Metalworking fluids (MWF) are coolants and lubricants used in industrial machining and grinding operations and have been linked with several cancers. Components found in MWFs, including polycyclic aromatic hydrocarbons found in oil-based straight fluids and nitrosamines found in water-based synthetic fluids, have been linked with mammary gland as well as respiratory organ tumors in laboratory animals. Epidemiologic evidence for these cancers, however, is less conclusive. Despite the provocative toxicologic evidence, there are few epidemiologic studies on MWF exposure and breast cancer. Results for lung cancer have been null for oil-based MWF and appeared inversely related to synthetic fluids, possible due to presence of endotoxin. The United Autoworkers-General Motors (UAW-GM) study includes an occupational cohort of 46,316 hourly workers in automotive manufacturing and extensive MWF exposure data, with annual average exposure estimates available for workers’ full employment history, and over two decades of cancer incidence follow-up. To better understand the exposure-response relationship between MWF exposure and these cancer outcomes, this dissertation set out to improve upon prior studies by first evaluating the presence of a common source of bias in occupational epidemiology, the healthy worker survivor effect (HWSE) in the UAW-GM cohort, and second collecting additional data and using advanced methods to evaluate these relationships.

Chapter 1 is the assessment of the presence of the HWSE in cancer studies of the UAW-GM cohort. The HWSE can affect the validity of occupational studies when data are analyzed incorrectly. HWSE depends on three underlying conditions: (1) leaving work predicts future exposure, (2) leaving work is associated with disease outcome, and (3) prior exposure increases probability of leaving work. If all these conditions are satisfied employment status is a time-varying confounder affected by prior exposure and standard methods will produce bias. I evaluated the presence of these conditions for select cancer outcomes, including lung cancer, in the UAW-GM cohort and found evidence for all three conditions. This suggested that standard methods may underestimate the exposure-response for lung cancer and therefore a g-method should be applied to control for employment status as a time-varying confounder affected by prior exposure. A secondary analysis examining breast cancer among female workers found insufficient evidence for condition (3), indicating that standard methods are appropriate for this outcome and will not produce bias due to HWSE.

In Chapter 2, I evaluated the exposure-response relation between cumulative MWF exposure and breast cancer incidence among female workers in the UAW-GM cohort. Additional data was obtained by extending follow-up four more years for female cohort members using data linkage with the Michigan Cancer Registry and the National Death Index. We identified 221 total incident breast cancer cases among 4,503 female workers. Risks associated with exposure to the three types of MWFs, straight, soluble, and synthetic, were evaluated using Cox proportional hazards models for all breast cancer cases as well as for pre-menopausal cases defined by age at diagnosis. Results suggest an increasing exposure-response curve for straight fluids and breast cancer. While the number of the pre-menopausal cancers is small, results are modestly suggestive of an increased risk associated with higher synthetic fluid exposure, suggesting a different mechanism for the younger cases.

In Chapter 3, I examined the relationship between lung cancer mortality and exposure to straight and synthetic MWF, as well as to the biocides that are added to water-based fluids to control microbial growth. Using the parametric g-formula, ratios were estimated comparing cumulative risk of lung cancer mortality under the hypothetical interventions always high exposure while at work and always unexposed. We also intervened on both synthetic MWF and biocides simultaneously to estimate independent effects. Results from this study suggest slightly elevated lung cancer mortality related to straight MWF exposure, albeit with wide confidence intervals. Our results do not support a negative association for synthetic fluids, as reported in earlier studies; instead, biocide in the fluid, a marker for the release of endotoxin, was associated with decreased lung cancer.

The hypotheses addressed in this dissertation are of public health importance in light of the extremely high incidence of both breast and lung cancer in the US. These studies provide information on the relationships between MWF exposure, breast and lung cancer incidence, adding to the scientific literature that informs regulatory measures and protects workers from the risks of MWF exposure.

Estimates of the risk of occupational exposures are typically based on observational workplace studies that are subject to bias due to the healthy worker survivor effect (HWSE), a ubiquitous process that results in the healthiest workers accruing the most exposure. This body of work is concerned with the estimation of causal effects of occupational exposures from observational workplace studies, in the context of the HWSE.

