About

Ellen Eisen is a professor at the UC Berkeley School of Public Health, with expertise in biostatistics, occupational and environmental health, and epidemiology. Her research focuses on methods in occupational epidemiology, particularly addressing analytical limitations in exposure-response modeling and healthy worker survivor bias using causal methods. She is interested in questions about the role of work in health, including the contribution of occupational exposures to disease risk, and has a specific focus on chronic heart and lung disease related to workplace exposures, as well as the risks of opioid mortality and suicide in manufacturing and mining populations. Her work aims to improve understanding of occupational health risks and develop methodological approaches to better assess these risks.

Research topics

• Internal medicine
• Medicine
• Environmental health
• Biology
• Genetics
• Emergency medicine
• Engineering
• Ecology
• Demography
• Mechanical engineering

Selected publications

• Associations between prenatal and early-life air pollution exposure and lung function in young children: exploring influential windows of exposure on lung development

  ISEE Conference Abstracts · 2022

  • Environmental health
  • Medicine
  • Biology

  Background and Aim: Evidence in the literature suggests that air pollution exposures experienced prenatally and early in life can be detrimental to normal lung development, however potentially more influential windows during development are not fully understood. We evaluate air pollution exposures during the prenatal and early-life period in association with lung function at ages 7-9, in an effort to identify potentially influential windows of exposure for lung development. Methods: Our study population consisted of 222 children aged 7-9, from the Fresno-Clovis metro area in California with spirometry data collected between May 2015 and May 2017. We used distributed-lag non-linear models to flexibly model the exposure-lag-response for monthly average exposure to fine particulate matter (PM2.5) and ozone (O3) during the prenatal months and first three years of life in association with forced vital capacity (FVC), and forced expiratory volume in the first second (FEV1), adjusted for covariates. Results: PM2.5 exposure during the period spanning pregnancy and the first 3-years of life was associated with lower FVC and FEV-1. Specifically, an increase from the 5th percentile of the observed monthly average exposure (7.55 μg/m3) to the median observed exposure (12.69 μg/m3) was associated with 0.42 L lower FVC (95% confidence interval (CI): -0.81, -0.03) and 0.38 L lower FEV1 (95% CI: -0.74, -0.01). The shape of the lag-response indicated that the second half of pregnancy may be a more influential window of exposure. Associations for ozone were not as strong and typically CIs included the null. Conclusion: Our findings indicate that prenatal and early-life exposures to PM2.5 are associated with decreased lung function later in childhood. Exposures during the latter months of pregnancy may be especially influential. Keywords: air pollution, lung development, distributed-lag models, exposure windows

  PublisherDOI

• S-186 The impact of job loss on self-injury mortality in a cohort of autoworkers: application of a novel causal approach

  2021

  Senior authorCorresponding
  • Medicine
  • Demography
  • Emergency medicine

  <h3>Introduction</h3> Recent increases in national rates of suicide and fatal overdose have been linked to a deterioration of economic and social stability. The American auto industry experienced these same pressures beginning in the 1980s with the emergence of a competitive global market. <h3>Objectives</h3> Using the United Autoworkers – General Motors cohort as a case study, we examine the impact of employment loss on these self-injury mortality events. <h3>Methods</h3> For 29,764 autoworkers employed on or after January 1, 1970, we apply incremental propensity score (IPS) interventions, a novel causal inference approach, to examine how proportional shifts in the odds of leaving active GM employment affect the cumulative incidence of self-injury mortality. <h3>Results</h3> Cumulative incidence of self-injury mortality was 0.87% (258 cases) at the observed odds of leaving active GM employment (δ = 1) over a 45-year period. A 10% decrease in the odds of leaving active GM employment (δ= 0.9) results in an estimated 7% drop in self-injury mortality (239 cases) while a 10% increase (δ= 1.1) results in a 19% increase in self-injury mortality (306 cases). We reject the null hypothesis of no incremental effect over the range of effects considered (δ in [0.75, 1.25]; P = 0.02). <h3>Conclusion</h3> These results suggest that leaving active employment increases the risk of death due to suicide or drug overdose.

