SPH Theses and Dissertations

Breast cancer is a leading cause of cancer and the second leading cause of cancer death among women in the US. Although many risk factors for breast cancer are known, few are modifiable and little is known about ways to prevent its incidence. Early-life body size is inversely associated with both premenopausal and postmenopausal breast cancer risk, suggesting an excess risk in lean girls. In a prospective analysis within the Nurses’ Health Study (NHS) II, Chapter 1 examines whether adolescent physical activity mitigates the excess risk of breast cancer associated with early-life body leanness. Lean girls were at higher risk of breast cancer, regardless of the level of adolescent physical activity; however, the association was slightly, though not significantly, attenuated among the most active girls. Breast cancer is hormone-related cancer; estrogen metabolites (EM) are both estrogenic and genotoxic, suggesting factors that alter the pattern of estrogen metabolism may contribute to breast carcinogenesis. With the application of advanced technology that measures 15 different individual estrogens and EM in urine, Chapter 2 examines the associations of dietary fiber and macronutrients intake with detailed estrogen metabolism in a cross-sectional analysis within the NHSII. Few significant associations were identified: a positive association between total fiber intake and 4-methoxyestradiol, an inverse association between total fiber intake and 17-epiestriol, and inverse associations for polyunsaturated and trans-fat intakes with 17-epiestriol. The tissue-specific responsiveness to potentially carcinogenic hormones, estrogen and progesterone, is partially regulated by the tissue expression of receptors that bind these hormones. Using benign breast biopsy samples collected in a nested case-control study within the NHS and NHSII, Chapter 3 assesses estrogen receptor (ER), progesterone receptor (PR), and proliferative marker Ki67 expression in normal breast tissue in relation to subsequent breast cancer risk. In this case-control analysis, PR expression in normal breast tissue was significantly positively associated with breast cancer risk in premenopausal women. ER and Ki67 expression was not significantly associated with breast cancer risk; however, our power was limited.
Results of this dissertation help elucidate the underlying biologic mechanisms of breast cancer and enhance our understanding of the link between risk factors and breast cancer risk.

The last decade witnessed the important scaled-up of malaria control interventions in sub-Saharan Africa (SSA). There is now renewed impetus to achieve the long-term goal of malaria elimination and reducing vectorial capacity of the Anopheles mosquito is a necessary first step towards this objective. Relying solely on the two pillars of malaria vector control (i.e., insecticide-treated nets and indoor residual spraying) will be insufficient to achieve elimination in much of SSA, however. Larval Source Management, and larviciding in particular, could play an important role in areas where breeding habitats are ‘few, fixed, and findable’ or where malaria vectors exhibit exophagic and exophilic behaviors, and in settings where insecticide resistance has emerged. Yet, only few contemporary studies have investigated the effectiveness of larviciding for malaria control despite historical success. Using the wealth of data from Dar es Salaam’s Urban Malaria Control Program (2004-2008), this dissertation will first assess the impact of a community-based larviciding program on prevalence of malaria infection in 15 urban wards of Dar es Salaam (Tanzania). The cost-effectiveness of this intervention will then be estimated from both a provider and a societal perspective. Finally, in a context of accelerated malaria control, the effect of reducing malaria transmission on disease-related behavior and knowledge will be examined.

Results suggest that the larviciding intervention had a significant protective effect, decreasing by 21% the odds of being infected with malaria. Larviciding was found to be cost-effective for incidences as low as 40 infections per 1,000 individuals per year but the cost-effectiveness ratios were highly dependent on the assumed baseline malaria incidence rates. Such a successful intervention could also bring about further challenges to sustaining gains in reducing malaria transmission as the larviciding intervention was found to negatively affect bednet usage and knowledge of disease symptoms. Collectively, these results imply that larviciding should be considered as part of an Integrated Vector Management approach in SSA, if local eco-epidemiological conditions are suitable, and that there is a need to sustain behavioral change communication following successful vector control interventions.

Major cardiovascular risk factors have changed over the past 3-4 decades throughout the world. While adiposity and diabetes are rising worldwide, blood pressure and cholesterol are declining in high-income and even some middle-income countries, possibly due to improvements in diet or better diagnosis and treatment; the same risk factors have remained unchanged or even increased in low-income countries. To formulate effective prevention and health system policies, there is need to understand the implications of these diverse trends for cardiovascular diseases (CVDs).

