Contexts are important social determinants of individual health trajectories and population level patterns of health disparities. This dissertation examines three types of contexts—social networks, physical environments, and social positions—using innovative quantitative approaches. Chapter 1 examines the intersectional social positions created by interlocking social identities—race/ethnicity, sex, income, education, and age—and their relationship to health disparities in the obesity epidemic. We outline an innovative analytic approach to evaluating intersectionality using multilevel models. After adjustment for the contributions of the main effects, a large intersectional effect remains. While clear social patterning emerges, interactions are not necessarily  patterned  according  to  ‘multiple  jeopardy’   and  ‘multiplicative  benefit’  as  might  have  been  expected. These findings reveal the complex social patterning of the obesity epidemic, and challenge us to consider possible refinements to intersectionality theory. Chapter 2 evaluates whether U.S. adolescent social networks are segregated by family income level. Network segregation or integration may affect adolescent health trajectories through a variety of pathways, yet the extent to which networks are socioeconomically segregated is poorly understood. We approach the evaluation of income segregation through a novel lens by explicitly considering three scales of analysis within social networks: the network community level, the dyadic level, and a level in between. We find evidence of income segregation at all three levels, though this segregation is neither extreme nor universal. Family income appears to be a socially salient factor in the structure of adolescent social networks. In Chapter 3, three contexts of relevance to the adolescent obesity epidemic—schools, neighborhoods, and social networks—are examined simultaneously. Using a novel combination of social network community detection and cross-classified multilevel modeling, we compare the contributions of each of these contexts to the total variation in adolescent body mass index. After adjusting for relevant covariates, we find that the school-level and neighborhood-level contributions to the variance are modest compared with the network community-level. These results are robust to multiple sensitivity tests. This study highlights the salience of adolescent social networks and indicates that they may be a promising context to address in the design of health promotion programs.

Background: Smoking and obesity are the leading causes of preventable death in the United States and known cancer risk factors, yet the patterning of these conditions within low-income and immigrant populations is not well understood. The distribution of social-environmental stressors is also poorly understood in these populations. Furthermore, the relationship between socioeconomic status (SES) and stress is understudied. Therefore, this dissertation explores the association between SES, psychosocial stress and social-environmental stressors, self-reported smoking and measured obesity. Methods: Papers 1 and 2 used cross-sectional data from the Health in Common study of racially/ethnically diverse adults living in low-income housing in the Boston metropolitan area (n=828). Paper 1 examined associations between SES, chronic stress, and smoking while paper 2 explored the relationship between SES, chronic stress, and obesity. In papers 1 and 2, SES and chronic stress were conceptualized as latent variables where chronic stress was represented by perceived stress and physical characteristics of the neighborhood and external home environment. Analyses were conducted using structural equation modeling (SEM) adjusting for sex, age, and nativity. For paper 3, we performed a targeted and systematic review of the literature on the relationship between SES and perceived stress and then conducted a meta-analysis. Results: The meta-analysis found a one unit standard deviation change in education resulted in a -.13 change in perceived stress. SEM results suggest residents with higher SES had significantly lower odds of being a current smoker (standardized coefficient=-0.31, p<0.05), but not of being obese (standardized coefficient=0.01, p=0.80). Furthermore, residents demonstrating higher chronic stress had significantly higher odds of being a current smoker (standardized coefficient=0.24, p<0.05) and being obese (standardized coefficient=0.13, p<0.05). SES was not associated with chronic stress in either model. Conclusions: SES impacts stress and stress impacts health and health behavior in vulnerable populations. More research is needed to explore the complex and multifactorial relationships between SES, stress, and health, particularly among different population segments. This dissertation work is significant because identifying stress/stressors that act as structural barriers, exacerbating inequities, could result in interventions demonstrating population health impact.

This three-paper dissertation examined the association between social and economic factors and HIV risk behavior among adolescents and young adults in Cape Town, South Africa. Studies in South Africa have demonstrated an important association between receiving financial or material resources and sexual risk behaviors. Most research has focused on economic exchanges that occur as part of transactional sex, and not economic support in main partnerships—which may not be transactional. Furthermore, little research has focused on men. For the first paper, cross-sectional data from Black African and Coloured substance-using women (N = 671) were analyzed to examine the association between the frequency of economic support and unprotected sex in main partnerships. Negative binomial and multinomial logistic regression analyses were conducted to model the number of unprotected sex acts and the consistency of condom use, respectively. Results indicated that receiving daily or almost daily support was positively associated with the number of unprotected sex acts; an association was not observed with consistency of condom use. For the second paper, binomial logistic regression was used to examine whether receiving economic support from a main partner was associated with unprotected last sex, and to test whether gender modified this association among young Black African and Coloured women and men (N = 1,456). Economic support from a main partner was not associated with sexual risk; however, receiving economic support from someone in the household other than a main partner was protective for women. Employment was protective for men and women. The third paper used discrete-time hazard analyses to examine whether employment and schooling were associated with sexual initiation among Black African, Coloured, and White adolescents (N = 1,326). Results indicated that although employment was not associated with sexual initiation, working while in school was associated with greater odds of sexual initiation. These papers provide a better understanding of the social and economic factors that may impact sexual risk behavior. The association between employment and risk behavior is complex, and how sexual risk is defined and measured matters. Findings also suggest the importance of assessing multiple sources of economic support, including frequency of support.