We estimate the effect of cumulative exposure to straight, soluble, and synthetic metalworking fluids (MWFs) on the incidence of colon cancer in the United Autoworkers-General Motors (UAW-GM) cohort. We use longitudinal targeted minimum loss-based estimation (TMLE) to compute the 25-year risk difference if always exposed above compared to if always exposed below an exposure cutoff while at work. Exposure cutoffs were selected a priori at the median of exposed person-years among colon cancer cases. Risk differences are 0.038 (95% CI = 0.022 to 0.054), 0.002 (95% CI = -0.016 to 0.019), and 0.008 (95% CI = 0.002 to 0.014) for straight, soluble, and synthetic MWFs, respectively. By control of the time-varying confounding on the casual pathway that characterizes the HWSE, TMLE estimated effects that were undetectable in earlier reports.

Most workers in UAW-GM were hired decades before the reporting of incident cancers began. Incident cancers that occurred before the start of reporting were left filtered. We show that if ignored, left filtering can lead to downward bias in exposure effect estimates. Further, we propose a novel delayed-entry adjusted Kaplan-Meier estimator that controls for time-varying confounding, and permits delayed risk-set entry. The estimator results in little bias in simulated datasets when the outcome is sufficiently rare.

In addition to dynamic (realistic) interventions that assign exposure according to workers' employment status, causal contrasts can be defined under static (etiologic) interventions that additionally prevent leaving work. Causal effect estimates of the two classes of interventions can differ substantially. While ideally the choice of intervention would be driven by the research question, in practice it may be dictated by the available data. Furthermore, when estimates of the long-term etiologic effects of occupational exposures are not available, guidelines for exposure limits may be based on studies that estimated effects of realistic interventions. In a simulation study we investigate the conditions under which the two effect measures are comparable, and identify factors that drive the differences between the two.

Work has long been recognized as a fundamental determinant of health, and stressful work is linked with symptoms of psychiatric illness such as distress, sadness, hopelessness, and depressive disorder. Blue-collar work emerges in the literature as an important example of this connection. Not only do repetitive, monotonous, and physically demanding tasks predispose workers to disability and injury, the erratic schedules and long hours oftentimes associated with blue-collar jobs may uniquely predispose workers to depression. Blue-collar jobs, in contrast with clerical and white-collar jobs, are also frequently characterized by exposure to a range of physical hazards, including heat, noise, chemicals, and particulate matter.

Women now represent over one-fourth of the 12 million workers in manufacturing jobs in the U.S. Female blue-collar workers may have less opportunity to adjust work demands or modify their duties following injury and take longer work-related sick leave. However, our current understanding of the effects of blue-collar work on mental health outcomes, such as depression are based largely on the experiences of men. Because depression is more common and more persistent in women, the lack of evidence regarding risk for depression among female blue-collar workers constitutes an important gap in the literature.

In response, this dissertation is focused on the implications of gender and aspects of the work environment for the mental health of workers in the U.S. aluminum industry. The American Manufacturing Cohort (AMC) study database is comprised of multiple, linked administrative datasets that collectively provide detailed, longitudinal follow-up for understanding work-life exposures, health and economic outcomes in a cohort of U.S. light and specialty metals workers. The study data are the product of an academic-corporate partnership forged in 1997 between Alcoa, Inc. and the Occupational and Environmental Medicine Program at the Yale University School of Medicine with the goal of incorporating research findings into company policy in an effort to improve worker safety. Study data consist of several distinct administrative datasets, and records for individual workers can be deterministically linked across datasets with a unique, encrypted identifier. Complete medical claims data are available for workers enrolled in their local preferred provider organization (PPO) health insurance plan (approximately 97% of the workforce). The analyses described herein are based on records from 30,035 men and 7,148 women employed in blue- and white-collar jobs by Alcoa Inc. between January 2003 and December 2013. They are the first to explicitly focus on the experiences of blue-collar women in the AMC Study.

For Chapter 1, I compare trends in depression by gender and occupational class in the study population of 37,183 workers. Chapters 2 and 3 are focused only on blue-collar workers. For Chapter 2, I characterize the association between the workplace gender composition and treatment for depression among male and female blue-collar workers. More specifically, I use g-computation to estimate risk differences that contrast the predicted risk of treatment for depression under hypothetical interventions on gender composition as compared with the observed data. Finally, in Chapter 3 I examine the effects of layoffs on remaining blue-collar workers’ use of mental healthcare services and injury risk with a difference-in-differences approach. Collectively, these three chapters provide a novel examination of the association between specific aspects of blue-collar work and risk for depression, with an emphasis on the unique experiences of women, in an unusually rich and unique administrative database.