This dissertation focuses on quantifying the impact of multiple risk factors and preventive interventions on CVDs and their disparities at the population level. Answering this question requires information on how much of the effects of adiposity on CVDs are mediated through other metabolic risk factors (i.e. high blood pressure, high serum cholesterol and high blood glucose), which themselves have other determinants. The first paper quantifies the direct as well as the mediated effects of excess weight on coronary heart disease (CHD) and stroke through blood pressure, serum cholesterol and blood glucose. The analyses use data of 97 prospective cohorts with more than 1.8 million participants. This allows for assessing whether the extent of mediation is modified by geographical region, study period, and other characteristics of study populations. The second paper revisits the above question using causal inference models and further quantifies the role of inflammatory markers as potential mediators. The analyses use individual-level data from 9 prospective cohort studies that have high-quality measurements of metabolic and inflammatory biomarkers. The third paper uses national data sources in the United States (US) and estimates the distributions of 10-year risk of fatal CHD by race. It also assesses the effects of different population-wide and targeted interventions on CHD risk distributions and their disparities between blacks and whites.

Our findings suggest that nearly half of excess risk for CHD and three-quarters of excess risk for stroke due to excess weight were mediated through three metabolic risk factors: blood pressure, cholesterol, and glucose. Inflammatory biomarkers had much smaller roles than the combination of metabolic risk factors. In the US, the distribution of 10-year CHD risk was shifted to the right among blacks compared to whites and had a heavier tail, leading to a substantially larger proportion of blacks in the high-risk group. A risk-based intervention that identifies and treats these individuals could substantially reduce both the overall risk of CHD and its racial disparities. These results together provide the quantitative evidence on the impact of cardiovascular risk factors and selected interventions on CVDs and their disparities.

Cardiovascular disease (CVD) is the number one cause for mortality and morbidity around the world. Meanwhile, diabetes mellitus (DM) has become an emerging epidemic, causing 1.5 million deaths in 2012, with 80% occurring in low- and middle-income countries. Substantial evidence has linked both lifestyle and metabolic risk factors to increased risk of CVD and death, with suboptimal diet being the single leading modifiable cause of poor health (Lim. SS, et al, Lancet 2012). Of 20 top individual causes of disease burden worldwide, 8 are related to poor nutrition, including suboptimal intakes of various dietary fatty acids. Although previous studies have found divergent health effects of different dietary fatty acids on health, gaps still exist in terms of the scientific knowledge (e.g. how individual circulating vs. dietary trans fatty acid subtypes affect health) and related disease burdens (e.g. national CHD mortality burdens attributable to suboptimal intakes of fatty acids). These gaps have motivated my dissertation researches. In chapter one and two, I investigated the prospective associations of five subtypes of plasma phospholipid trans fatty acid (TFA) levels with the risk of various disease endpoints, including total, CVD and non-CVD mortality, incident coronary heart disease (CHD) and DM. In chapter two, I also examined the prospective associations of total and subclasses dietary TFA with risk of DM. The analyses were conducted using the Cardiovascular Health Study, a community-based multicenter prospective cohort consisted of older Americans. The risks were estimated using the Cox proportional hazard model.
The study in chapter three was a collaborative effort of the Nutrition and Chronic Diseases Expert Group as part of the 2010 Global Burden of Diseases, Injuries, and Risk Factors study. Using the comparative risk assessment framework, I comprehensively quantified the CHD mortality attributable to suboptimal intakes of saturated fat, omega-6 polyunsaturated fat, and TFA in 186 countries in 1990 and 2010, by age and sex groups. I also estimated the regional and country level trends of these attributable CHD burdens from 1990 to 2010. The findings of this study are relevant for informing regional and country level public health policy priorities.

The effects of temperature and air pollution on public health are comprehensive and ubiquitous. Therefore, this dissertation deals with the comprehensive topic of climate change and air pollution and their effects on public health. The first chapter examines the effect of temperature on mortality in 148 cities in the U.S. from 1973 through 2006. We focused on the timing of exposure to unseasonal temperature and temporal and spatial acclimation. The second chapter incorporated AOD data from satellite imagery with other predictors such as meteorological variables, land use regression, and spatial smoothing to predict the daily concentration of PM2.5 at a 1 km2 resolution across the southeastern United States, covering the seven states of Georgia, North Carolina, South Carolina, Alabama, Tennessee, Mississippi, and Florida for the years from 2003 through 2011. As the sequel of the result from the second chapter, the last chapter investigated the acute effect of PM2.5 on mortality in the entire population of North Carolina, South Carolina, and Georgia between 2007 and 2011 using the predictions from the second topic as PM2.5 exposure.

