florida.

MORRIS, M. A., HULME, C., CLARKE, G. P., EDWARDS, K. L. & CADE, J. E. 2014. What is the cost of a healthy diet? Using diet data from the UK Women's Cohort Study. Journal of Epidemiology and Community Health.

MURAKI, I., IMAMURA, F., MANSON, J. E., HU, F. B., WILLETT, W. C., VAN DAM, R. M. & SUN, Q. 2013. Fruit consumption and risk of type 2 diabetes: results from three prospective longitudinal cohort studies. Bmj, 347, f5001.

MURPHY, S. L., XU, J. & KOCHANEK, K. D. 2013. Deaths: Final Data for 2010. National vital statistics reports. Hyattsville, MD: National Center for Health Statistics.

MURSU, J., VIRTANEN, J. K., TUOMAINEN, T.-P., NURMI, T. & VOUTILAINEN, S. 2014. Intake of fruit, berries, and vegetables and risk of type 2 diabetes in Finnish men: the Kuopio Ischaemic Heart Disease Risk Factor Study. American Journal of Clinical Nutrition, 99, 328-333.

NABISCO 2014. Nabisco Chips Ahoy! Original Chocolate Chip Nutrition Information. In: WORKS, S. (ed.). http://www.snackworks.com/products/product-detail.aspx?product=4400003219.

NELSON, R. G., EVERHART, J. E., KNOWLER, W. C. & BENNETT, P. H. 1988. Incidence, prevalence and risk factors for non-insulin-dependent diabetes mellitus. Prim Care, 15, 227-50.

NETTLETON, J. A., LUTSEY, P. L., WANG, Y., LIMA, J. A., MICHOS, E. D. & JACOBS, D. R. 2009a. Diet Soda Intake and Risk of Incident Metabolic Syndrome and Type 2 Diabetes in the Multi-Ethnic Study of Atherosclerosis (MESA). Diabetes Care, 32, 688-694.

NETTLETON, J. A., ROCK, C. L., WANG, Y., JENNY, N. S. & JACOBS, D. R. 2009b. Associations between dietary macronutrient intake and plasma lipids demonstrate criterion performance of the Multi-Ethnic Study of Atherosclerosis (MESA) food-frequency questionnaire. Br J Nutr, 102, 1220-7.

NETTLETON, J. A., SCHULZE, M. B., JIANG, R., JENNY, N. S., BURKE, G. L. & JACOBS, D. R. 2008a. A priori– defined dietary patterns and markers of cardiovascular disease risk in the Multi-Ethnic Study of Atherosclerosis (MESA). American Journal of Clinical Nutrition, 88, 185-94.

NETTLETON, J. A., STEFFEN, L. M., MAYER-DAVIS, E. J., JENNY, N. S., JIANG, R., HERRINGTON, D. M. & JACOBS, D. R. 2006. Dietary patterns are associated with biochemical markers of inflammation and endothelial activation

116

in the Multi-Ethnic Study of Atherosclerosis (MESA). American Journal of Clinical Nutrition, 83, 1369-1379.

NETTLETON, J. A., STEFFEN, L. M., NI, H., LIU, K. & JACOBS, D. R. 2008b. Dietary patterns and risk of incident type 2 diabetes in the Multi-Ethnic Study of Atherosclerosis (MESA). Diabetes Care, 31, 1777-1782.

NI MHURCHU, C., BLAKELY, T., JIANG, Y., EYLES, H. C. & RODGERS, A. 2010. Effects of price discounts and tailored nutrition education on supermarket purchases: a randomized controlled trial. Am J Clin Nutr, 91, 736-47.

NIEBYLSKI, M. L., REDBURN, K. A., DUHANEY, T. & CAMPBELL, N. R. 2015. Healthy food subsidies and unhealthy food taxation: A systematic review of the evidence. Nutrition, 31, 787-95.

O’HARA, J. & MUSICU, A. 2015. Big Soda vs. Public Health. Center for Science in the Public Interest: How the Industry Opens Its Checkbook to Defeat Health Measures available at: https://cspinet.org/file/6465/download?token=qLEj8qfC.

OGDEN, C. L., CARROLL, M. D., KIT, B. K. & FLEGAL, K. M. 2014. Prevalence of Childhood and Adult Obesity in the United States, 2011-2012. JAMA, 311, 806-814.

OGDEN, C. L., KIT, B. K., CARROLL, M. D. & PARK, S. 2011. Consumption of Sugar Drinks in the United States, 2005–2008. NCHS data brief, no 71. Hyattsville, MD: National Center for Health Statistics.

OKRENT, A. & ALSTON, J. M. August 2012. The Demand for Disaggregated Food-Away-From-Home and Food-at-Home Products in the United States. ERR-139, U.S. Department of Agriculture, Economic Research Service.

OSLER, M., HELMS ANDREASEN, A., HEITMANN, B., HOIDRUP, S., GERDES, U., MORCH JORGENSEN, L. & SCHROLL, M. 2002. Food intake patterns and risk of coronary heart disease: a prospective cohort study examining the use of traditional scoring techniques. European Journal of Clinical Nutrition, 56, 568-574.

PEREIRA, M. A., JACOBS, D. R., VAN HORN, L., SLATTERY, M. L., KARTASHOV, A. I. & LUDWIG, D. S. 2002. Dairy consumption, obesity, and the insulin resistance syndrome in young adults: the CARDIA Study. JAMA, 287, 2081-2089.

PFEUFFER, M. & SCHREZENMEIR, J. 2007. Milk and the metabolic syndrome. Obes Rev, 8, 109-18.

PIERNAS, C. & POPKIN, B. M. 2009. Snacking Increased among U.S. Adults between 1977 and 2006. Journal of Nutrition, 140, 325-332.

POWELL, L. M., AULD, M. C., CHALOUPKA, F. J., O'MALLEY, P. M. & JOHNSTON, L. D. 2007a. Access to fast food and food prices: relationship with fruit and vegetable consumption and overweight among adolescents. Adv Health Econ Health Serv Res, 17, 23-48.

POWELL, L. M., AULD, M. C., CHALOUPKA, F. J., O'MALLEY, P. M. & JOHNSTON, L. D. 2007b. Associations between access to food stores and adolescent body mass index. Am J Prev Med, 33, S301-7.

POWELL, L. M. & BAO, Y. 2009. Food prices, access to food outlets and child weight. Econ Hum Biol, 7, 64-72.

117

POWELL, L. M. & CHALOUPKA, F. J. 2009. Food Prices and Obesity: Evidence and Policy Implications for Taxes and Subsidies. The Milbank Quarterly, 87, 229-257.

POWELL, L. M., CHRIQUI, J. & CHALOUPKA, F. J. 2009a. Associations between state-level soda taxes and adolescent body mass index. Journal of Adolescent Health, 45, S57-S63.

POWELL, L. M., CHRIQUI, J. F., KHAN, T., WADA, R. & CHALOUPKA, F. J. 2013. Assessing the potential effectiveness of food and beverage taxes and subsidies for improving public health: a systematic review of prices, demand and body weight outcomes. Obes Rev, 14, 110-28.

POWELL, L. M., WADA, R. & KUMANYIKA, S. K. 2014. Racial/ethnic and income disparities in child and adolescent exposure to food and beverage television ads across the U.S. media markets. Health Place, 29, 124-31.

POWELL, L. M., ZHAO, Z. & WANG, Y. 2009b. Food prices and fruit and vegetable consumption among young American adults. Health & Place, 15, 1064-1070.

RAO, M., AFSHIN, A., SINGH, G. & MOZAFFARIAN, D. 2013. Do healthier foods and diet patterns cost more than less healthy options? A systematic review and meta-analysis. BMJ Open, 3, e004277.

RASHAD, I. 2007. Obesity and diabetes: the roles that prices and policies play. Adv Health Econ Health Serv Res, 17, 113-28.

RASSEN, J. A., BROOKHART, M. A., GLYNN, R. J., MITTLEMAN, M. A. & SCHNEEWEISS, S. 2009. Instrumental variables I: instrumental variables exploit natural variation in nonexperimental data to estimate causal relationships. Journal of Clinical Epidemiology, 62, 1226-1232.

RATHOD, A. D., BHARADWAJ, A. S., BADHEKA, A. O., KIZILBASH, M. & AFONSO, L. 2012. Healthy Eating Index and mortality in a nationally representative elderly cohort. Arch Intern Med, 172, 275-7.

REAVEN, G. M. 1995. Pathophysiology of insulin resistance in human disease. Physiol Rev, 75, 473-86.

REED, J., FRAZÃO, E. & ITSKOWITZ, R. 2004. How Much Do Americans Pay for Fruits and Vegetables? U.S. Department of Agriculture, Economic Research Service, AIB-790.

REEDY, J., KREBS-SMITH, S. M., MILLER, P. E., LIESE, A. D., KAHLE, L. L., PARK, Y. & SUBAR, A. F. 2014. Higher diet quality is associated with decreased risk of all-cause, cardiovascular disease, and cancer mortality among older adults. J Nutr, 144, 881-9.

REHM, C. D., MONSIVAIS, P. & DREWNOWSKI, A. 2011. The quality and monetary value of diets consumed by adults in the United States. Am J Clin Nutr, 94, 1333-9.

REHM, C. D., MONSIVAIS, P. & DREWNOWSKI, A. 2015. Relation between diet cost and Healthy Eating Index 2010 scores among adults in the United States 2007–2010. Preventive Medicine, 73, 70-75.

