Cohort Profile: The American Manufacturing Cohort (AMC) study

Why was the cohort set up?

The American Manufacturing Cohort (AMC) database consists of multiple, linked datasets that collectively provide an invaluable resource for understanding work-life exposures, health and economic outcomes in a cohort of US light and specialty metalworkers. The study data are the product of an academic-corporate partnership forged in 1997 between Alcoa, Inc., and M.R.C. (then Chief Medical Officer for Alcoa and also Director of the Occupational and Environmental Medicine Program at the Yale University School of Medicine) which continued until Alcoa separated into two distinct companies in 2016. During the period of active partnership, Alcoa worked with M.R.C. to incorporate research findings into company policy in an effort to prioritize worker safety.

The first papers on the health and safety of Alcoa workers produced by the present research Consortium were published in 2006. Early research focused primarily on traditional occupational health and safety questions, including those related to job-level chemical and physical exposures, and work hours in relation to ischaemic heart disease and individual-level biometrics, such as pulmonary function and body mass index (BMI). Later, research funded primarily by the National Institute on Aging (NIA) and the National Institute for Occupational Safety and Health (NIOSH) examined the social and economic determinants of chronic disease more broadly. By 2015, study personnel had developed sufficient infrastructure—including a secure, remote computing environment and detailed data documentation—which can support new users outside the core research Consortium at Yale, Stanford and Berkeley.

Although the formal working partnership with Alcoa has ended, the AMC database continues to expand through external linkages with data from the National Death Index (NDI), Centers for Medicare and Medicaid Services (CMS), Internal Revenue Service (IRS) and Census databases (IPUMS) at the level of individual workers. These linkages offer opportunities to better characterize the determinants of disease, disability and mortality with an emphasis on the social, economic and physical exposures related to the workplace.

Who is in the cohort?

Study follow-up spans from 1996 to 2013, and includes facilities originally owned by Alcoa, as well as several that were acquired either before or during the academic-corporate partnership period from other primary or fabricated metals product manufacturers (including Alumax, Howmet and Reynolds). Company personnel files included records for 142 257 active US workers employed for at least some time between 1996 and 2013, and 29 779 retirees. Medical claims data include an additional 65 613 spouses and 101 782 dependants from 2003 to 2013. Depending on the specific study question, investigators may focus their analysis to a subset of workers based on factors such as job type, facility size or type, duration of employment and choice of health plan. For example, several studies published to date on the health effects of fine particulate matter (PM_2.5) are restricted to a subset of 12 facilities for which PM_2.5 measurements were obtained for research, in addition to total particulate (TPM) which is more broadly available.

Table 1 summarizes the number of facilities, workers and basic demographic characteristics for active workers employed between 1996 and 2013 across: all US facilities; all US facilities with 250 or more workers; and the 12 US facilities with PM_2.5 measurements. We depict the number of active workers employed across all US facilities for each year of follow-up in Figure 1.

Table 1.

Demographic characteristics for workers employed across US facilities with 50 more workers and with 250 or more workers, and in the 12 facilities with PM_2.5 measurements^a

	All US facilitiesc,d	Facilities with ≥250 active workerse	Facilities selected for PM2.5 assessment
Facilities, n	675	102	12
Workers, n	142257	113849	25182
Employment type, n (%)b
Hourly	102223 (71.9)	87050 (76.4)	25182 (100.0)
Salaried	39917 (28.1)	26799 (23.6)	−
Male, n (%)	107747 (75.7)	86284 (75.8)	21244 (84.36)
Race/ethnicity
White	99225 (69.8)	81331 (73.7)	20210 (80.3)
Black	18286 (12.9)	15255 (13.8)	2554 (10.2)
Hispanic	12826 (9.0)	9523 (8.6)	2042 (8.1)
Other	5207 (3.7)	4253 (3.9)	268 (1.4)
Age, mean (SD)	41.4 (13.1)	41.5 (13.4)	41.3 (11.2)

SD, standard deviation.

^aIncludes all active workers between 1 January 1996 and 31 December 2013.

^bWorkers who were employed in both hourly and salaried jobs are classified here as hourly workers.

^cExludes employees with unknown location (n = 4500).