Observational studies of newly marketed medications can add important safety and effectiveness information to what is known from pre-approval randomized controlled trials. As new medications are often prescribed to select group of patients, confounding can be particularly large in this setting. Multivariable modeling of the exposure or the outcome is commonly used to reduce confounding, but the modeling can be challenged by rapidly evolving prescribing patterns and the limited number of patients exposed to the new medication. Furthermore, while adjustment for empirically identified potential confounders, and proxies thereof, has been shown to reduce confounding, it is difficult to empirically identify potential confounders in settings with small number of outcomes. In this thesis, we explored two new methods to improve confounding control in the setting of newly marketed medications with few exposures and outcomes and many potential confounders. The first method is the high dimensional disease risk score (DRS) developed using an historical cohort of comparator drug initiators. We developed prediction models for the outcomes of interest in an historical cohort and applied the models to the concurrent study cohort of new and comparator drug initiators. We then used individual patient’s predicted risk score to balance the baseline risk between the two groups. In chapter 1, we compared combinations of shrinkage and dimension reduction approaches to reduce model over-fitting when developing DRSs from large numbers of potential covariates. In chapter 2, we compared the performance of the high dimensional DRSs to the standard high dimensional propensity score (hdPS) approach. In chapter 3, we developed a new method that augments the hdPS variable selection process with data from an historical cohort and compared this approach to standard hdPS. All evaluations were conducted in example comparative studies of newly marketed medications using large US claims database. The hdPS with variable selection augmented by historical data showed good performance in confounding adjustment even in small outcome settings. Future studies should evaluate the use of this method in other settings and should explore improvements in the use of high dimensional DRSs.

This dissertation examines the relationship between life course social mobility and socio-economic disparities in health using data from the Health and Retirement Study. Chapter 1 explores how eligibility for the Korean War GI Bill, which provided socio-economic benefits to veterans, impacted the mental health of veterans compared to non-veterans. Using coarsened exact matching (CEM) to rigorously control for selection into the military, Korean War era veterans were matched to non-veterans. Regression models were used to predict number of depressive symptoms (NDS) and an indicator for depression. Pooled results were null, however veterans from low socio-economic backgrounds benefited from military service while veterans from high socio-economic backgrounds did not, resulting in a reduction in socio-economic disparities in NDS and depression prevalence. Chapter 2 examines if the benefits from Korean War era military service had a spillover effect to wives of veterans. Using CEM, veteran wives were matched to wives of non-veterans. Pooled results were null, however, among women from high socio-economic backgrounds, veteran wives had fewer NDS than non-veteran wives; there was no effect of husband’s military service among wives from low socio-economic backgrounds.
Chapter 3 examines the relationship between life course social mobility and markers for stress (C-reactive protein), blood sugar (hemoglobin A1c), cholesterol (high density lipoprotein), and kidney function (Cystatin C). Results from linear regression models indicate that all outcomes followed a social gradient whereby individuals who experienced high SES at both time points had the best biomarker levels, while individuals who experienced low SES at both time points had the worst biomarker levels. Individuals who experienced upward mobility over the life course had statistically equivalent biomarker levels as individuals who experienced high SES at both time points, while individuals who experienced downward mobility had statistically equivalent biomarker levels as individuals who experienced low SES at both time points. These results suggest that the deleterious effects of a low SES childhood may be ameliorated or eliminated by upward social mobility in adulthood for these outcomes.

This dissertation explores the role of biologic factors and a novel method of measuring cognitive function in investigating mechanisms and risk factors of cognitive aging. With a rapidly increasing aging population, the public health burden of dementia is expected to rise in the future. Therefore, it is important to both better understand the etiology and identify novel risk factors, in order to reduce the incidence of new cases and develop effective treatments. Blood-based biomarkers, such as amyloid-beta and sex hormones, can provide an objective measure of factors likely associated with dementia risk. Additionally, computerized cognitive testing can provide efficient and accurate measurement of cognitive function, yet is seldom used in epidemiologic studies.

Our first chapter involves the first systematic review and meta-analysis of prospective studies exploring the association between plasma amyloid-beta and incident Alzheimer’s disease or dementia. While preclinical prediction of Alzheimer’s disease is important for effective intervention, studies have considerably varied in design, assays and sample size, making it difficult to interpret the overall data and thus necessitating a systematic review and meta-analysis.