RICKMAN, J. C., BARRETT, D. M. & BRUHN, C. M. 2007. Nutritional comparison of fresh, frozen and canned fruits and vegetables. Part 1. Vitamins C and B and phenolic compounds. J Sci Food Agric, 87, 930-944.

ROBBINS, J. M., VACCARINO, V., ZHANG, H. & KASL, S. V. 2001. Socioeconomic status and type 2 diabetes in African American and non-Hispanic white women

118

and men: evidence from the Third National Health and Nutrition Examination Survey. American Journal of Public Health, 91, 76-83.

ROSE, D. & RICHARDS, R. 2004. Food store access and household fruit and vegetable use among participants in the US Food Stamp Program. Public Health Nutr, 7, 1081-8.

RUMMO, P. E., MEYER, K. A., GREEN HOWARD, A., SHIKANY, J. M., GUILKEY, D. K. & GORDON-LARSEN, P. 2015. Fast food price, diet behavior, and cardiometabolic health: Differential associations by neighborhood SES and neighborhood fast food restaurant availability in the CARDIA study. Health & Place, 35, 128-135.

RUNGE, C. F., JOHNSON, J. & RUNGE, C. P. 2011. Better Milk than Cola: Soft Drink Taxes and Substitution Effects. Choices, 26.

RYDEN, P., MATTSSON SYDNER, Y. & HAGFORS, L. 2008. Counting the cost of healthy eating: a Swedish comparisonof Mediterranean-style and ordinary diets. International Journal of Consumer Studies, 32, 138-146.

RYDEN, P. J. & HAGFORS, L. 2011. Diet cost, diet quality and socio-economic position: how are they related and what contributes to differences in diet costs? Public Health Nutr, 14, 1680-92.

SAHNI, S., TUCKER, K. L., KIEL, D. P., QUACH, L., CASEY, V. A. & HANNAN, M. T. 2013. Milk and yogurt consumption are linked with higher bone mineral density but not with hip fracture: the Framingham Offspring Study. Arch Osteoporos, 8, 119.

SAMARA, A., HERBETH, B., NDIAYE, N. C., FUMERON, F., BILLOD, S., SIEST, G. & VISVIKIS-SIEST, S. 2013. Dairy product consumption, calcium intakes, and metabolic syndrome-related factors over 5 years in the STANISLAS study. Nutrition, 29, 519-24.

SAMUEL, VARMAN T. & SHULMAN, GERALD I. 2012. Mechanisms for Insulin Resistance: Common Threads and Missing Links. Cell, 148, 852-871.

SANCHEZ-MORENO, C., CANO, M. P., DE ANCOS, B., PLAZA, L., OLMEDILLA, B., GRANADO, F. & MARTIN, A. 2003. Effect of orange juice intake on vitamin C concentrations and biomarkers of antioxidant status in humans. Am J Clin Nutr, 78, 454-60.

SARAFIDIS, P. A., LASARIDIS, A. N., NILSSON, P. M., PIKILIDOU, M. I., STAFILAS, P. C., KANAKI, A., KAZAKOS, K., YOVOS, J. & BAKRIS, G. L. 2007. Validity and reproducibility of HOMA-IR, 1/HOMA-IR, QUICKI and McAuley's indices in patients with hypertension and type II diabetes. J Hum Hypertens, 21, 709-16.

SCHRODER, H., MARRUGAT, J. & COVAS, M. I. 2006. High monetary costs of dietary patterns associated with lower body mass index: a population-based study. Int J Obes (Lond), 30, 1574-9.

SCHULZE, M. B., MANSON, J. E., LUDWIG, D. S., COLDITZ, G. A., STAMPFER, M. J., WILLETT, W. C. & HU, F. B. 2004. Sugar-Sweetened Beverages, Weight Gain, and Incidence of Type 2 Diabetes in Young and Middle-Aged Women. JAMA, 292, 927-934.

SHAHAR, D. R., SCHWARZFUCHS, D., FRASER, D., VARDI, H., THIERY, J., FIEDLER, G. M., BLUHER, M., STUMVOLL, M., STAMPFER, M. J. & SHAI,

119

I. 2010. Dairy calcium intake, serum vitamin D, and successful weight loss. American Journal of Clinical Nutrition, 92, 1017-1022.

SIMPLY ORANGE JUICE COMPANY. 2015. Simply Orange Pulp Free Nutrition Facts. Available: http://www.simplyorangejuice.com/product/pulp-free-orange-juice.jsp.

SMITH, K. L., STRAKER, L. M., KERR, D. A. & SMITH, A. J. 2014. Overweight adolescents eat what? And when? Analysis of consumption patterns to guide dietary message development for intervention. J Hum Nutr Diet.

SMITH, T. A., LIN, B.-H. & LEE, J.-Y. 2010. Taxing Caloric Sweetened Beverages: Potential Effects on Beverage Consumption, Calorie Intake, and Obesity. United States Department of Agriculture, Economic Research Service, 1-33.

SONESTEDT, E., WIRFALT, E., WALLSTROM, P., GULLBERG, B., ORHO-MELANDER, M. & HEDBLAD, B. 2011. Dairy products and its association with incidence of cardiovascular disease: the Malmo diet and cancer cohort. Eur J Epidemiol, 26, 609-18.

SOWERS, J. R. 1998. Obesity and cardiovascular disease. Clin Chem, 44, 1821-5. STEWART, H., DONG, D. & CARLSON, A. 2013. Why Are Americans Consuming

Less Fluid Milk? A Look at Generational Differences in Intake Frequency. United States Department of Agriculture, Economic Research Service, 149, 1-29.

STEWART, H., HYMAN, J., BUZBY, J. C., FRAZAO, E. & CARLSON, A. February 2011. How much do fruits and vegetables cost? United States Department of Agriculture, Economic Research Service, EIB-71.

STOCK, J. H., WRIGHT, J. H. & YOGO, M. 2002. A Survey of Weak Instruments and Weak Identification in Generalized Method of Moments. Journal of Business & Economic Statistics, 20, 518-529.

STORE BRANDS Dollar sales share of the private label frozen and canned fruit and vegetables category in U.S. retail stores in 2015, by segment. Available at: http://www.statista.com/statistics/419674/us-dollar-sales-share-of-private-label-frozen-and-canned-fruit-and-vegetables-by-segment/ Last accessed: May 18, 2016.

STURM, R. 2002. The effects of obesity, smoking, and drinking on medical problems and costs. Health Aff (Millwood), 21, 245-53.

STURM, R. & DATAR, A. 2008. Food prices and weight gain during elementary school: 5-year update. Public Health, 122, 1140-1143.

SYMPHONY IRI. 2015. Who We Are [Online]. http://www.iriworldwide.com/About/WhoWeAre.aspx. [Accessed March 12, 2015.

TANDE, D. L., MAGEL, R. & STRAND, B. N. 2010. Healthy Eating Index and abdominal obesity. Public Health Nutr, 13, 208-14.

TERZA, J. V., BASU, A. & RATHOUZ, P. J. 2008. Two-Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling. Journal of health economics, 27, 531-543.

THE BRITISH FROZEN FOOD FEDERATION, LOCAL AUTHORITY CATERERS ASSOCIATION & SHEFFIELD HALLAM UNIVERSITY 2009. Frozen Foods - Use and Nutritional Acceptability in Primary School Lunch Provision, 2009 Report. Food Innovation.

120

THE CAPITOL FORUM 2014. Albertsons/Safeway: Significant Overlaps Likely To Lead to Divestitures; FTC Bureau of Competition Chief Could Lead to More Industry-Friendly Review, but State AGs Present Additional Deal Risk. Retrieved January 26, 2015 from http://thecapitolforum.com/wp-content/uploads/2014/04/ALB-SAFEWAY-2014.03.20.pdf.

THE COUNCIL FOR COMMUNITY AND ECONOMIC RESEARCH (C2ER) 2015. Cost of Living Index Manual. Available at: https://www.coli.org/surveyforms/colimanual.pdf; Last accessed: June 25, 2016.

THOW, A. M., DOWNS, S. & JAN, S. 2014. A systematic review of the effectiveness of food taxes and subsidies to improve diets: Understanding the recent evidence. Nutrition Reviews, 72, 551-565.

TODD, J. E., LEIBTAG, E. & PENBERTHY, C. June 2011. Geographic Differences in the Relative Price of Healthy Foods. U.S. Department of Agriculture, Economic Research Service, EIB-78.

TREMBLAY, A. & GILBERT, J. A. 2009. Milk products, insulin resistance syndrome and type 2 diabetes. J Am Coll Nutr, 28 Suppl 1, 91s-102s.

TRICHOPOULOU, A., COSTACOU, T., BAMIA, C. & TRICHOPOULOS, D. 2003. Adherence to a Mediterranean Diet and Survival in a Greek Population. New England Journal of Medicine, 348, 2599-2608.

TRICHOPOULOU, A., KOURIS-BLAZOS, A., WAHLQVIST, M. L., GNARDELLIS, C., LAGIOU, P., POLYCHRONOPOULOS, E., VASSILAKOU, T., LIPWORTH, L. & TRICHOPOULOS, D. 1995. Diet and overall survival in elderly people. Bmj, 311, 1457-60.