Rationale: Exposure to ambient air pollution is a major environmental risk factor for chronic diseases such as asthma. Children with asthma can be even more susceptible to the effects of air pollution since their respiratory system is not fully developed and some of the air pollutants can trigger asthma attacks. Over the past decades, scientists and researchers recognized the need to improve our understanding in the biological response mechanisms. The exact underlying mechanisms linking air pollution to disease outcomes, however, are not clear.

Objectives: The overarching aims of this thesis are to investigate the association between exposure to ambient air pollutants and adverse health effects among minority children and identify potential biological pathways from exposure to health end points by considering genetic ancestry and, asthma endotype (atopy) as effect modifiers of the relation between air pollution and telomere length. In Chapter 1, we investigated the association between ambient air pollutants and asthma exacerbations in urban minority children, as well as effect modification by atopy status and African ancestry. In Chapter 2, we conducted a pilot study to gather preliminary information about how telomere length varies in relation to polycyclic aromatic hydrocarbons (PAH) exposure in children living in a highly polluted city. In Chapter 3, we examined the association between ambient air pollutants and telomere length in minority children to understand the potential damage caused by air pollution at the molecular level.

Methods: In Chapter 1, air pollutant exposures were estimated based on residence using U.S. EPA monitoring data and inverse distance weighting. The associations between average daily exposures and asthma exacerbations were estimated by the incident rate ratio (IRR) from a negative binomial regression model. In Chapter 2, we selected asthmatic and non-asthmatic subjects based on their annual average PAH level and described patterns of telomere length, measured by using uniplex polymerase chain reaction (PCR). In Chapter 3, the annual average daily exposure to each of four air pollutants was examined in relation to telomere length.

Results: In chapter 1, exposure to ambient O3 and NO2 were associated with asthma exacerbations. Results for PM2.5 were null. Exposure-response relationships were linear for O3 and NO2 among non-atopic subjects and inconsistent among atopic subjects. Effect modification by African genetic ancestry was present only for O3; the impact of exposure appeared to be larger for those with higher African ancestry. In chapter 2, we found an inverse linear relationship between PAH and telomere length in a small pilot study. In chapter 3, the association between ambient SO2 and telomere length was significantly negative, whereas results for PM2.5, NO2 and O3 were null.

Conclusions: Our results provide further evidence that exposure to ambient air pollution is a serious environmental risk factor that causes adverse health outcomes among minority children.

This dissertation focuses on assessing the effect of shift work on worker health, specifically identifying and evaluating selected working time characteristics as risk factors for incident hypertension and Type II diabetes mellitus. Shift work is a common occupational exposure across many sectors of the economy with a prevalence of near 20% of the US workforce. Shift work has also been consistently associated with many adverse health outcomes, including cancer. While most research has focused on the associations of night shift work and female breast cancer, recent evidence suggests that shift work may also cause an increased risk of reproductive disorders, gastrointestinal ulcers and cancers, prostate cancer, hypertension and Type II diabetes. In order to protect worker’s health while maintaining the efficiency of 24-hour work, it is important to identify the potentially harmful characteristics of shift work.

In chapter 1 we discuss definitions for shift work and present current hypotheses for how shift work impacts human health. We also discuss the challenges facing environmental health scientists and epidemiologists as they define and characterize a complex exposure, such as shift work. An effective assessment of the impact of shift work on human health is dependent on development of metrics of shift work that are designed to capture the biological impact of shift work, rather than workplace schedule variation.

In chapter 2 we develop and characterize the prevalence of selected working hour characteristics that may impact circadian rhythms in the American Manufacturing Cohort. In this first description of shift work in a US manufacturing workforce, we demonstrate that working hour characteristics such as shift type, duration, intensity, rotational direction, and social aspects of work need to be considered concurrently. Furthermore, these working hour characteristics vary by annual shift schedule. We identify, as expected, that permanent day workers have the lowest percentage of quick returns and rotations (as a permanent schedule would imply). However, working the day shift does not provide absolute protection from potentially disruptive characteristics of working hours since long work hours, as well as quick returns and rotations occurred when workers switched from morning to afternoon shifts (both considered day work). Notably, we also demonstrate that older workers are more likely to work permanent day schedules, while racial minorities such as African American workers are more likely to perform work with rotations.