The second chapter explores the association between endogenous hormone levels and cognitive function in a population of older women. While many studies have investigated the role of hormone therapy in cognitive aging, fewer studies have investigated endogenous hormones, which may provide a more objective measure of hormonal status. We therefore present a prospective cohort study investigating the association between several endogenous sex hormones and their prohormones at baseline, and cognitive function over 20 years later.

The third chapter evaluates the feasibility and performance of a self-administered computerized cognitive battery. Although most epidemiologic studies of cognitive aging rely on traditional neuropsychological testing to measure cognitive outcomes, such methods can be prone to error, interviewer bias, and demand substantial time and cost. Therefore, we present results demonstrating for the first time the feasibility and performance of an unsupervised self-administered computerized cognitive battery in a population of older men.

Work and family stress is increasingly pervasive in today’s workforce, and studies reveal that it may be related to worse employee cardiovascular health. However, it is unclear whether the combined stress from work and home relate to individual cardiometabolic risk factors over time and whether the workplace influences these risk factors. Further, no research has examined if these demands are associated with higher levels of inflammation in the body, which are meaningful indicators of cardiovascular disease (CVD). This dissertation seeks to address these gaps in the literature. We draw upon data from 1,524 predominantly female, ethnically and racially diverse extended-care employees who provide biological as well as self-reported data on a variety of sociodemographic, health and work and family variables at four study waves (baseline, 6 months, 12 months and 18 months). Specifically, Chapter 1 examines the observational effects of work and family conflict, conceptualized to represent perceived stress, on five cardiovascular risk factors (blood pressure, glycosylated hemoglobin, cholesterol, body mass index and cigarette consumption) used to establish a cardiometabolic risk score (CRS) over the 18 month study period. Chapter 2 investigates the effects of a workplace intervention designed to increase flexibility, schedule control and workplace support on employee markers of inflammation from baseline to 12 months as part of a prospective, randomized field experiment. Chapter 3 assesses manager- and worksite-level influences on the aforementioned CRS and behavioral and biological variables representing CVD risk, all measured at the employee-level at baseline. We employ multilevel level modeling techniques to account for the nesting of employees within manager workgroups and worksites as well as multiple measures per employee where appropriate. Overall, we find that work and family conflict may relate to certain measures of cardiometabolic risk, such as BMI and cholesterol. Additionally, employees who work within the same worksite may have more similar cholesterol levels than employees working at different worksites. We do not find evidence that this particular workplace intervention lead to changes in employee levels of inflammation over time. Given the public health burden of CVD, we recommend that future research continue to examine the effects of work and family stressors on cardiovascular outcomes in a variety of settings.

In 2013, approximately 289,000 women died from pregnancy-related causes and 2.8 million newborns died within the first 28 days of life. The vast majority of these deaths occur in resource-limited settings. This work examines risk and protective factors for the development of several perinatal complications that put mothers and their infants at risk for adverse health outcomes. We explored determinants of preeclampsia and gestational hypertension among women in Dar es Salaam, Tanzania. We also examined the effects of pregnancy spacing intervals on perinatal outcomes in that group of women. We used log binomial regression to obtain risk ratios and 95% confidence intervals for the development of the adverse pregnancy outcomes of interest. We also looked at the efficacy of an intervention aimed at improving neonatal outcomes by conducting a systematic review and meta-analysis of the effects of kangaroo mother care on neonatal morbidity and mortality.

        We found that nulliparity, history of hypertension, urinary tract infection, low calcium intake, history of preeclampsia, and history of preterm birth were associated with an increased risk of developing preeclampsia among women in Dar es Salaam. Risk factors for gestational hypertension included a history of diabetes, elevated blood pressure at study enrollment, increased mid-upper arm circumference, high hematocrit, low mean corpuscular volume, a history of miscarriage or stillbirth, and older age at first pregnancy. Twin gestation and increased body mass index were risk factors for both types of hypertensive disorders of pregnancy among women in Tanzania. After a live birth, inter-pregnancy intervals less than six months were associated with an increased the risk of having a low birth weight baby in the next pregnancy; while after a stillbirth, short inter-pregnancy intervals were associated with increased risk of stillbirth and perinatal death. Providing kangaroo mother care to infants after birth was associated with decreased neonatal morbidity and mortality and increased likelihood of exclusive breastfeeding when compared to conventional care. 

        These findings can help identify women and infants at increased risk for developing pregnancy-related complications and contribute to informing development of evidence-based maternal, newborn, and family planning programs and policies.