TUOMILEHTO, J., LINDSTROM, J., ERIKSSON, J. G., VALLE, T. T., HAMALAINEN, H., ILANNE-PARIKKA, P., KEINANEN-KIUKAANNIEMI, S., LAASKO, M., LOUHERANTA, A., RASTAS, M., SALMINEN, V. & UUSITUPA, M. 2001. Prevention of type 2 diabetes mellitus by changes in lifestyle among subjects with impaired glucose tolerance. New England Journal of Medicine, 344, 1343-1350.

TURCSIK, R. 2014. Wiping out the competition [Online]. Available: http://www.groceryheadquarters.com/2014/11/paper-products/ [Accessed June 17, 2015.

UNITED STATES CENSUS BUREAU 2012. The 2012 Statistical Abstract, Health & Nutrition: Food Consumption and Nutrition.

UNITED STATES DEPARTMENT OF AGRICULTURE 2014a. Food Expenditures - Table 15. In: SERVICE, E. R. (ed.). http://www.ers.usda.gov/data-products/food-expenditures.aspx.

UNITED STATES DEPARTMENT OF AGRICULTURE 2014b. National Nutrient Database for Standard Reference Release 27, Basic Report: 01016, Cheese, cottage, lowfat, 1% milkfat. http://ndb.nal.usda.gov/ndb/foods/show/16?fg=&man=&lfacet=&format=&count=&max=25&offset=&sort=&qlookup=cottage+cheese.

UNITED STATES DEPARTMENT OF AGRICULTURE 2014c. USDA National Nutrient Database for Standard Reference. The National Agricultural Library http://ndb.nal.usda.gov/.

121

UNITED STATES DEPARTMENT OF AGRICULTURE 2015. Scientific Report of the 2015 Dietary Guidelines Advisory Committee. Available from https://health.gov/dietaryguidelines/2015-scientific-report/pdfs/scientific-report-of-the-2015-dietary-guidelines-advisory-committee.pdf. Last accessed November 20, 2016.

UNITED STATES DEPARTMENT OF AGRICULTURE & AGRICULTURAL RESEARCH SERVICE. 2014. What We Eat in America, NHANES 2011-2012, individuals 2 years and over (excluding breast-fed children), day 1 dietary intake data, weighted. Food Patterns Equivalents Database (FPED) 2011-2012. [Online]. Available: http://www.ars.usda.gov/SP2UserFiles/Place/80400530/pdf/fped/Table_1_FPED_GEN_1112.pdf [Accessed August 3, 2015.

UNITED STATES DEPARTMENT OF AGRICULTURE & ECONOMIC RESEARCH SERVICE. 2013. Urban Influence Codes: Documentation [Online]. Available: http://www.ers.usda.gov/data-products/urban-influence-codes/documentation.aspx [Accessed June 22, 2015.

UNITED STATES DEPARTMENT OF AGRICULTURE & UNITED STATES DEPARTMENT OF HEALTH AND HUMAN SERVICES 2010. Dietary guidelines for Americans, 2010. 7th Edition.

UNITED STATES DEPARTMENT OF AGRICULTURE & UNITED STATES DEPARTMENT OF HEALTH AND HUMAN SERVICES 2015. Dietary guidelines for Americans, 2015-2020. 8th Edition.

UNITED STATES FOOD AND DRUG ADMINISTRATION 2014. CFR - Code of Federal Regulations Title 21, 21CFR101.12. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=101.12.

USDA/CNPP 2008. Development of the CNPP prices database. Available from: http://www.cnpp.usda.gov/sites/default/files/usda_food_plans_cost_of_food/PricesDatabaseReport.pdf; Last accessed: June 25, 2016.

VAZQUEZ, G., DUVAL, S., JACOBS, D. R. & SILVENTOINEN, K. 2007. Comparison of Body Mass Index, Waist Circumference, and Waist/Hip Ratio in Predicting Incident Diabetes: A Meta-Analysis. Epidemiologic Reveiws, 29, 115-128.

VOLPE, R. & OKRENT, A. 2012. Assessing the Healthfulness of Consumers’ Grocery Purchases. EIB-102, U.S. Department of Agriculture, Economic Research Service.

WALKER, R. E., KEANE, C. R. & BURKE, J. G. 2010. Disparities and access to healthy food in the United States: A review of food deserts literature. Health Place, 16, 876-84.

WANG, C. Y., COXSON, P., SHEN, Y.-M., GOLDMAN, L. & BIBBINS-DOMINGO, K. 2012. A penny-per-ounce tax on sugar-sweetened beverages would cut health and cost burdens of diabetes. Health Affairs, 31, 199-207.

WANG, D. D., LEUNG, C. W., LI, Y. & ET AL. 2014. Trends in dietary quality among adults in the united states, 1999 through 2010. JAMA Internal Medicine, 174, 1587-1595.

WATERLANDER, W. E., STEENHUIS, I. H., DE BOER, M. R., SCHUIT, A. J. & SEIDELL, J. C. 2012. Introducing taxes, subsidies or both: the effects of various

122

food pricing strategies in a web-based supermarket randomized trial. Prev Med, 54, 323-30.

WILKINSON, C. P., FERRIS III, F. L., KLEIN, R. E., LEE, P. P., AGARDH, C. D., DAVIS, M., DILLS, D., KAMPIK, A., PARARAJASEGARAM, R. & VERDAGUER, J. T. 2003. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology, 110, 1677-1682.

WILLETT, W. C. & HU, F. B. 2006. Not the Time to Abandon the Food Frequency Questionnaire: Point. Cancer Epidemiology Biomarkers &amp; Prevention, 15, 1757-1758.

WILSON, P. W., D'AGOSTINO, R. B., SULLIVAN, L., PARISE, H. & KANNEL, W. B. 2002. Overweight and obesity as determinants of cardiovascular risk: the Framingham experience. Arch Intern Med, 162, 1867-72.

ZAHEDI, H., KELISHADI, R., HESHMAT, R., MOTLAGH, M. E., RANJBAR, S. H., ARDALAN, G., PAYAB, M., CHINIAN, M., ASAYESH, H., LARIJANI, B. & QORBANI, M. 2014. Association between junk food consumption and mental health in a national sample of Iranian children and adolescents: the CASPIAN-IV study. Nutrition, 30, 1391-7.

ZIMMERMAN, F. J. 2011. Using marketing muscle to sell fat: the rise of obesity in the modern economy. Annual Review of Public Health, 32, 285-306.

ZOU, G. 2004. A Modified Poisson Regression Approach to Prospective Studies with Binary Data. American Journal of Epidemiology, 159, 702-706.

123

APPENDIX 1 : FOR CHAPTER 2

APPENDIX 1.1 Choice of products in the IRI dataset

The following criteria were used to select items (UPCs) within products from the IRI

supermarket price dataset:

1. Selected based on high volume sales or unit sales

Within category, the item had to be in the top Nth percentile (percentile varied

according to the product) of volume sales sold in 2010 (sales equalized to package units)

or top Nth percentile of package units sold in 2010. Item rankings varied by year and so

2010 was used as the index year because our health data come primarily from that year.

Ideally we wanted to use the top 20th percentile for all product categories but had to

select a lower percentile because either (a) a product category had less brand-name

dominance and/or the brands frequently altered their packaging and so UPCs were not

consistent over many years. High volume of unit sales was defined as the top 20th

percentile for carbonated beverages, chocolate candy, and cookies. High volume of unit

sales was defined as the top 40th percentile for salty snacks, frozen vegetables, and

cottage cheese. Milk was not subset according to high volume or unit sales.

2. Selected based on high market saturation

Within categories, for at least 2 out of 4 years, the item had to be sold at a

relatively high percent of IRI supermarkets. This was determined by examining IRI's

analysis of the "all commodity volume" or ACV. We iteratively determined the ACV

threshold so as to obtain a sufficient number of items per category-brand. The minimum

threshold for >2-year ACV was: chocolate candy ≥ 80% of all IRI stores, carbonated

124

beverages ≥ 60%, cookies ≥ 50%, salty snacks ≥ 50%, orange juice ≥ 50%, yogurt ≥

50%, frozen vegetables ≥ 20%, cottage cheese ≥ 20%, and milk ≥ 10%.

3. Diversity of brands

We selected brand-name products because we needed products that had high

potential for being sold in stores across the U.S. Per category, we selected a mean of 5.5

brands (range 3-11) and 15 UPCs (range 5-25). The categories with only 3 brands were

cottage cheese and frozen vegetables, and the lowest number of UPCs was for cottage

cheese (n=9). After brand selection, the researchers confirmed that selected brands were

noted in industry literature as being the top brands.

4. Multiple items in a particular package size

After fulfilling criteria 1-3 above, items were further sorted by product category,

brand name, package size, 2010 volume sales (equivalence), and minimum ACV between

2009-2012 (descending order so highest minimum value). In order to be able to pool

price information we attempted to select at least 2 items with the same package size and

items that were preferably different brands.

Note to criteria #4: Our analyses used a handful of index food product categories

and small number of index UPCs to compare food prices between regions and between

product categories. Because price per ounce declines as package size increases, ideally

we would select items that are similar across product categories according to "intended

servings per package" and "package size". However, this restriction was not feasible

given criteria 1-3 above. Thus, we selected a diversity of package sizes and standardized

to servings per package and package size before averaging price within a product

125

category and before comparing relative prices between regions and between product

categories. Mean servings per package was 2-5 for chocolate candy/cottage

cheese/frozen vegetables and 7-8 for carbonated beverages and salty snacks.