  PublisherOA PDFDOI

• P-280 Metalworking fluids and cancer incidence in the UAW-GM autoworkers cohort

  Poster presentations · 2021-10-22

  articleOpen accessSenior author

  <h3>Introduction</h3> Metalworking fluids (MWF) are complex mixtures of oils and chemical additives used to cool and lubricate metal machining operations. Previous studies have reported increased risk of specific cancers associated with MWF exposure. <h3>Objectives</h3> This report broadly examines cancer incidence in the United Auto Workers-General Motors (UAW-GM) cohort exposed to MWFs with extended follow-up (through 2015). The outcomes of interest were melanoma, leukemia, non-Hodgkin lymphoma and cancers of the colon, rectum, pancreas, esophagus, stomach, larynx, lung and bronchus, breast, prostate, kidney and renal pelvis, and bladder. <h3>Methods</h3> The cohort includes 39,132 workers followed for cancer incidence in Detroit area Surveillance, Epidemiology, and End Results Program (SEER) and Michigan cancer registries, from 1973–2015. Cox models yielded estimates of adjusted hazard ratios (HR) with categorical variables for lagged cumulative exposure to each MWF (straight, soluble, and synthetic). <h3>Results</h3> There were 7,809 cancer cases of interest. Over 43 years of follow-up, the incidence of several types of cancers was significantly elevated in relation to at least one type of MWF; exposure–response patterns were consistent with prior reports from this cohort. We found significantly increased incidence of stomach and kidney cancer associated with higher levels of straight fluid exposure and increased rectal and pancreatic cancer with increasing synthetic fluid exposure. Only Non-Hodgkin lymphoma was associated with soluble MWF, with HRs significantly elevated in the highest exposure category at 1.70 (95% Confidence Interval (CI): 1.13–2.54). <h3>Conclusions</h3> Our results provide further evidence of associations between MWF exposure and several types of cancer. This study summarizes information on the incidence of the fourteen cancer types with reduced bias from both the healthy worker hire effect and left truncation. However, the HRs presented do not address potential downward bias from the healthy worker survivor effect which may be necessary to correct in future targeted analyses.

  PublisherOA PDFDOI

• Metalworking fluids and cancer mortality in a US autoworker cohort (1941–2015)

  Scandinavian Journal of Work Environment & Health · 2020 · 19 citations

  Senior authorCorresponding
  • Medicine
  • Environmental health
  • Internal medicine

  Objectives This report describes the extended follow-up (1941-2015) of a cohort of 38 549 automobile manufacturing workers with potential exposure to metalworking fluids (MWF). The outcomes of interest were mortality from cancers of the esophagus, stomach, intestine, rectum, bladder, liver, pancreas, larynx, lung, skin, prostate, brain, and female breast, as well as leukemia. This report includes 5472 deaths from cancer, more than ten times the numbers of deaths in our last summary report published 20 years ago. Methods Standardized mortality ratios were computed for the entire study period. Adjusted hazard ratios (HR) were estimated in Cox proportional hazard models with categorical variables for cumulative exposure to each type of MWF. Results Exposure-response patterns are consistent with prior mortality reports from this cohort. We found increased risk of skin and female breast cancer with straight fluids. For the first time, we found elevated risk of stomach cancer mortality. Overall, many of the exposure-response results did not suggest an association with MWF. Conclusions Mortality is a poor proxy for cancer diagnosis for treatable cancers and not the optimal outcome measure in etiological studies. Although the HR presented here handle bias from the healthy worker hire effect and left truncation, they do not handle bias from healthy worker survivor effect, which likely results in underestimates of the health impacts of MWF. Although this updated summary provides some information on the risk of cancer from MWF, targeted future analyses will help clarify associations.

  PublisherOA PDFDOI

• O1E.5 Short-term disability leave and employment termination: using marginal structural models to estimate counterfactual risks