Current prevention recommendations for ovarian cancer are limited, and the underlying etiology is not fully elucidated. The associations of common modifiable factors, such as psychosocial and lifestyle factors, with ovarian cancer risk need to be more fully evaluated. Thus, I examined the association of ovarian cancer with depression, physical activity, hypertension, and antihypertensive medication use among participants from two prospective cohort studies: the Nurses' Health Study and Nurses' Health Study II. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for these associations. Depression was associated with about 30% increased risk of ovarian cancer (HR = 1.30, 95% CI 1.05-1.60). Higher risk was also observed among women with persistent positive depression status than women intermittently positive or persistently negative for depression. Contrary to the hypothesis that physical activity may lower ovarian cancer risk, we observed that both low and high physical activity was associated with a modest increase in ovarian cancer risk (HR for ≥27 [approximately equivalent to 1 hr/day of brisk walking] versus 3-9 MET-hrs/week = 1.26, 95% CI 1.02, 1.55; HR for <3 versus 3-9 MET-hrs/week = 1.19, 95% CI 0.94, 1.52). However, these associations were only restricted to premenopausal physical activity, and postmenopausal activity was not associated with ovarian cancer risk. While hypertension was not associated with risk (HR = 1.03, 95% CI 0.87, 1.21), use of thiazide diuretics was associated with an increased risk of ovarian cancer (HR = 1.35, 95% CI 1.04, 1.74), and use of calcium channel blockers was associated with a suggestively lower risk (HR = 0.73, 95% CI 0.53, 1.01). Our results need to be confirmed by other studies, but suggest that these common modifiable factors may have a moderate impact on ovarian cancer risk. This represents an opportunity to broaden our understanding of ovarian cancer etiology and potentially improve prevention strategies for ovarian cancer.

In this dissertation, we first assess the associations between circulating metabolites and body mass index in two U.S. prospective cohorts. Then, we propose several methods to analyze multiple phenotypes in genetic association studies based on summary statistics.

In the Chapter 1, we investigate the associations between circulating metabolites and repeatedly measured body mass index (BMI) among Caucasian men and women. We employed linear mixed models with a random intercept term to account for the within-subject correlation and adjust for potential baseline confounders including baseline obesity status, smoking, physical activity, alcohol intake, total caloric intake, age at blood draw and follow-up years. We didn't find the associations were heterogeneous between Caucasian men and women, and therefore we used fixed effect meta-analysis to combine evidences of associations from these two cohorts. We found that valine, leucine and isoleucine were positively associated with BMI, while acetylglycine was negatively associated with BMI.

In Chapter 2, we present linear mixed model based score tests for the detection of pleotropic genetic variants that are associated with multiple correlated traits based on summary statistics. Our tests are robust to effect heterogeneity and correlation structures among multiple traits. We conducted simulation studies to compare the proposed methods with existing methods. We also applied our methods to a global lipids GWAS summary statistics data set and identified hundreds of novel genetic variants.

In Chapter 3, we propose a geometric perspective on the powers of principal component association tests based on GWAS summary statistics. Utilizing eigen-analysis of the correlation matrix and asymptotic power analysis, we investigate when PCA is powerful and when it is not to detect the genetic signals. We further apply our methods to a global lipids level genome-wide association study data set and identify hundreds of novel genetic variants that were missed by conventional single-trait analysis approaches. Our results can help guide researchers to choose powerful PCAT methods in multiple phenotype association studies and also better interpret the association results.

Statement of Problem: Despite significant advancements in occupational health and safety in recent decades, injury rates in commercial construction remain high. New programs that address the complexity of the construction work environment are needed to keep workers healthy and safe.

Methods: The first step of this dissertation was to explore associations between organizational programs and policies, as measured by a Contractor Safety Assessment Program (CSAP) score, and worker safety climate scores. Next, a safety communication and recognition program was developed and piloted. It was evaluated through a mixed methods approach in a randomized controlled trial. Primary outcome measures included safety climate, awareness, communication, and teambuilding. Additionally, the dynamic nature of the construction site was quantified through an analysis of the determinants of length of stay of construction workers on the worksite.