126

APPENDIX 1.2 Details on healthy and unhealthy products included in the price calculations

Details of products included in each category

Healthy food prices are represented by a combination of the following products:

milk, frozen vegetables, yogurt, and cottage cheese. Brands that had significant regional

and national penetration were selected in order to maintain product uniformity across all

metro areas being examined. Milk includes 15 products (All ½ gallon products. One 1

gallon product was initially included but was not available in 66% of stores, and because

one gallon products had significantly different prices per serving compared with half

gallon products, it was removed). Data include regular and organic milk, and reduced fat

(1% and 2%) and skim milk. Frozen vegetables include 23 products with package sizes

range from 7 to 24 ounces. Cottage cheese includes 9 products, all of which were either

16 or 24 ounces and included 1%, 2%, or 4% milkfat. Yogurt includes 17 products with

package sizes ranging from 6 to 48 ounces and includes low-fat and non-fat products.

Orange juice includes 12 not from concentrate 100% orange juice products with package

sizes ranging from 13.5 to 128 fluid ounces.

Unhealthy foods are represented by the following products: chocolate candy,

cookies, salty snacks, and soda. Chocolate candy includes single and multiple serving

packages of candy bars/pieces ranging from 1.4 ounces (e.g., Hershey’s milk chocolate

bar with almonds) to 12.6 ounces (a bag of M&Ms). Products include packaging of

regular size and “fun” size pieces across 22 different products. Cookies include 13

products in package sizes ranging from 7.5 to 31.8 ounces. These include Oreo, Chips

Ahoy, Nilla Wafers, among other popular products. Salty snacks include 23 products,

127

including pretzels, potato chips, tortilla chips and other similar products. Product sizes

range from 6.4 ounces (Pringles) to 18 ounces (Tostitos). Soda includes 25 products

(excluding diet or reduced calorie items), including single 20 ounce and 2-liter bottles, as

well as 12-packs of 12 ounce cans.

Rationale for the selection of unhealthy foods

Representatives for the unhealthy food category were chosen for their long shelf

life and high use of preservatives. Additionally, each of these foods has poor nutritional

quality; they are high in added sugar, saturated fat, and/or sodium, with very little protein

or vitamins. Increased consumption of these foods has been associated with obesity,

diabetes, and other diseases, detailed below.

Soda: Soda may be the ultimate unhealthy food, as it supplies no nutritional value

yet is calorically dense, while ironically providing low satiety (Malik et al., 2006). With

all of its 140 calories per 12 ounce can coming from refined sugars, most often high

fructose corn syrup, and containing no vitamins or minerals, soda has been referred to as

liquid candy (Jacobson, 2005). The USDA recommends no more than 48 grams of solid

fats and added sugars in a 2,000 calorie diet (United States Department of Agriculture

and United States Department of Health and Human Services, 2010), yet a single can of

soda includes 33 grams of added sugar alone United States Department of Agriculture

(2014c). Despite the lack of nutritional value, soda consumption is higher in the United

States than anywhere else in the world, with approximately one-half of the US population

consuming a sugar drink on any given day (Ogden et al., 2011). Because of its sweet

taste, poor nutritional value, low price, and high rates of consumption, soda consumption

128

has been linked to a number of diseases, including obesity, cardiovascular disease, and

diabetes (Bray and Popkin, 2014, Dhingra et al., 2007, Hu, 2013, Hu and Malik, 2010,

Malik and Hu, 2012, Malik et al., 2010b, Schulze et al., 2004, de Koning et al., 2011b,

Malik et al., 2010a, Berkey et al., 2004, Ludwig et al., 2001). Thus soda is well suited to

be considered a representative of unhealthy food.

Salty snacks, cookies, and chocolate candy: The remaining unhealthy food

representatives (chocolate candy, salty snacks, and cookies) were chosen because they

are commonly considered “junk foods” due to their low nutritional quality (Brunello et

al., 2014, Gittelsohn et al., 1998, Magee, 2014, Smith et al., 2014, Zahedi et al., 2014).

Specifically because these foods are high in sodium, added sugar and/or saturated fat

(Brunello et al., 2014). The health effects associated with consumption of such junk food

range from an increased risk of diabetes (Gittelsohn et al., 1998) to poorer mental health

(Zahedi et al., 2014). For an example of the poor nutritional content, consider a regular

1.55 oz Hershey’s milk chocolate bar. A single bar contains 210 calories, 8 grams of

saturated fat (13g of total fat, 20% of the RDA), and 24 grams of sugar (Hershey's, 2014).

Additionally, a 30 gram serving of Chips Ahoy! Original Chocolate Chip cookies,

approximately 3 cookies, contains 144 calories, 7.3 grams of fat, and 10 grams of sugar,

and under 1 gram of protein (Nabisco, 2014). And for salty snacks, a single serving of

approximately 12 Nacho Cheese Doritos chips (30 grams) contains 150 calories, 8.6

grams of fat, and 225 mg of sodium (Frito Lay, 2014). All three of these foods, which are

typical of their overall product category, are calorically dense and high in added sugar,

saturated fat, and/or sodium, in addition to being high in preservatives resulting in a long

shelf life, and hence considered unhealthy in our study.

129

Rationale for the selection of healthy foods

Foods selected to represent the healthy food category are those that can be

considered a proxy for all fresh foods that have a limited shelf life, such as fresh fruits

and vegetables, and include vegetables (frozen), fruit juice (as a proxy for fruits), and

dairy products. Additionally, many of the foods being considered healthy have been

associated with positive health outcomes, including lower rates of diabetes, obesity, and

heart disease, among others.

Frozen vegetables: Although this study did not have access to the price of fresh

fruits and vegetables we were able to capture the price of frozen vegetables sold in these

supermarkets. The price of frozen vegetables is very similar to that of their fresh

counterparts, with some frozen vegetables slightly more expensive than fresh, while

others are somewhat cheaper (Stewart et al., February 2011). Although the shelf life of

frozen vegetables is longer than their fresh counterparts, they have similar health benefits

(Rickman et al., 2007, The British Frozen Food Federation et al., 2009, Favell, 1998).

Frozen vegetables have high nutritional content; for example, one serving (85

grams)(United States Food and Drug Administration, 2014) of mixed vegetables from

Birds Eye, which includes frozen carrots, corn, peas and green beans, contains 2.4 grams

of fiber, 2.3 grams of protein, 120% of Vitamin A recommended daily allowance (RDA),

and just 67 calories (Birds Eye, 2016). Furthermore, consumption of vegetables (alone or

studied in combination with fruit consumption) has been linked to health benefits,

including lower rates of diabetes in women (Bazzano et al., 2008) and in men (Mursu et

al., 2014), as well as reduced rates of cardiovascular disease (Hung et al., 2004) and

130

coronary heart disease (He et al., 2007). Because of these reasons, frozen vegetables

serve as a proxy for healthy food in general.

Orange juice: Orange juice is included as a health food representative mainly to

act as a proxy for the price of fresh fruit. The fruit juice included in this study is 100%

orange juice and thus the price of oranges is directly reflected in the price of the juice.

The price of an edible cup of an orange is exactly the same whether it is from a fresh

navel orange or from ready to drink orange juice ($0.34 per edible cup) (Stewart et al.,

February 2011). And over time the price of orange juice tracks closely with the price of

raw oranges: from 2001 through the fall of 2011 the correlation between the price of

oranges and retail 100% orange juice was 74% (Morris, 2011). While the health benefits

of fresh fruit, including oranges, is well established, the health effects of drinking orange

juice are mixed. Some studies have found a modest increase in risk of diabetes with

greater consumption of fruit juice (Bazzano et al., 2008, Muraki et al., 2013), while

others have found benefits of increased vitamin C and antioxidant biomarker

concentrations (Sanchez-Moreno et al., 2003) and increased HDL levels (Kurowska et

al., 2000). Thus, orange juice is a proxy for healthy foods because of its pricing

similarities to the fresh, healthy, fruit it is created from, rather than its own health

benefits.

Dairy products (milk, yogurt, and cottage cheese): Dairy products were chosen

because of their short shelf life, similar to that of fresh fruits and vegetables, and because

of their high nutritional value and health benefits. Milk and other dairy contain numerous

vitamins and minerals, including vitamin D and calcium, among many others (Runge et

al., 2011). While milk may contain a significant number of calories – a 12 ounce glass of

131

1% low-fat milk will contain 154 calories – 32% comes from protein, 21% from fat, and

the remaining amount from the natural sugar lactose (United States Department of

Agriculture, 2014c). This is a stark contrast to the 100% of calories in soda derived from

added sugars. The consumption of skim milk been shown to satiate the appetite and lead

to fewer consumed calories compared with fruit drinks in the morning (Dove et al.,

2009), and thus a healthy option at breakfast. Vitamin D and dairy calcium are associated

with diet-induced weight loss (Shahar et al., 2010) and reduced odds of insulin resistance

syndrome in overweight persons (Pereira et al., 2002), all of which are applicable to the

outcomes being studied in this research.

Yogurt is similar to fluid milk in that it has a limited shelf life and is rich in

nutrients, with the additional benefit of probiotics. There have been numerous studies of

the potential health benefit of yogurt consumption, particularly within older adults (El-

Abbadi et al., 2014), i.e., the population of interest in this dissertation. Yogurt

consumption has been associated with improved bone mineral density (Sahni et al.,

2013), a lower risk of cardiovascular disease (Sonestedt et al., 2011), and a lower risk of

type 2 diabetes (Margolis et al., 2011) within older populations. Cottage cheese is the

third dairy product to be included, which has many of the same benefits of yogurt and

fluid milk. Although cottage cheese consumption in the United States is fairly small when

compared to other dairy products (Davis et al., 2010), it has significant nutritional value.