  Occupational and Environmental Medicine · 2019-04-01

  articleOpen accessSenior author

  Introduction Short-term disability leave can be considered as a measure of not being well enough to work. The American Manufacturing Cohort, followed 1996–2013, consists of employees of a light-metal company that provided short-term disability insurance to all employees: coverage to replace wages for up to 6 months of work absence due to medical issues. We hypothesized that since brief short-term disability leave allows workers time to recover from illness or injury without losing their jobs, it should be protective against employment termination. Methods We analyzed 18 386 (83% male, 80% white) hourly employees. We censored workers once their accumulated disability leave exceeded 6 weeks because longer time spent on short-term disability leave suggests more serious illness or injury that may prevent return to work. To analyze the effect of short-term disability leave on employment termination, we applied a marginal structural pooled logistic model that allowed for a time-varying hazard function. We adjusted for time-varying confounding by occupational exposures and health-related variables using inverse probability weighting. Using the estimated coefficients, we compared the predicted probabilities (by person-month) of terminating employment with the corresponding counterfactual probabilities if the worker had never taken disability leave. These probabilities yielded estimated survival curves under the two scenarios. Results The average worker was followed for 5.5 years. Approximately 42% of the workers took at least one day of disability leave, and 48% terminated employment during follow-up. We estimated that 1058 (29%) more workers would have terminated employment within 5 years from cohort entry if the company had had no disability leave benefit than were predicted under the natural course. Conclusion Short-term disability leave is a potentially relevant health variable for occupational epidemiologists. This analysis suggests that short-term disability leave can help employees retain their jobs when a temporary health issue prevents them from working.

  PublisherOA PDFDOI

• 0415 Working around the clock: an exposure response relationship between night work and incident hypertension

  2017-08-01

  articleOpen accessSenior author

  Discussion Both those taking IHR and MS showed a slight improvement in this aspect of lung function during the course of their careers, generally supporting earlier findings of West Sussex and London firefighters.

  PublisherOA PDFDOI

• 0190 Occupational exposure to crystalline silica and death from lung cancer: g-estimation of structural accelerated failure time models

  2017-08-01

  articleOpen access

  <h3>Background</h3> Occupational exposure to crystalline silica has been linked to mortality risk for lung cancer. The healthy worker survivor effect may bias effect estimates downward unless special methods are used to adjust for employment status, time off work, and co-exposures. Unlike traditional regression, g-estimation adjusts correctly for such time-varying confounders. <h3>Methods</h3> We applied g-estimation of structural accelerated failure time models to estimate the number of years of life that could have been saved for mortality from all natural causes and from lung cancer if exposure to crystalline silica had been prevented among 2342 white male workers in the diatomaceous earth industry (1942–2011). Exposures were lagged 17 years because exposure data were only available through 1994; this also accounts for disease latency. Analyses adjusted for calendar year, age, Latino ethnicity, smoking status, duration of employment in the diatomaceous earth industry and exposure to crystalline silica before entering follow-up, prior exposure to crystalline silica, prior cumulative exposure to dust and asbestos, time taken off work, and employment status. <h3>Results</h3> If all workers had been unexposed to crystalline silica, we estimated that workers who died of natural causes would have survived, on average, 1.1 years longer (95% CI: 0.3, 2.3) overall. Workers who died from lung cancer would have survived an estimated average of 9.0 years longer (95% CI: 4.4, 16.2) if they had been unexposed. <h3>Conclusions</h3> A ban on exposure to crystalline silica in this cohort would have resulted in longer survival for workers, particularly those who died of lung cancer.

  PublisherOA PDFDOI

• 0137 Exposure-lag-response in occupational epidemiology: application of distributed non-linear lag models in a cohort of diatomaceous earth workers exposed to crystalline silica

  2017-08-01

  articleOpen access

  Occupational exposures extending over a long working life can have complex relationships with health outcomes, as timing, duration, and intensity of exposure are all potentially relevant. Simple measures of cumulative, or average intensity of exposure typically considered in occupational studies may not fully capture these relationships. We applied distributed non-linear lag models to examine the association of crystalline silica exposures with mortality from lung cancer and non-malignant respiratory disease. We fitted Cox proportional hazard models for each cause of interest to data from a cohort study of 2342 California diatomaceous earth workers exposed to crystalline silica. Our models combined various functions for exposure-response and lag-response including linear, piece-wise constant and spline functions. Models with a spline function for exposure-response and a constant term for the lag-response appeared to have the best fit for lung cancer, while models with spline functions for both exposure-response and lag-response had the best fit for non-malignant respiratory disease. Hazard ratios (HR) from these best fitting models corresponding to average daily exposures of 275 µg/m3 during lag years 11–40 prior to the age of observed cases were 1.96 (95% confidence interval (CI) 0.95–4.06) and 2.01 (95% CI: 1.02–3.97) for the two outcomes respectively. HRs from simple models with linear exposure-response and constant lag-response terms for the same exposure scenario were 1.15 (95% CI: 0.88–1.49) and 1.21 (95% CI: 1.01–1.44) respectively. Occupational studies of longitudinal cohorts with detailed exposure histories could benefit from methods allowing for non-linearities and the disentanglement of intensity, duration and timing of exposure.