My research examines the adoption and implementation processes involved in transferring a global health policy into national-level practice. More specifically, I consider how high-level stakeholders adopted and street-level, private-sector retailers implemented the Phase I pilot of the Affordable Medicines Facility- malaria (AMFm) between mid-2009 and end-2011. The AMFm — a large-scale program housed at the Global Fund to Fight AIDS, TB and Malaria — sought to improve access to high-quality malaria treatment through financing and delivery strategies using the public and private sectors. To date, the median implementation outcomes have been considered in the Independent Evaluation commissioned by the Global Fund but country-level processes and nuanced considerations of outcomes have gone unexplored. To better understand the AMFm pilot in Ghana, I collected both quantitative and qualitative data between August and December 2011. To consider adoption, I first use a grounded, qualitative approach to address: What explains the stands taken by national stakeholders towards and against participating in the AMFm’s Phase I? I generate explanatory categories about the different views — stands — key stakeholders in Ghana took about joining Phase I. Public health goals; indirect policy goals; and concerns about personal, organization, and national reputation help to explain the views of different high-level stakeholders. Second, I consider the actions taken by different stakeholders: To what extent can a multiple-streams approach to policy adoption help clarify Ghana’s decision to join in the AMFm’s pilot? I find the Multiple-Streams Approach cannot be used to explain adoption of the AMFm pilot in Ghana. However, a modified version accounting for the global and national levels simultaneously can explain this case. To consider implementation, I ask: Do retailers in Northern Region comply with Ghana’s the advertised AMFm Recommended Retail Price among for-profit, private-sector retailers? And, does non-compliance vary systematically with features of retailer structure or conduct? I find high compliance based on reported retail prices. I also find that neither measures of spatial competition nor having seen regulation enforced in the past explain the pattern of non-compliance. Rather, variation in the terminal supply price is highly associated with a retailer’s decision to charge at or above the RRP.

Obesity has become a global epidemic, and obesity prevalence rose from 4.8% to 9.8% in men and from 7.9% to 13.8% in women between 1980 and 2008 1. Preventing obesity and related chronic diseases, especially type 2 diabetes (T2D) and cardiovascular disease (CVD), is of crucial public health significance 2. Identification of dietary factors that are beneficial to health is of high priority. This dissertation focused on two kinds of foods, coffee and soy food, and their associations with obesity-related health outcomes. In Chapter 1, the association of coffee consumption with cardiovascular disease has been investigated in numerous epidemiological studies 3-5. However, a key issue that remains to be resolved is the dose-response relationship of long-term coffee consumption with CVD risk, including incidence of coronary heart disease, stroke, and heart failure, and CVD mortality. In the current study, I examined the association between coffee consumption and risk of CVD by meta-analyzing results from 36 prospective cohort studies with 1,279,804 study participants and 36,352 CVD cases. In Chapter 2, the study described in Chapter 1 showed that coffee consumption was non-linearly associated with risk of CVD: moderate coffee consumption was associated with lower risk of CVD, with the lowest CVD risk at 3 to 5 cups per day, and heavy coffee consumption was not associated with risk of CVD. However, whether the non-linear association was due to a true biological effect or confounding of smoking is not known. In this chapter, with 208,501 participants and 31,956 deaths in three large cohort studies, I prospectively examined the associations of coffee consumption with total mortality and cause-specific mortality among the overall population as well as never smokers. In Chapters 3 and 4, the association of isoflavone intake with risk of T2D has been assessed, and mixed results were observed 6-8. Clinical trials showed that isoflavone supplements did not improve glucose control 9-10, however, these clinical trials were limited by small sample size and short duration of follow-up. Therefore, I further examined the association of soy food with risk of type 2 diabetes. Two different approaches were used to assess soy food intake. First, soy food assessed by food frequency questionnaire (FFQ) was used as main exposure, and the association of soy food consumption with risk of type 2 diabetes was examined prospectively in three Harvard cohorts. Second, urinary isoflavones excretion was used as main exposure, and the association of urinary isoflavones concentration with risk of type 2 diabetes was assessed using a nested case-control design.