A single serving of 1% milk-fat cottage cheese contains 14 grams of protein, but just 1

gram of fat and 79 calories total, in addition to being a significant source of calcium,

phosphorus, riboflavin, and vitamin B-12 (United States Department of Agriculture,

132

2014b). Cottage cheese consumption is also more prevalent in older adults (Davis et al.,

2010), making it applicable to the population in our study.

Overall, milk products, including cottage cheese and yogurt, are associated with

decreased prevalence of insulin resistance and type-2 diabetes (Tremblay and Gilbert,

2009, Fumeron et al., 2011), metabolic syndrome (Fumeron et al., 2011, Pfeuffer and

Schrezenmeir, 2007, Samara et al., 2013) and obesity.(Major et al., 2008) Because of

these reasons, and their relatively short shelf life, all three of these dairy products are

strong candidates for representatives of the healthy food category.

133

APPENDIX 1.3 Principal component analysis

A two-stage principal component analysis (PCA) was conducted to identify

products that statistically grouped together in order to reduce the dimensionality of the

data and the number of outcomes analyzed. We identified product-item prices that were

highly correlated with each other and distinct from other product groups. All retained

items had component loading scores ≥0.60. This analysis was done in two steps.

The first PCA was conducted by analyzing all individual items (e.g., 20 ounce

Coca-Cola Classic) within each product (e.g., soda). Consistent across all products, two

factors were identified within each product according to the volume or weight of the

items; one factor included all lower volume/weight items (e.g., 2-liter and 20-oz

individual bottles of soda) while the other included all higher volume/weight items (e.g.,

cases of twelve 12-oz cans and 6-packs of 20-oz sodas). The only exception to this was

cottage cheese for which all items (16-oz and 24-oz) grouped into a single factor. Two

price variables were then created for each product by grouping items with similar

volume/weight according to the results of the PCA (with the exception of one for cottage

cheese).

The second PCA was then conducted using the newly created product variables.

The PCA found that healthy food products grouped into two categories: dairy (made up

of yogurt, cottage cheese, and milk) and fruits and vegetables (comprised of orange juice

and frozen vegetables). For unhealthy foods, cookies and chocolate candy grouped

together, while soda and salty snacks remained independent of other products, resulting

in three unhealthy product classes. The different sizes within each product identified in

the first PCA always grouped together in the second step, thus all items from the same

134

product category were kept together. The five product categories resulting from the

second PCA were used for the remainder of all analyses.

Supplemental Table 1 Rotated component matrix from the principal component analysis of healthy foods illustrating the component loading scores of each item for dairy (milk, yogurt and cottage cheese) and fruits and vegetables (orange juice and frozen vegetables)

Component 1 2

Milk (half-gallon) .764 .295 Orange juice (≤59 ounces) .138 .830 Orange juice (≥89 ounces) .450 .727 Yogurt (≤24 ounces) .740 .244 Yogurt (≥32 ounces) .892 .133 Cottage cheese .833 .288 Frozen vegetables (≤10 ounces) .615 .632 Frozen vegetables (≥12 ounces) .204 .826

Extraction Method: Principal Component Analysis. Rotation Method: Varimax with Kaiser Normalization. Rotation converged in 3 iterations.

Supplemental Table 2 Rotated component matrix from the principal component analysis of unhealthy foods illustrating the component loading scores of each item for soda, sweets (chocolate and cookies) and salty snacks

Component

1 2 3 Soda (multi pack) .149 .914 .189 Soda (single bottle) .364 .837 -.011 Chocolate candy (≤5 ounces) .848 .118 .069 Chocolate candy (≥9 ounces) .775 .173 .173 Cookies (≤12 ounces) .795 .515 -.005 Cookies (≥15 ounces) .719 .360 -.183 Salty snacks (≤11 ounces) .489 .039 .713 Salty snacks (≥13 ounces) -.162 .102 .879

Extraction Method: Principal Component Analysis. Rotation Method: Varimax with Kaiser Normalization. Rotation converged in 5 iterations.

135

APPENDIX 1.4 Distribution of the healthy-to-unhealthy price ratio

The healthy-to-unhealthy price ratio followed a roughly normal distribution as

illustrated in the histogram below (Figure 1). The normal distribution of the data allowed

for meaningful analysis when considering the impact of a one standard deviation change

in the measure on the outcomes examined in Chapters 3 and 4.

Figure 1 Distribution of the health-to-unhealthy price ratio for the 1953 stores in the IRI database. The ratio had a mean of 1.97 and standard deviation of 0.19.

136

APPENDIX 1.5 Validation of the IRI price dataset against C2ER

As a measure of external validity of the price data we looked at correlations with

the Cost of Living Index (COLI) from the Council for Community and Economic

Research (C2ER). Store prices from the IRI data were rolled-up to the level of core-based

statistical area (CBSA), also referred to as a metropolitan area, to match the spatial

resolution of the COLI data. A subset of products from IRI were then chosen to most

closely match those included in the COLI calculations; this included ½ gallon milk (in

IRI this was 1%, 2%, and fat free, while in COLI it is whole milk), frozen corn (12 to 16

oz. from IRI vs. 16 oz. from COLI), orange juice (48 to 89 oz. in IRI vs. 64 oz. in COLI;

included only Tropicana and Florida’s natural for both data sources), soda (2-liter of

Coca-Cola from both sources), and potato chips (8.5 and 9.5 oz. from IRI vs. 12 oz. plain

chips from COLI). Spearman rank correlations were performed and found moderate

correlations for orange juice (r=0.305, p<0.001), potato chips (r=0.425, p<0.001), and

soda (r=0.540, p<0.001), low correlation for frozen corn (r=0.149, p<0.001), and no

correlation for milk (r=0.013, p=0.592). The lack of correlation for milk may have been

due to only branded and organic low and non-fat products included in the IRI data vs. any

type of whole milk in COLI.

Supplemental Table 3 Spearman rank correlations for products found in the IRI dataset and C2ER

Spearman rank

correlation p-

value Orange juice a 0.305 <0.001 Frozen corn b 0.149 <0.001 Milk c 0.013 0.592 Soda d 0.540 <0.001

137

Spearman rank

correlation p-

value Potato chips e 0.425 <0.001

a Orange juice includes 48 o 89 ounce Tropicana or Florida's Natural from IRI and 64 ounce packages of the same brands from C2ER b Frozen corn includes 12 to 16 ounces packages from IRI and 16 ounce packages from C2ER c Milk includes branded half gallon milk (1%, 2%, fat free) from IRI and any half gallon whole milk from C2ER d Soda includes 2-liter bottles of Coca-Cola from IRI and C2ER e Potato chips includes 8.5 and 9ounce Baked Lays, Baked Ruffles, Kettle Cooked Lays from IRI and 12 ounce plain chips from C2ER

138

APPENDIX 1.6 Additional supplemental tables for Chapter 2

Supplemental Table 4 Product selection criteria for each of the food categories used in this study

Category Number brands

Number UPCs Example brands Exclusions

Carbonated Beverages

9 25 7-UP, A&W, Canada Dry, Coca Cola, etc.

Chocolate Candy 11 22 Hershey, Kit Kat, M&M, Milky Way, Nestle, etc.

Cookies 4 13 Keebler, Little Debbie, Nabisco, Pepperidge Farm

Cottage cheese 3 9 Breakstone, Daisy, General Mills

Specialty products, such as lactose free cottage cheese, were not included.

Orange juice 3 12 Florida's Natural, Simply, Tropicana

Added sugar, not 100% orange juice

Milk 5 14 Farmland, Hood, Horizon, Organic Valley, Stonyfield

Full fat; specialty milks: soy, almond

Frozen vegetables 3 23 Bird's Eye, Green Giant, Pictsweet

Seasoned, glazed, creamed

Salty snacks 10 23 Lays, Funyuns, Pringles, Snyders, Tostitos, etc.

Yogurt 5 17 Dannon, Fage, Stonyfield, Yoplait

Products marketed towards children (e.g., Trix brand yogurt) or those that are highly sweetened (e.g., crunch and whips).