In chapter 3 we present an analysis examining the impact of recent night and rotation work exposure and risk of hypertension. We apply the same definitions of night and rotational work described in chapter 2, and present evidence that combinations of recent night and rotational work increase the risk of hypertension. In particular, we identify higher risk of hypertension among all levels of recent night work compared with non-night workers. The highest risk of hypertension were among those with 95-100% night work, workers who would normally be considered ‘permanent night workers’. This suggests that permanent night workers are experiencing circadian rhythm disruption even though their work schedules are not rotating. Furthermore, we observe elevated hazard ratios for all combinations of night and rotational work compared to non-night workers. In particular, the hazard ratio was almost 4-fold for workers with mostly night work and frequent rotations compared with non-night workers. Even those workers with mostly non-night work and infrequent rotations had a 2-fold risk of hypertension, indicating the potential importance of recent shift work and hypertension risk.

In chapter 4 we analyzed the association between cumulative months of night work and average percentage of night shifts over follow-up using Cox proportional hazard models. The results from the Cox proportional hazard models provide some modest evidence that night work exposure may be associated with an increased risk of diabetes. This association was seen when night work exposure is classified as either the cumulative number of months of night work or the average percentage of night shifts over follow-up.

Chapter 5 concludes the dissertation with a summary of the results from each chapter, the strengths and limitations of the current work, and a discussion of the next steps in shift work research.

Abstract

Respiratory exposures to solvents and metalworking fluids in relation to chronic health outcomes: strategies for reducing bias within conditional modeling approaches

by

Stella Fay Beckman

Doctor of Philosophy in Epidemiology

University of California, Berkeley

Professor Ellen A. Eisen, Chair

Background

Occupational injury and illness are significant contributors to the global burden of disease, with more than 2 million deaths caused by work-related hazards estimated to occur worldwide each year. The goal of occupational epidemiology is to identify and characterize the causes of work-related injury and illness in order to inform workplace health and safety interventions. While occupational groups provide opportunities to study exposures at higher levels than those seen in the general population and with well-characterized exposures, studies of chronic diseases among workers are subject to a unique form of bias: the healthy worker effect. This results in downward bias, which can even make a truly hazardous exposure appear protective, under two circumstances: 1) when workers, who are overall healthier, are compared to the general population and 2) when workers with more robust health (survivors) remain in highly exposed jobs and accrue more exposure than those who leave work. While causal inference techniques are required to fully address this effect, there are approaches that may mitigate the bias within a standard conditional model.

Methods

In this dissertation, I explore the associations between occupational exposures and health outcomes with particular attention to bias caused by the healthy worker effect. In the second chapter, I estimate the association between acquired color vision defects and past exposure to n-hexane in a cross-sectional study of active and retired automobile repair workers in the San Francisco Bay Area, California. Color vision defects are hypothesized to be an early indicator of neurotoxicity caused by n-hexane. In the third and fourth chapters, I consider exposure to metalworking fluid (MWF) in a large cohort of machinists. I investigate the association between MWF and chronic obstructive pulmonary disease mortality in chapter three, using indirect adjustment for confounding by cigarette smoking status as well adjustment for cardiovascular disease mortality as a potential competing risk. In chapter four, I estimate the association between mineral oil-based MWF and mortality due to natural causes and cardiopulmonary disease in weighted Cox proportional hazards models, using inverse probability weights to address informative censoring and selection bias in a cross-sectional sample of the cohort. I contextualize all findings in within subject matter literature and discuss limitations of the methods chosen to address bias due to the healthy worker effect.

Conclusion

The chapters of this dissertation explore the associations between common industrial exposures (solvents and metalworking fluid) and health outcomes occurring many years after exposure. I find an elevated, though not statistically significant, association between n-hexane exposure and acquired color vision defects among automobile repair mechanics exposed to levels of n-hexane below current regulatory limits. For exposure to MWF, I find elevated risk of chronic obstructive pulmonary disease with 95% confidence intervals excluding 1 when adjusted for cardiovascular disease as a competing risk in Cox proportional hazards models. I report modestly but inconsistently elevated risks of natural cause and cardiopulmonary disease mortality with exposure to mineral oil-based MWF in models weighted to adjust for informative censoring and sample selection. Given the evidence of residual downward bias due to elements of the healthy worker survivor effect that can only be addressed with causal inference methods, these results taken as a whole suggest that current occupational safety and health regulations are not sufficiently protective of worker health, and should be reconsidered in light of this evidence.