Chapter one of this manuscript identifies the sources, composition, and temporal variability of fine (PM2.5) and coarse (PM2.5-10) particles. A Harvard Impactor was used to collect daily particle samples from 2002-2010 at a single site in Boston, MA. Filters were analyzed for elements, black carbon (BC), and total PM mass. Positive Matrix Factorization (PMF) identified BC, S, Pb, V, and Ni to be associated mostly with the fine particle mode, and Ca, Mn (road dust), and Cl (sea salt) mostly with the coarse mode. PMF identified six source types for PM2.5, including regional pollution (48%), motor vehicles (21%), wood burning (19%), oil combustion (8%), crustal/road dust (4%), and sea salt (<1%). Three source types were identified for PM2.5-10, including crustal/road dust (62%), motor vehicles (22%), and sea salt (16%). A linear decrease in PM concentrations with time was observed for both fine (-5.2%/yr) and coarse (-3.6%/yr) particles. That PM2.5 is decreasing more sharply than PM2.5-10 over time suggests the increasing importance of PM2.5-10 and traffic-related sources for PM exposure and future policies. Chapters two and three use variables related to PM2.5 concentrations to quantify PM2.5 exposure where PM monitoring does not exist. In chapter two, validation of a model calibrating the PM2.5-visibility relationship using data collected in Kuwait from 2004-2005 demonstrated the ability to predict PM2.5 exposure at monitoring sites. In chapter three, validation of a model calibrating the visibility-AOD relationship using data collected from 2006-2007 in Iraq demonstrated the ability to predict visibility between monitoring sites at a fine resolution (1x1 km), and in turn estimate PM2.5 spatially. Applying these relationships to predict PM2.5 at locations exterior to the calibration sites showed high temporal and spatial variability in PM2.5 exposure (10-365 µg/m3). The ability to obtain precise estimates of PM2.5 concentrations in Southwest Asia and Afghanistan is of high utility for epidemiologists seeking to understand the relationship between chronic PM2.5 exposure and respiratory diseases among deployed military personnel stationed throughout the region. That predicted PM2.5 varies widely over space and time supports the ability to successfully utilize our estimates to understand this relationship in the region.

It is well established that sugar-sweetened beverages (SSBs) are associated with obesity and chronic diseases. Although there is some emerging evidence that consumption of added sugars is declining in the United States, on average Americans’ consumption still exceeds recommended levels. Consequently, it is imperative that researchers continue to delve further into the question of exactly how SSBs influence obesity and associated chronic diseases, as well as consider creative and novel strategies for reducing their impact on consumers’ health. Several important gaps in the research are addressed by this dissertation. Chapter one considers the role of SSBs and overall diet quality with respect to the growing body of evidence that demonstrates an association between sleep duration and obesity. We used linear regression to examine the associations of sleep duration with dietary indicators in elementary school students taking part in a multi-sector, community-based obesity prevention intervention (the Massachusetts Childhood Obesity Research Demonstration Project). We found that students who reported sleeping <10 hours/day consumed soda more frequently and vegetables less frequently compared with students who reported optimal sleep. Chapter two assesses whether fast food customers are worse at estimating the caloric content of their meal when their purchase includes a high-calorie beverage (HCB). We used linear regression to examine the association between purchasing HCB and calorie estimation among adult and adolescent fast food customers, and found that among adults, drinking HCB contributes to underestimating calories. HCB may be influencing calorie estimation in a unique way compared to high-calorie food items. Chapter three considers the relevance of SSBs with respect to proposed changes to the Supplemental Nutrition Assistance Program (SNAP). Using sales data from a large supermarket chain in the Northeast, we used multivariate analysis of variance to determine whether there is an association between SNAP receipt and shopping patterns. We found that SNAP shoppers spent more than non-SNAP shoppers on sugar-sweetened beverages, red meat, and cold convenience foods, and spent less on fruits, vegetables and poultry. Each chapter lends additional support for a focus on SSB consumption in obesity prevention efforts and will inform the development of prevention strategies in the future.