  PublisherOA PDFDOI

• 0140 Breast cancer incidence and metalworking fluid exposure in a cohort of female autoworkers

  2017-08-01 · 1 citations

  articleOpen accessSenior author

  Breast cancer is the leading cancer diagnosed among women and environmental studies have produced few leads on modifiable risk factors. Following an Institute of Medicine recommendation for occupational studies of highly exposed women, we took advantage of an existing cohort of 4503 female hourly autoworkers exposed to metalworking fluid (MWF), complex mixtures of oils and chemicals widely used in metal manufacturing worldwide. Cox proportional hazards models were fit to estimate hazard ratios (HR) for incident breast cancer and cumulative exposure (20 year lag) to straight mineral oils (a known human carcinogen), and water-based soluble and synthetic MWF. Because the state cancer registry began in 1985, decades after the cohort was defined, we restricted analyses to sub-cohorts hired closer to the start of cancer follow-up. Among those hired after 1969, the HR associated with an increase of one interquartile range in straight MWF exposure was 1.13 (95% confidence interval: 1.03, 1.23). In separate analyses of premenopausal breast cancer, as defined by age at diagnosis, the HR was elevated for exposure to synthetic MWF, chemical lubricants with no oil content, suggesting a different mechanism for the younger cases. This study adds to the limited literature regarding quantitative chemical exposures and breast cancer risk.

  PublisherOA PDFDOI

• 0344 Ischaemic heart disease mortality, diesel exhaust, and respirable particulate matter exposure in the diesel exhaust in miners study (dems)

  2017-08-01 · 1 citations

  articleOpen access

  We examined the impact of exposure among non-metal miners to respirable elemental carbon (REC), a diesel exhaust surrogate, and respirable particulate matter from mine and ore dust (RPM), on ischaemic heart disease (IHD) mortality in the Diesel Exhaust in Miners Study (DEMS). DEMS was conducted by National Cancer Institute and National Institute for Occupational Safety and Health. Among males at 8 US mines, we estimated IHD mortality hazard ratios (HR) for cumulative exposure and for average intensity to REC and RPM among the 10 070 miners hired since dieselization. In addition, we employed the parametric g-formula to assess the impact of hypothetical REC and RPM interventions on IHD mortality adjusting for time-varying employment status to address healthy worker survivor bias. The HR (95% confidence interval (CI)) for the highest category versus lowest category of exposure were 1.18 (0.56, 2.24) for cumulative REC, 1.25 (0.78, 2.01) for cumulative RPM, 0.75 (0.39, 1.44) for average REC, and 2.58 (1.26, 5.28) for average RPM. Using the parametric g-formula, we estimated the cumulative risk under a hypothetical intervention where annual average daily exposures to REC is set to 0 and a joint intervention consistent with REC and RPM exposure limits of 0 and 0.5 mg/m³ respectively. The ratios comparing the risk under the intervention on REC alone and for the joint intervention, each compared to the observed risk, were 0.86 (0.62, 1.17) and 0.84 (0.71, 0.98) respectively. Our study indicates that exposure to REC and PM may increase IHD mortality among workers in this cohort.

  PublisherOA PDFDOI

Recent grants

• Occupational Exposure to PM2.5 and Cardiovascular Disease(CVD)

  NIH · $4.7M · 2011–2020

• Disease, Disability and Death in an Aging Workforce

  NIH · $12.9M · 2006–2024

Frequent coauthors

• Sadie Costello

  University of California, Berkeley

  31 shared

• Andreas M. Neophytou

  Colorado State University

  24 shared

• Sally Picciotto

  Berkeley Public Health Division

  22 shared

• Daniel Brown

  20 shared

• Mark R. Cullen

  11 shared

• Katie M. Applebaum

  George Washington University

  10 shared

• Elizabeth M. Noth

  University of California, Berkeley

  10 shared

• John R. Balmes

  University of California, Berkeley

  10 shared