The main objective of this dissertation was to investigate the cardiovascular disease (CVD) consequences of structural racism operating implicitly as discriminatory employment and labor policies and practices. Public health interventions targeting proximal risk factors such as health behavior have effectively reduced overall morbidity and mortality; yet, disparities in CVD mortality between Black and white populations persist even with a focus on working populations, which would presumably have similar baseline education, health, and healthcare access within skill levels.

The American Heart Association has stated that structural racism is a fundamental cause of disparities in CVD risk as it creates mutually reinforcing systems of unequal distribution of wealth, resources, opportunities, and power. First, we provide a background on the relationship between structural racism and cardiovascular health wherein ongoing interactions between macrosystems and institutions constrain the opportunities, resources, and power of socially racialized groups. This functions to prevent the elimination of Black-white cardiovascular health gaps. In the first chapter, we make explicit how structural racism is posited to function through the workplace in the form of labor and employment organizational practices and norms that impact CVD mortality. Next, we assess structural racism’s influence on workforce policies and practices, particularly workplace racial diversity, which may present upstream targets for assessing and reducing racial health disparities in CVD. We evaluate whether a more racially balanced workforce would be protective against CVD mortality in a retrospective study of nearly 40,000 hourly Michigan autoworkers. We found that there was a protective effect of increasing exposure to racial diversity that was strongest and most consistent among Black workers. In this first chapter, our findings provided evidence that workplace practices encouraging diverse hiring and retention have the potential to improve all workers’ health, particularly the socially racialized groups in that workforce.

Work as a social determinant of cardiovascular health and workplace exposures to particulate matter both remain understudied among Black working populations. In the next chapter, we aimed to assess the risk of CVD mortality among Black autoworkers from the United Autoworkers – General Motors (GM) cohort study under several hypothetical scenarios: increasing exposure to workplace racial diversity, reduced metalworking fluid exposures, and both. Using longitudinal data from Black workers from the Detroit GM plant, we applied the parametric g-formula to assess the risk of CVD mortality under hypothetical scenarios with set values for plantwide racial diversity, and selected exposure limits for metalworking fluid, separately and jointly. Our findings were consistent with the hypothesis that interventions to increase racial diversity and decrease metalworking fluid would reduce CVD mortality risk. Our study underscored the importance of a racially diverse workplace in addition to reducing occupational hazard concerns for the prevention of CVD deaths among racialized populations.

Structural racism not only influences racial diversity, but also the ongoing racial segregation of Black Americans which has been associated with several cardiovascular outcomes. In recent decades, racial segregation has focused almost exclusively on measures of dissimilarity in residential spaces; however, little research has explored segregation beyond the residence. We present an analysis on the occupational structures and organization of work by examining the impact of job-based racial segregation and job mobility. This analysis used the index of dissimilarity to measure segregation and estimated exposure to cumulative job trajectories. We found that the relationship between increasing segregation and CVD mortality risk was highest among Black workers with lower job mobility. Black workers with higher job trajectories also experienced the harmful impacts of racial segregation, though at a lower magnitude. Among white workers, increasing racial segregation was harmful only for those with higher job mobility. Importantly, the higher-grade job assignments also have the most exposure to the GM occupational hazard of interest, metalworking fluid. Thus making our interaction between segregation and job mobility a cruicial method which can isolate pathways from segregation to CVD mortality via occupational hazards and psychosocial stress. Our findings provided evidence that workplace racial segregation may increase CVD mortality risk among Black workers, regardless of their career growth, and that addressing workplace racial segregation in concert with increasing opportunities for job mobility may be important to improving the long-term health of racialized working populations.

The hypotheses addressed in this dissertation are of public health importance considering the persistent racial disparities in CVD mortality and the gaps in the literature regarding the long-term health impacts of structural racism in the workplace. These studies provide information on the relationships between racial discrimination, occupational hazards, and CVD mortality, adding to the scientific literature that informs workplace policies and practices that aim to reduce racial disparities in preventable chronic disease.

Objectives

Methods

Results

Conclusions