139

Supplemental Table 5 Nutrition information for each of the healthy and unhealthy food categories included in this study

FDA serving size1

Key nutrition information2

Healthy Foods

Frozen vegetables 88 g 60 calories, 2 g protein, 2 g fiber Low-fat milk 8 fl oz 102 calories, 8 g protein, 13 g natural sugar, 2 g

saturated fat Orange juice 8 fl oz 110 calories, 2 g protein, 23 g sugar Yogurt 225 g 132 calories, 11 g protein, 16 g sugar, 2 g

saturated fat Cottage cheese 110 g 79 calories, 14 g protein, 1 g saturated fat

Unhealthy foods

Soda 8 fl oz 93 calories, 26 g added sugar, 0 g protein Cookies 30 g 144 calories, 7 g fat, 10 g sugar, 1 g protein Salty snacks 30 g 150 calories, 9 g fat, 225 mg sodium, 2 g

protein Chocolate candy 40 g 190 calories, 7 g saturated fat

1: Source: (United States Food and Drug Administration, 2014) 2: Nutrition information for each product is represented by one item within that group: frozen vegetables (Birds Eye mixed vegetables(Birds Eye, 2016)), low-fat milk (1% milk-fat (United States Department of Agriculture, 2014c)), orange juice (Simply Orange pulp free (Simply Orange Juice Company, 2015)), yogurt (Dannon All Natural Plain Lowfat (Dannon, 2015)), cottage cheese (1% milk-fat (United States Department of Agriculture, 2014b)), soda (Coca-Cola Classic (United States Department of Agriculture, 2014c)), cookies (Chips Ahoy! Original (Nabisco, 2014)), salty snacks (Doritos Nacho Cheese (Frito Lay, 2014)), and chocolate candy (Hershey’s milk chocolate bar (Hershey's, 2014)) Abbreviations: g = grams, fl oz = fluid ounces, mg = milligrams

140

Supplemental Table 6 Serving sizes and weights for composite price calculations using two different weight calculations – based on national consumption averages and using equal weights within each food class

Serving size

Servings per day Weight a Weight b

Healthy foods Fruits and vegetables c (rep. by OJ and frozen veg)

1 cup 2.63 0.617 0.5

Dairy d 8 fl oz 1.63 0.383 0.5 Unhealthy foods

Soda e 8 fl oz 0.73 0.450 0.33 Sweets f,g 30g/40g 0.62 0.383 0.33 Salty snacks h 30 g 0.27 0.167 0.33

a Weight based on national consumption averages (servings per day column) b Equal weight given to all products in their respective class c NHANES 2011-2012 survey found adults 20 years and older consumed 1.64 cups of vegetables per day and 0.99 cups of fruit per day(United States Department of Agriculture and Agricultural Research Service, 2014) d NHANES 2011-2012 survey found adults 20 years and older consumed 1.63 cups of dairy per day. (United States Department of Agriculture and Agricultural Research Service, 2014) e Adults ≥60 years old consume 68 calories from SSBs on a daily basis.(Kit et al., 2013) A 12 ounce can of Coca-Cola contains 140 calories,(Coca-Cola, 2014) thus 68 calories equates to 5.8 ounces, or 0.73 daily servings f According to a Canadian agriculture report, 2010 cookie sales was 1 million tons in the United States,(Agriculture and Agri-food Canada, 2012) which equates to 0.27 servings per day. g According to the latest report from the International Cocoa Organization, in 2008 Americans consumed 5.09 kilograms of chocolate on average, roughly 11.2 pounds, which equates to 0.49 ounces per day, or 0.35 servings.(International Cocoa Organization) h A 2004 report from the USDA found that 95.5% of households consumed any chips (potato, corn, etc.) and of those that did eat chips they consumed 7.0 pounds per capita.(Kuchler et al., 2004) After accounting for those that did not buy any chips this becomes 6.7 pounds per capita annually, 0.29 ounces per day, or 0.27 servings. Abbreviations: g = grams, fl oz = fluid ounces

141

Supplemental Table 7 Food prices according to weights based on national consumption averages and equal weights

All healthy food

(weight 1)a All healthy food

(weight 2)b Mean SD Mean SD Aggregate healthy food price $0.590 $0.056 $0.619 $0.065 Aggregate unhealthy food price $0.298 $0.024 $0.294 $0.021 Healthy-to-unhealthy price ratio 1.99 0.19 2.11 0.21

a Weight 1: Aggregate healthy food prices and unhealthy food prices use weights according to national consumption estimates of vegetables, fruits, and dairy for healthy foods, and sugar sweetened beverages, chocolate, cookies, and chips (potato, corn, etc.) for unhealthy food. b Weight 2: Aggregate healthy food prices, unhealthy food prices use equal weights within each category (i.e., fruits & vegetables and dairy each received a weight of 1/2 for the healthy food price; and soda, sweets, and salty snacks each received a weight of 1/3 for the unhealthy price

142

Supplemental Table 8 Price of each product category by demographic characteristics

Healthy foods Unhealthy foods Ratio

# of

stores

Fruits + vegetables Dairy

All healthy food a Soda Sweets Salty Snacks

All unhealthy food a

Healthy-to- Unhealthy a

Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD

Overall 1953 $0.489 $0.051 $0.760 $0.100 $0.590 $0.056 $0.217 $0.029 $0.375 $0.037 $0.290 $0.014 $0.298 $0.024 $1.990 $0.190

Neighborhood SES quantile b Lowest quintile (least advantaged)

391 $0.460 $0.042 $0.788 $0.089 $0.581 $0.047 $0.219 $0.024 $0.370 $0.032 $0.297 $0.017 $0.299 $0.019 1.950 0.175

Second quintile 391 $0.476 $0.040 $0.759 $0.092 $0.581 $0.048 $0.215 $0.025 $0.370 $0.034 $0.292 $0.013 $0.295 $0.020 1.974 0.177

Middle quintile 390 $0.487 $0.039 $0.760 $0.102 $0.589 $0.053 $0.213 $0.024 $0.373 $0.028 $0.288 $0.011 $0.295 $0.017 2.003 0.196

Fourth quintile 391 $0.496 $0.040 $0.747 $0.106 $0.590 $0.057 $0.213 $0.025 $0.373 $0.026 $0.285 $0.011 $0.294 $0.016 2.014 0.204 Highest quintile (most advantaged)

390 $0.526 $0.065 $0.747 $0.103 $0.611 $0.068 $0.226 $0.043 $0.389 $0.055 $0.287 $0.013 $0.306 $0.039 2.010 0.190

Proportion Black/Hispanic quintile Lowest quintile (0.8% to 14.0% Bl/Hisp) 390 $0.501 $0.060 $0.741 $0.086 $0.592 $0.058 $0.216 $0.039 $0.378 $0.047 $0.286 $0.013 $0.297 $0.033 $2.002 $0.182 Second quintile (14.0% to 24.0%) 392 $0.499 $0.051 $0.743 $0.096 $0.591 $0.058 $0.219 $0.031 $0.375 $0.037 $0.288 $0.012 $0.298 $0.026 $1.990 $0.187 Middle quintile (24.0% to 37.8%) 390 $0.494 $0.045 $0.760 $0.112 $0.593 $0.060 $0.216 $0.026 $0.374 $0.034 $0.289 $0.012 $0.297 $0.022 $2.004 $0.203 Fourth quintile (37.9% to 58.4%) 391 $0.486 $0.046 $0.759 $0.102 $0.587 $0.054 $0.217 $0.024 $0.373 $0.032 $0.292 $0.013 $0.297 $0.019 $1.981 $0.194 Highest quintile (58.5% to 98.8%) 390 $0.465 $0.046 $0.798 $0.090 $0.588 $0.050 $0.217 $0.023 $0.374 $0.033 $0.294 $0.017 $0.299 $0.018 $1.974 $0.184

Urban classification

Large metro (pop. ≥1 million) 1606 $0.492 $0.055 $0.769 $0.102 $0.596 $0.058 $0.217 $0.031 $0.377 $0.039 $0.288 $0.013 $0.298 $0.026 $2.007 $0.192

Small metro (pop. <1 million) 296 $0.476 $0.028 $0.725 $0.083 $0.568 $0.042 $0.221 $0.018 $0.368 $0.027 $0.296 $0.013 $0.298 $0.015 $1.912 $0.163

Rural (pop <50k) 51 $0.467 $0.019 $0.701 $0.055 $0.553 $0.023 $0.217 $0.018 $0.354 $0.026 $0.301 $0.014 $0.291 $0.015 $1.904 $0.139

Region

143

Healthy foods Unhealthy foods Ratio

# of

stores

Fruits + vegetables Dairy

All healthy food a Soda Sweets Salty Snacks

All unhealthy food a

Healthy-to- Unhealthy a

Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD

Northeast 207 $0.509 $0.087 $0.730 $0.095 $0.594 $0.074 $0.237 $0.056 $0.396 $0.069 $0.288 $0.016 $0.314 $0.052 $1.912 $0.179

Midwest 173 $0.489 $0.032 $0.782 $0.038 $0.599 $0.026 $0.172 $0.020 $0.383 $0.034 $0.277 $0.011 $0.279 $0.014 $2.155 $0.106

South 1009 $0.468 $0.024 $0.703 $0.052 $0.555 $0.024 $0.217 $0.019 $0.365 $0.030 $0.296 $0.013 $0.295 $0.018 $1.888 $0.138

West 564 $0.519 $0.057 $0.867 $0.090 $0.650 $0.043 $0.225 $0.015 $0.382 $0.026 $0.283 $0.007 $0.302 $0.013 $2.152 $0.143 a Composite healthy and unhealthy food prices were weighted based on national consumption averages of each food category b Neighborhood SES was derived from log of the median household income; log of the median value of housing units; the percentage of households receiving interest, dividend, or net rental income; the percentage of adults 25 years of age or older who had completed high school; the percentage of adults 25 years of age or older who had completed college; and the percentage of employed persons 16 years of age or older in executive, managerial, or professional specialty occupations.