Results: Correlations between CSAP scores and safety climate scores were weak at best, thus highlighting a gap in communication between management and workers. The B-SAFE program, a safety communication and recognition program was developed to meet this gap. It used data from safety inspection scores to provide feedback to workers on hazards and controls, and provided a reward when the site met a pre-determined safety inspection threshold (a measure that was fair, consistent, attainable and fair). In the final program design, the whole site was treated as the unit of analysis. B-SAFE led to many positive changes, including a statistically significant increase in safety climate scores of 2.29 points (p-value=0.012), when adjusting for time-varying parameters and worker characteristics. Workers at the B-SAFE sites noted increased levels of safety awareness, communication, and teamwork, when compared to control sites. The composition of workers on-site at any given month changed by approximately 50%, and the length of stay on-site was associated with race/ethnicity, union status, title, trade, and musculoskeletal pain (p-values<0.05).

Conclusions: The construction work environment is dynamic, with over half of the population on-site changing each month. This makes applying and evaluating traditional worksite based interventions challenging. Interventions like B-SAFE that are developed to address the complexities can have a positive impact on site safety measures.

The objective of this dissertation is to contribute to ongoing research on prenatal metals exposure, in terms of arsenic, lead, and manganese, and infant health and growth, and to deepen the understanding of the complexity of such problems. We seek to do so in three parts. First we examine the association between combined prenatal metals exposure and infant birth weight and head circumference. Then, we look at the effect on birth weight from the HFE gene variants and its interaction effects with arsenic. Finally, we look at the association of prenatal metals exposure and child growth up to 36 months. The study populations of all three of our studies are taken from mothers enrolled in 2 hospitals affiliated with Dhaka Community Hospital in Bangladesh. They were given self-administered questionnaires at time of enrollment and are followed after birth. Child measurements were taken at time of birth, and the biomarker for these studies are cord blood metal measurements. In the first part of this dissertation, through multivariate linear regression, including a metal interaction term, we found that prenatal arsenic and manganese exposure individually associated with lowered birth weight and birth head circumference. We also found evidence of interactions between the two metals, suggesting that joint exposure creates greater deficit in birth outcomes. In the next part, looking at gene-environment interactions, we found significant modification effects of multiple SNPs on the HFE gene that increased the association between arsenic and birth weight. We also found direct effect of less studied HFE genes to lower birth weight. Finally, we assessed the effect of prenatal metals exposure on early growth in children through longitudinal analysis. In following the weight and height of the child from birth up to 36 months of age, our results indicated adverse association between arsenic and manganese on growth.

Background: Heart Rate Variability (HRV) as a research outcome has the potential for misclassification due to its inability to account for changes in the heart rate. HRV can be parsed into Acceleration Capacity (AC) and Deceleration Capacity (DC) which address these limitations. Objectives: To investigate the associations between AC and DC with short-term and long-term metal PM2.5 exposures; and examine if these associations are mediated by inflammation. Methods: A panel of 45-50 male welders, mean age 39-40 years, had continuous PM2.5 exposure during typical welding work shifts for 4-6 hours repeated 2-5 times over sampling periods in 2010-2012. We also obtained continuous recordings of digital electrocardiograms (ECG) over their work shift using Holter monitors during the same time; and analyzed blood samples before and after each welding shift for potential mediators of inflammation. In our first analysis, we used linear mixed models to assess the association between hourly PM2.5 exposure and each of simultaneously measured hourly AC and DC, controlling for covariates. Then, mediation analysis was done using linear mixed models to assess the associations between shift PM2.5 exposure, potential mediator levels, and AC and DC, controlling for relevant covariates in order to deduce the direct and indirect effects (via the mediator) of PM2.5 on AC and DC. In our final analysis, we used linear regression to assess the association between CEI PM2.5 exposure and each of current AC and DC, controlling for confounders. Results: Negative exposure-response associations were found for AC and DC with increased PM2.5 both in the short-term and long-term exposure after adjusting for covariates. In our mediation models, the proportion of the total effect of PM2.5 on AC or DC (indirect effect) mediated through IL-6 on AC was at best 4%. Conclusions: There are sustained acute and chronic effects of metal particulates on AC and DC even after exposure has ceased. These findings suggest that there may be more pathways that sustain response following exposure other than a direct effect of metal particulates on AC and DC. Furthermore, there may be complex mediating pathways involving interleukin 6.

Introduction: Smoking is the leading cause of preventable death in the United States, and obesity is the second cause. Both are risk factors for various types of cancer. Specifically, smoking is linked to lung, esophageal, bladder, kidney, and stomach cancers; and obesity is linked to postmenopausal breast, colorectal, endometrial, pancreatic, and gallbladder cancers. Multiple factors affect health behavior, and this dissertation investigated the role of media and health communication as modifiable influences related to smoking and obesity.