There is a growing body of research that suggests that risk factors in the prenatal period and early childhood are important to the development and prevention of overweight in childhood and its sequelae across the life course. The objective of this dissertation is to elucidate the relationships between socially-patterned maternal exposures and early childhood weight. The three studies draw on the Early Childhood Longitudinal Study-Birth Cohort, a nationally-representative sample of children born in the United States in 2001 and followed from birth through kindergarten. Chapter 1 examines the relationship between maternal pregnancy intention status and subsequent child’s weight at ages 9 months, 2 years, and 4 years using linear regression and multilevel models. Children from pregnancies reported by the mother as unwanted were more likely to have higher weight-for-length z-scores than children from pregnancies reported as intended only at age 9 months, and breastfeeding duration partially mediated the relationship. Chapter 2 examines whether timing of mother’s return to work after childbirth is associated with early childhood weight and height at ages 9 months, 2 years, and 4 years, and whether this association varies by household socioeconomic status and maternal race and ethnicity. Results from linear regression models suggest that mother’s earlier return to work is associated with lower height for children from lower socioeconomic status households compared to their peers from high socioeconomic status households. Chapter 3 investigates the associations between enactment of state-level breastfeeding promotion laws, specifically pertaining to breastfeeding in public, breastfeeding or pumping in the workplace, and breastfeeding mothers being exempt from jury duty, and child’s weight-for-length z-score at 9 months of age, and examines whether these relationships vary across mother’s racial and ethnic groups. Results from multilevel models examining each of the state-level breastfeeding laws were null. Using a nationally-representative longitudinal dataset, this dissertation contributes new scientific evidence and sheds light on potential mechanisms that can affect early childhood weight, which can help inform policies and programs aimed at reducing inequalities in early childhood health.

This dissertation explores the relationships between island food consumption, sociodemographic variables, and cardiovascular risk using data from the Pacific Islander Health Study (PIHS). Chapter 1 explores the associations between self-reported level of island food consumption and key covariates. Island food consumption was modeled using Poisson regression and adjusted for demographic, socioeconomic, and cultural characteristics. Increased Pacific Island cultural affinity was the strongest predictor of increased island food consumption while being formerly married was associated with decreased island consumption. These results speak to cultural norms and lifestyle choices that influence dietary choice. Chapter 2 examines how the associations between Pacific Island cultural affinity and island food consumption is moderated by demographic covariates. Following exploratory factor analysis, two separate factors emerged. The scale was split into two subscales measuring cultural activity participation and culturally relevant media consumption. No significant interactions were found between the activity subscale and any of the covariates, but significant interactions were found between the media subscale and ethnicity and birth country. Differences in media consumption – and therefore food advertising or food portrayal – by birth country may drive the moderation seen in these interactions. Chapter 3 assesses the association between island food consumption and cardiovascular risk using linear regression. A 9-point cardiovascular risk score was constructed. After adjustment for four blocks of covariates island food consumption was not significantly associated with cardiovascular risk. In bivariate analyses, island food consumption was significantly associated with higher fruit and vegetable consumption, but not fast food or sugar sweetened beverage consumption. These findings suggest that the type of island foods consumed by PIHS participants may include high levels of fruits and vegetables that are part of a heart-healthy diet. This dissertation is a first step in understanding dietary patterns of Pacific Islander Americans and has generated several hypotheses that could be used to inform future work. Showing how island food consumption in this small, but growing population will one day inform both policy makers and targeted dietary interventions.

Malaria is a major cause of morbidity and mortality worldwide, particularly among young children in Sub-Saharan Africa. There exists a very effective treatment for malaria, a class of drugs known as artemisinin-based combination therapies (ACTs). However, it is estimated that only 12-22% of children with malaria were treated with an ACT in 2015 and, even when people take ACTs, they do not always complete the full treatment course. This dissertation examines malaria treatment behaviors in Central and Eastern Uganda. The first two chapters of this dissertation use a randomized controlled trial to evaluate interventions designed to increase patient adherence to the ACT treatment regimen. We find that 35% of patients who purchase subsidized ACTs from local drug shops in central Uganda do not complete the full 3-day treatment course. In the first chapter, we show that patients who felt better mid-way through treatment were more likely to stop taking the medication than those who were still unwell. However, short messages promoting adherence, delivered via stickers affixed to the standard ACT package, increased adherence rates by approximately 9%. The second chapter examines the impact of diagnostic testing for malaria on adherence to the malaria treatment regimen. Since most people do not receive a confirmed diagnosis of malaria before beginning treatment, uncertainty about the true cause of illness may be a factor in nonadherence. We find that patients who tested positive for malaria on a rapid diagnostic test did not have higher odds of adherence. However, we present some evidence that patients who tested positive were more likely to consume a few additional pills of the drug. The third chapter of this dissertation shows that the probability that a febrile patient has malaria declines substantially with the age of the patient and increases with the local village prevalence rate. We find, however, that ACT treatment rates are invariant to age and local prevalence, and beliefs about whether an illness is malaria also do not vary with age or prevalence. This work suggests that mis-perceptions of malaria risk may be contributing to the under-treatment of malaria in young children.