144

Supplemental Table 9 Sensitivity analysis of the hierarchical models examining price outcomes on neighborhood socioeconomic status (SES) and the proportion of individuals who are Hispanic or black using a 2-mile buffer

Estimate a

95% CI Model Lower Upper P-value SES MODELS: Mean difference in price per 20 percent change in neighborhood socioeconomic index

Healthy food price Unadjusted b 0.00434 0.00320 0.00548 <0.0001

Adjusted c 0.00073 -0.00093 0.00240 0.3881

Fruits and veggies Unadjusted b 0.01342 0.01215 0.01469 <0.0001

Adjusted c 0.01080 0.00890 0.01271 <0.0001

Dairy Unadjusted b -0.01304 -0.01514 -0.01093 <0.0001

Adjusted c -0.01869 -0.02192 -0.01546 <0.0001

Unhealthy food price Unadjusted b 0.00035 -0.00015 0.00085 0.1741

Adjusted c 0.00020 -0.00049 0.00089 0.5649

Soda Unadjusted b 0.00038 -0.00023 0.00098 0.2209

Adjusted c 0.00038 -0.00057 0.00132 0.4357

Sweets Unadjusted b 0.00231 0.00145 0.00317 <0.0001

Adjusted c 0.00107 -0.00002 0.00215 0.0547

Salty snacks Unadjusted b -0.00226 -0.00262 -0.00191 <0.0001

Adjusted c -0.00136 -0.00192 -0.00079 <0.0001

Healthy-to-Unhealthy price ratio Unadjusted b 0.01218 0.00826 0.01610 <0.0001

Adjusted c 0.00188 -0.00447 0.00824 0.5615

BLACK/HISPANIC MODELS: Mean difference in price per 20 percent change in neighborhood proportion of Black/Hispanic

Healthy price Unadjusted b -0.00295 -0.00412 -0.00178 <0.0001

Adjusted c 0.00021 -0.00148 0.00190 0.8071

Fruits and veggies Unadjusted b -0.01005 -0.01141 -0.00869 <0.0001

Adjusted c -0.00013 -0.00205 0.00180 0.8973

Dairy Unadjusted b 0.01067 0.00850 0.01284 <0.0001

Adjusted c 0.00115 -0.00214 0.00444 0.4925

Unhealthy price Unadjusted b 0.00079 0.00027 0.00130 0.0027

Adjusted c 0.00199 0.00128 0.00269 <0.0001

Soda Unadjusted b 0.00035 -0.00027 0.00096 0.2676

Adjusted c 0.00068 -0.00028 0.00165 0.1669

Sweets Unadjusted b -0.00019 -0.00108 0.00069 0.6671

Adjusted c 0.00344 0.00234 0.00455 <0.0001

Salty snacks Unadjusted b 0.00231 0.00194 0.00267 <0.0001

145

Estimate a

95% CI Model Lower Upper P-value

Adjusted c 0.00121 0.00064 0.00179 <0.0001

Healthy-to-Unhealthy price ratio Unadjusted b -0.01546 -0.01944 -0.01148 <0.0001

Adjusted c -0.01149 -0.01798 -0.00500 0.0005 a per 20 percentile change in neighborhood SES or Black/Hispanic b Models did not include covariates but were adjusted for county and state via model nesting c Includes covariates: Age, region, urbanicity, population density, supermarket density, toilet paper price, and either race (in the SES models) or neighborhood SES (in the race models).

146

APPENDIX 2 : FOR CHAPTER 3

APPENDIX 2.1 Supplemental tables for Chapter 3

Supplemental Table 10 Characteristics of included and excluded individuals in the analysis of diet quality

Included

participants Excluded

individuals

N /

Mean Col %

/ SD n /

Mean % / SD Number of participants (N) 2765 1951 MESA recruitment site (n, %) a

Forsyth County, NC 539 19.5% 274 14.0% New York, NY 538 19.5% 272 13.9% Baltimore, MD 464 16.8% 194 9.9% St. Paul, MN 8 0.3% 763 39.1% Chicago, IL 651 23.5% 225 11.5% Los Angeles, CA 565 20.4% 223 11.4%

Region of residence (n, %) Northeast 534 19.3% 241 13.0% Midwest 642 23.2% 937 50.5% South 1021 36.9% 452 24.3% West 568 20.5% 227 12.2%

Total supermarket density (3 mile) (mean, SD) 1.19 1.42 0.79 1.19 Female (n, %) 1466 53.0% 1048 53.7% Age (mean, SD) 70.3 9.5 69.4 9.5 Race/ethnicity (n, %)

White 1101 39.8% 824 42.2% Chinese American 359 13.0% 182 9.3% Black 834 30.2% 417 21.4% Hispanic 471 17.0% 528 27.1%

Education (n, %) High school diploma or less 777 28.1% 718 36.8% Some college 761 27.5% 610 31.3% Bachelor’s degree or more 1227 44.4% 615 31.5%

Per capita household income (in $10k) 2.6 1.9 2.3 1.7 Wealth index 2.6 1.2 2.5 1.2 Income/wealth index 5.1 2.2 4.8 2.2 Marital status (n, %)

147

Not married or living with partner 1107 40.0% 825 42.3% Married/Living with partner 1658 60.0% 1126 57.7%

BMI (mean, SD) 28.2 5.6 28.9 5.7 <25 (n, %) 855 30.9% 512 26.2% 25-29.9 (n, %) 1043 37.7% 710 36.4% ≥30 (n, %) 867 31.4% 729 37.4%

Smoking status (n, %) Never smoked 1281 46.3% 841 43.1% Former smoker 1283 46.4% 938 48.1% Current smoker 201 7.3% 172 8.8%

Physical activity, MET min per week (mean, SD) 2773.7 3552.0 2548.0 3366.9 a This is the MESA location of the participant, not necessarily their area of residence

148

Supplemental Table 11 Results of sensitivity analyses using equal weights for the price outcomes, using a 5-mile radius to capture prices for all individuals, and using a 1-mile radius for those living in New York City and a 3-mile radius for all others

Exposure of interest Healthy-to-unhealthy ratio Healthy food price Unhealthy food price 95% CI 95% CI 95% CI

Odds ratio Lower Upper p-

value Odds ratio Lower Upper p-

value Odds ratio Lower Upper p-

value Using equal weights for food prices a

Model 1: region, age, gender 0.92 0.76 1.11 0.3905 0.96 0.86 1.09 0.552 1.00 0.92 1.10 0.9625 Model 2: Model 1 plus income/wealth, education level, smoking status, and race

0.84 0.69 1.03 0.0869 0.97 0.86 1.10 0.6684 1.02 0.93 1.12 0.6297

Final Model: Model 2 plus neighborhood SES and neighborhood supermarket density

0.72 0.57 0.91 0.0051 1.05 0.90 1.23 0.5163 1.16 1.01 1.32 0.0314

Using a 5-mile radius for food prices b Model 1: region, age, gender 1.01 0.84 1.21 0.9487 0.97 0.85 1.10 0.6063 0.98 0.88 1.09 0.6974 Model 2: Model 1 plus income/wealth, education level, smoking status, and race

0.83 0.68 1.01 0.0562 0.95 0.83 1.07 0.3860 1.01 0.91 1.13 0.8417

Final Model: Model 2 plus neighborhood SES and neighborhood supermarket density

0.78 0.63 0.96 0.0216 0.93 0.79 1.10 0.4071 1.05 0.91 1.21 0.5231

Using a 1-mile radius for those in NYC c Model 1: region, age, gender 0.98 0.85 1.14 0.8098 1.01 0.88 1.16 0.8980 1.03 0.91 1.15 0.6798 Model 2: Model 1 plus income/wealth, education level, smoking status, and race

0.89 0.76 1.04 0.1447 0.98 0.85 1.13 0.7611 1.06 0.93 1.19 0.3815

Final Model: Model 2 plus neighborhood SES and neighborhood supermarket

0.86 0.73 1.01 0.0608 0.97 0.83 1.13 0.6957 1.07 0.94 1.22 0.2792

149

Exposure of interest Healthy-to-unhealthy ratio Healthy food price Unhealthy food price 95% CI 95% CI 95% CI

Odds ratio Lower Upper p-

value Odds ratio Lower Upper p-

value Odds ratio Lower Upper p-

value density

a Each of the two healthy food products (fruits & vegetables, dairy) received a weight of 0.5, while each of the three unhealthy products (soda, chocolate candy & sweets, salty snacks) received a weight of 0.33. The radius for capturing prices remained at 3-miles for this sensitivity analysis. b All IRI supermarkets within 5-miles of a participant’s place of residence at Exam 5 were included c A 1-mile radius was used for individuals living in New York City, while a 3-mile radius was kept for all others

150

APPENDIX 3 : FOR CHAPTER 4

APPENDIX 3.1 Supplemental tables for Chapter 4

Supplemental Table 12 Patient characteristics of those who were included in the study (n=3,408) and those who were excluded (n=1,308)

Included participants Excluded

individuals

N /

Mean Col % /

SD N /

Mean Col % /

SD Number of participants (N) 3408 1308 MESA location (n, %) a

Forsyth County, NC 645 18.9% 168 12.8% New York, NY 665 19.5% 145 11.1% Baltimore, MD 570 16.7% 88 6.7% St. Paul, MN 7 0.2% 764 58.4% Chicago, IL 814 23.9% 62 4.7% Los Angeles, CA 707 20.7% 81 6.2%

Region of residence (n, %) Northeast 660 19.4% 115 8.8% Midwest 797 23.4% 782 59.8% South 1236 36.3% 237 18.1% West 715 21.0% 80 6.1%

Age (mean, SD) 70.2 9.4 69.1 9.5 Female (n, %) 1809 53.1% 705 53.9% Race/ethnicity (n, %)

White 1274 37.4% 651 49.8% Chinese American 481 14.1% 60 4.6% Black 1072 31.5% 179 13.7% Hispanic 581 17.0% 418 32.0%