Methods: Paper 1 examined the association between average hours of television viewing per day and smoking status among all Hispanic adults (n=675) and within the subgroups of Puerto Rican (n=182) and Dominican (n=396) adults using multivariable logistic regression models. Paper 2 explored how health information sources inform decision-making related to childhood obesity among Hispanic mothers during their children’s first 1000 days of life (conception-age 24 months) using seven focus groups. Paper 3 documented the development and uptake of a media competition (with 595 student participants) implemented in the context of a multi-sector community intervention targeting childhood obesity prevention through process evaluation. Paper 3 also examined community, organizational and provider characteristics that explain variation in implementation effectiveness and described diffusion of the media competition across community sectors using 54 key informant interviews.

Results: In Paper 1, Hispanic adults who watched 5+ hours (5-15 hours) of TV per day were more likely to be a smoker than those who watched ≤2 hours, with the same association among Puerto Ricans. In Paper 2, trusted health information sources for Hispanic mothers included health care providers, female and male family members, BabyCenter.com and other Internet sources, selected social media, and television. In Paper 3, salient themes that emerged as implementation facilitators were having a cascade of champions and adaptability through providing opportunity to participate in the media competition outside traditional class time.

Discussion: Papers 1 and 2 support the importance of understanding the nuances and differences in Hispanic adults in order to design appropriate media and health communication interventions. Paper 3 provides insight about how to replicate media competitions for children in other communities.

Type 2 diabetes has reached epidemic proportions globally, and accumulating evidence suggests extremes of sleep duration increase risk. Diet may be an important mechanism, yet few studies examine prospective relationships of sleep duration and diet quality or whether diet explains associations of sleep duration with childhood obesity or diabetes in adults. In Chapter One, we report a moderate correlation between self-reported sleep duration and actigraphy in Sueño, the sleep ancillary study to the Hispanic Community Health Study/Study of Latinos. Chapter Two identifies associations of chronic insufficient sleep duration since infancy with lower diet quality in mid-childhood in Project Viva: children with the least favorable diet and sleep have the highest body mass index z-scores in mid-childhood, but diet does not explain associations with adiposity. In Chapter Three, adherence to healthful dietary patterns reduces risk of diabetes in the Women’s Health Initiative; high quality diets are protective in all groups, but race/ethnicity modifies associations. In Chapter Four, we find that changes in sleep duration, increases in particular , are associated with diabetes and concomitant changes in diet quality, physical activity and weight in the Nurses’ Health Study. Each of these studies contributes new knowledge: Sueño represents the largest sleep validation to date, the only validation among Hispanic/Latinos and allows researchers to better understand the information contained in (and the limitations of) self-reported measures of sleep duration within subgroups. In the Women’s Health Initiative, we address limitations of the current literature on dietary patterns by calculating four dietary indices within the same cohort, standardizing the scores for comparison and examining associations across racial/ethnic groups. Project Viva is the first study to examine the influence of chronic insufficient sleep on diet quality in childhood when health behaviors and dietary preferences are being formed. Finally, examining changes in sleep duration and changes in diet quality, physical activity and weight in the Nurses’ Health Study represents a novel way to leverage repeated assessments. Results of this dissertation may help build the case for policy and intervention efforts to prevent and treat obesity and diabetes, particularly those that seek to improve both sleep and diet.

Measuring health through surveys is challenging. Participants may respond in a socially favorable but untruthful way, and responses across respondents may be difficult to compare. List experiments and anchoring vignette techniques have been proposed to improve survey measures, and this dissertation applied qualitative and quantitative methods to evaluate either the usefulness or validity of those techniques. The first paper explored how list-based questions perform compared to direct questions when measuring two behaviors, smoking and intravenous infusion use. The difference-in-differences between the two survey methods and two behaviors was non-significant. List experiments might introduce downward biases rather than alleviate them due to cognitive difficulty in responding. The second paper applied cognitive interviewing to the design of vignettes among Chinese students with objectively different visual acuities. Ten major problems were identified regarding vignette comprehension, judgment, and responses. Respondents rated vignette character’s vision differently from their own, demonstrating a norm of being “strict with oneself and lenient towards others” in an Asian context.
The third paper assessed the validity of three vignette survey methods (i.e., indirect comparison between self and vignettes, direct comparison between self and vignettes, and self-assessment primed by vignettes) in measuring distance vision. Surprisingly, it was found that vignette methods were not improvements over self-assessment, and indirect comparison performed the worst.
These three studies shed light on a series of cognitive difficulties related to advanced survey techniques. Traditional self-assessment may be as useful as the list experiment and more valid than some anchoring vignette techniques in these studies’ context.

Breast cancer is the most frequently diagnosed cancer and the second leading cause of cancer death among US women. Several hormonal, anthropometric, lifestyle, and genetic factors are known to be associated with breast cancer, though these associations may differ by menopausal status and molecular subtype. Studying risk factors in relation to these subtypes can help enhance our understanding of breast cancer etiology. Here, we aim to further explore the mechanisms through which several established and suspected risk factors may influence risk of breast cancer, with an emphasis on modifiable exposures, which may have direct implications for prevention strategies, particularly for premenopausal and non-luminal breast cancers. In Chapter 1, we evaluated the cross-sectional relationship between intakes of caffeine, coffee, decaffeinated coffee, and tea, and comprehensive profiles of urinary estrogens and estrogen metabolites in premenopausal women. We found significant associations between coffee intake and metabolites in the 2-hydroxylation pathway, suggesting a possible mechanism through which coffee may affect breast cancer risk. In Chapter 2, we examined associations between premenopausal plasma carotenoid levels and markers of oxidative stress and subsequent breast cancer risk. In contrast to previously published analyses, which largely focused on postmenopausal carotenoid exposure, we did not find significant inverse associations between circulating carotenoids and risk; additionally we did not observe positive associations between fluorescent oxidation products and risk. However, we did find some evidence that the effects of carotenoids on risk may be modified by single nucleotide polymorphisms in genes related to carotenoid availability and oxidative/antioxidative processes. In Chapter 3, we explored whether the associations of reproductive risk factors with breast cancer vary by intrinsic molecular subtype. We observed evidence that many risk factors are most strongly associated with the hormone receptor-positive luminal A subtype, which comprises the majority of breast cancers, though tests of heterogeneity did not reach significance in many cases. Consistent with previous studies, we observed that breastfeeding may reduce risk of basal-like tumors, and may represent a potential preventive strategy for this aggressive subtype. In conclusion, these analyses, while varied in scope, help elucidate mechanisms by which risk factors may influence breast cancer risk.

Nutritional factors have been intensively studied as important determinants of many diseases. Food frequency questionnaires (FFQ), dietary records, 24-hour dietary recalls, nutrient biomarkers are important dietary assessment methods, and are subject to various sources of measurement error. Given the limitations of these methods, much effort has been devoted to refining them and evaluating their ability to measure diet. This dissertation focused on evaluating the performance of a semi-quantitative FFQ (SFFQ), multiple web-based automated-self-administered 24-hour recalls (ASA24), 7-day dietary records (7DDR) and biochemical indicators in assessing nutrient intakes among women. Intraclass correlation coefficient, Spearman correlation coefficient, and validity coefficient calculated by method of triads were used to evaluate the reproducibility and validity of each dietary method.

The first paper evaluated the performance of a 152-item SFFQ comparing intakes of nutrients estimated by SFFQ with those measured by the average of two 7DDR, and of four ASA24s kept over a one-year period. The study SFFQ performed consistently well when compared with multiple diet records, and that modifications to the questionnaire over time have adequately taken into account the changes in the food supply and eating patterns that have occurred since 1980. Multiple ASA24s can provide similar estimates of validity as dietary records if day-to-day variation is taken into account.

The second paper explored the validity of long-term intakes of energy, protein, sodium and potassium assessed by SFFQ and ASA24s using recovery biomarkers and 7DDR as standards. The study SFFQ and averaged ASA24’s are reasonably valid measurements for energy-adjusted protein, sodium and potassium compared to multiple recovery biomarkers or dietary records. Recovery biomarkers should not be considered to be without error, including systematic within-person error.

Finally, the third paper further evaluated the validity of nutrient assessed by SFFQ and ASA24 compared with intakes by the 7DDR and plasma levels of fatty acids, carotenoids, retinol, tocopherols and folate. Again, the study SFFQ provides reasonably valid measurements for specific fatty acid, most carotenoids, alpha-tocopherol and folate compared to concentration biomarkers or dietary records. Compared to SFFQ, almost all nutrients estimated by averaged ASA24s had relatively low correlations with biomarkers, 7DDRs and estimated ‘true’ underlying intakes.