Family history of diabetes (n, %) 1529 44.9% 540 41.3% BMI (mean, SD) 28.2 5.6 29.2 5.8

<25 (n, %) 1040 30.5% 327 25.0% 25-29.9 (n, %) 1283 37.6% 470 35.9% ≥30 (n, %) 1085 31.8% 511 39.1%

Education level (n, %) HS/GED or less 1022 30.0% 473 36.2% Some college, Technical or Associate degree 945 27.7% 426 32.6% Bachelor’s degree or higher 1441 42.3% 401 30.7%

151

Included participants Excluded

individuals

N /

Mean Col % /

SD N /

Mean Col % /

SD Per capita income (in $10k) 2.6 1.9 2.2 1.6 Wealth index 2.6 1.2 2.6 1.2 Income/wealth index 5.0 2.2 4.9 2.1 Marital status (n, %)

Not married or living with partner 1395 40.9% 537 41.1% Married/Living with partner 2013 59.1% 771 58.9%

Smoking status (n, %) Never smoked 1575 46.2% 547 41.8% Former smoker 1585 46.5% 636 48.6% Current smoker 248 7.3% 125 9.6%

Physical activity, hours per day (mean, SD) 8.8 5.4 9.3 5.8 Daily calorie intake (mean, SD) 1643 782 1728 811 a This is the MESA location of the participant, not necessarily their area of residence

152

Supplemental Table 13 Analysis of interactions between SES and race with the price ratio for each of the three outcomes of interest a

Type 2 diabetes prevalence (n=3,408) OR Lower Upper p-

value p for

interaction Income/wealth

Lowest (1-4), n=1,264 0.759 0.612 0.942 0.0125 0.0647

Middle (5-6), n=1,011 1.227 0.895 1.683 0.2036

Highest (7-8), n=1,133 1.116 0.793 1.570 0.5292

Education level

HS or less, n=1,022 0.826 0.641 1.065 0.1409 0.8139

Some college, n=945 0.986 0.728 1.335 0.9255

BS or more, n=1,441 1.015 0.774 1.331 0.9142

Race

White, n=1,274 0.984 0.669 1.448 0.9348 0.7132

Chinese American, n=481 0.707 0.509 0.981 0.0381

Black, n=1,072 1.082 0.789 1.485 0.6248

Hispanic, n=581 0.847 0.633 1.134 0.2637

Type 2 diabetes incidence (n=2,829) RR Lower Upper p-

value p for

interaction Income/wealth

Lowest (1-4), n=977 1.097 0.722 1.665 0.6652 0.6736

Middle (5-6), n=861 1.180 0.712 1.958 0.5209

Highest (7-8), n=991 0.913 0.596 1.398 0.6755

Education level

HS or less, n=772 1.477 0.840 2.599 0.1758 0.5299

Some college, n=773 0.938 0.602 1.461 0.7757

BS or more, n=1284 1.120 0.727 1.726 0.6068

Race

White, n=1162 1.088 0.579 2.043 0.7940 0.6883

Chinese American, n=402 1.029 0.615 1.724 0.9126

Black, n=815 0.976 0.549 1.733 0.9329

Hispanic, n=450 1.401 0.806 2.437 0.2322

Insulin resistance (n=1,892) Estim

ate Lower Upper p-

value p for

interaction Income/wealth

Lowest (1-4), n=839 0.016 -0.064 0.097 0.6869 0.2544

Middle (5-6), n=686 0.071 -0.027 0.169 0.1552

Highest (7-8), n=828 0.064 -0.023 0.152 0.1483

Education level

153

HS or less, n=671 -0.068 -0.163 0.028 0.1635 0.1812

Some college, n=633 0.083 -0.012 0.178 0.0862

BS or more, n=1049 0.062 -0.014 0.138 0.1104

Race

White, n=931 0.036 -0.056 0.128 0.4454 0.6414

Chinese American, n=355 0.084 -0.032 0.201 0.1546

Black, n=679 0.006 -0.111 0.123 0.9216

Hispanic, n=388 0.012 -0.088 0.113 0.8108

OR, odds ratio; RR, risk ratio; HS, high school; BS, bachelor's degree. a Covariates include all covariates described for model 3 in the primary analysis (geographic region, age, gender, race/ethnicity, family history of diabetes, income/wealth index, education level, smoking status, physical activity, and neighborhood socioeconomic status).

154

APPENDIX 3.2 HOMA-IR validity

HOMA-IR was shown to be a valid measure of insulin resistance according to its

significant association with the insulin resistance index assessed by euglycemic-

hyperinsulinemic clamp (clamp IR) (Emoto et al., 1999). It was found that the log

transformed HOMA-IR measure had a stronger association than the non-transformed

value. Validation and reproducibility of the HOMA-IR measure were measured (again

using clamp IR as the gold-standard) in another study by Sarafidis et al (2007) (Sarafidis

et al., 2007). Correlation analysis demonstrated that HOMA-IR was strongly and

inversely correlated with the M-value (clamp IR index) (r =- 0.572, p<0.001), the

standardized M/I value (r=-0.768, p<0.0001) and the metabolic clearance rate (another

clamp-derived index) (r=-0.576, p<0.0001). The coefficient of variance (CV) for HOMA-

IR between visit 1 and visit 2 was 23.5% and the correlation between the two periods was

r=0.367 (p<0.05).

The thin-film adaptation of the glucose oxidase method on the Vitros analyzer

(Johnson & Johnson Clinical Diagnostics, Inc.) was used to measure fasting serum

glucose at each examination using. From each of the four examinations, 200 samples

were reanalyzed over a short period of time to recalibrate the original observations and

ensure consistency of the fasting glucose assay over the examinations. The RIA method

was used to determine fasting serum insulin levels were determined using the Linco

Human Insulin-Specific RIA Kit (Linco Research, Inc.). All assays were conducted at the

Collaborative Studies Clinical Laboratory at Fairview University Medical Center

(Abiemo et al., 2013, Nettleton et al., 2009a, Nettleton et al., 2008b).

155

VITA EDUCATION Doctor of Philosophy in Epidemiology, December 2016 Dornsife School of Public Health, Drexel University, Philadelphia, PA Master of Science in Biostatistics, May 2009 Mailman School of Public Health, Columbia University, New York, NY Bachelor of Science in Mathematics, June 2007 Drexel University, Philadelphia, PA RESEARCH EXPERIENCE Director of Life Sciences Research, HealthCore, Inc. 2016-Present Researcher; Sr. Researcher; Manager; Associate Director; HealthCore, Inc. 2010-2016 Associate Faulty of Biostatistics, Albert Einstein College of Medicine 2009-2010 Data analyst, Research Foundation for Mental Hygiene, 2008-2009 TEACHING EXPERIENCE Drexel University Dornsife School of Public Health Teaching Assistant, 2014-2015: PBHL 630: Intermediate Epidemiology; PBHL 621: Intermediate Biostatistics II Honors and awards: Teaching assistant excellence award nomination SELECT PEER-REVIEWED PUBLICATIONS Kern DM, Auchincloss AH, Robinson LF, Ballester LL. Neighborhood variation in the price of soda relative to milk and its association with neighborhood socioeconomic status and race. Public Health Nutr. Jun 2016; 30:1-11 Wu B, Bell K, Stanford A, Kern DM, Tunceli O, Vupputuri S, Kalsekar I, Willey V. Understanding CKD among patients with T2DM: prevalence, temporal trends, and treatment patterns – NHANES 2007-2012. BMJ Open Diab Res Care. Apr 2016; 11(4):e000154 Kern DM, Mellstrom C, Hunt PR, Tunceli O, Wu B, Westergaard M, Hammar N. Long-term cardiovascular risk and costs for myocardial infarction survivors in a US commercially insured population. Curr Med Res Opin. Apr 2016; 32(4):703-11. Kern DM, Davis J, Williams SA, Tunceli O, Wu B, Graham E, Strange C, Trudo F. Validation of an administrative claims-based diagnostic code for pneumonia in a commercially insured COPD population. Int J Chron Obstruct Pulmon Dis. Jul 2015; 10(1):1417-1425. Kern DM, Davis J, Williams SA, Tunceli O, Wu B, Hollis S, Strange C, Trudo F. Comparative Effectiveness of Budesonide/Formoterol Combination and Fluticasone/Salmeterol Combination among Chronic Obstructive Pulmonary Disease Patients New to Controller Treatment: A US Administrative Claims Database Study. Repir Res. Apr 2015; 16(1):52 Kern DM, Zhou S, Chavoshi S, Tunceli O, Sostek M, Singer J, LoCasale R. Treatment patterns, healthcare resource utilization and costs among chronic opioid users for non-cancer pain using a US administrative claims database. Am J Manag Care. Mar 2015;21(3):e222-e234 Kern DM, DeVore S, Kim J, Tunceli O, Wu B, Hirshberg B. Real world mortality, cardiovascular outcomes, and healthcare costs in type 2 diabetic patients at high risk for cardiovascular disease. Am J Accountable Care. Mar 2015; 3.15:55-61. Kern DM, Balu S, Tunceli O, Anzalone D. Statin Treatment Patterns and Clinical Profile of Patients with Risk Factors for Coronary Heart Disease (CHD) Defined by National Cholesterol Education Program (NCEP) Adult Treatment Panel (ATP) III. Curr Med Res Opin. Dec 2014; 30(12):2443-2551