^dRace/ethnicity is missing for 6713 workers; employment type is missing for 117 workers; gender is missing for 46 workers, and age is missing for 529 workers.

^eEver more than 250 employed in any year between 1996 and 2013.

Figure 1.

The total number of active employees by gender across all US facilities (N = 675) for each year of follow-up from 1996 to 2013 by gender. Increases in the size of the workforce beginning in 2001 are attributable in part to acquisition of new facilities, and decreases in the size of the workforce after 2008 are attributable in part to the sale of facilities.

What has been measured?

The AMC Study database consists of several distinct datasets—human resources, medical claims, industrial hygiene assessments, expert ratings of job demand, real-time injury surveillance system and external datasets not derived from data initially provided by the company, among others—that can be linked at the individual level using a unique, encrypted identifier. Company personnel records include all active workers employed by the company between 1996 and 2013. The availability of data for each employee varies based on a number of factors (including job type, facility, duration of employment, and choice of health plan). Specific datasets included in the AMC database are detailed below and summarized in Table 2.

Table 2.

Brief description of data sources that comprise the AMC database

Data Source	Years	Number of workers	Population subset	Brief description	Limitations
Personnel records, payroll and timeclock data
Personnel file	1996–2013	142257	All workers active at any point between 1996 and 2013.	Includes basic employee demographics (birth date, hire date, sex, race/ethnicity) and changes in employee status and job title over time	Employee status and job title are missing for several entries or are recorded inconsistently across different facilities and years of follow-up
Payroll data	2001–14	84362	Active hourly workers	Provides the number of regular versus overtime hours per worker per pay period	Does not include payroll data for salaried workers
Timeclock data	1999–2014	42512	Available for workers at facilities that used smart-time (1999–2009) or work-brain (2010–14) software to log working hours.	Timeclock data capture specific hours and duration of work for hourly employees	Does not include salaried workers or data for hourly workers at facilities that did not use smart-time or work-brain to log working hours
Industrial hygiene measures
Job exposure matrices for PM and chemicals	1980–2014	25182	Available for the subset of workers employed at one of the 12 PM2.5 facilities between 1996 and 2014	Exposures for distinct exposure groups based on measurements from company industrial hygiene sampling database. JEMs are available for total particulate matter, PM2.5, oil mist, individual polycyclic aromatic hydrocarbons (PAHs), surrogates for PAHs (i.e. coal tar pitch volatiles), groupings of individual PAHs, individual and groups of metals, and fluorides. Also available are exposure profiles based on both major manufacturing processes and mix of chemical exposures	Available for workers at a limited number of US plants; PM2.5 measurements are averaged at the job level rather than individual level
Job Demand Survey
Job Demand Survey	2003	16117	Available for subset of workers employed one of the 11 facilities from 2000 to 2007 which participated in the Job Demand Survey in 2003	Expert ratings of physical and psychosocial job demand by job title completed in the autumn of 2003	Only cross-sectional data on job-demand from 2003 are available and have been used to characterize job demand among all workers at JDS facilities from 2000 to 2007
Audiometry, spirometry, biometrics, and incident management
Audiometry	1992–2015	79298	Workers enrolled in company hearing conservation programme	Pure tone air conduction audiometric threshold testing	Available only for the subset of workers in the company hearing conservation programme
Spirometry	1981–2015	36962	Workers required to undergo pulmonary function tests.	Includes: forced expiratory volume at 1 s (FEV1); forced vital capacity (FVC); and test date. These variables can be used to calculate lung function change over time	Available only for the subset of workers in the company respiratory protection programme
Biometrics	1996–2013	19108 (smoking) 22044 (blood lipids) 36786 (BMI) 37320 (blood pressure)	Workers employed at facilities targeted for medical chart abstraction	Includes: smoking status; blood lipid levels including cholesterol and triglycerides; height, weight and body mass index (BMI); and blood pressure (BP)	Available for only a subset of workers at targeted facilities with varying degrees of missingness
Incident Management	1996–2013	42610	All workers	Includes the date and location for all work-related incidents, including both first aid and serious (i.e. OSHA-reportable) incidents	Complete data available for US facilities from 1996 through 2010. However, from 2011 through 2013, these data are available for workers from a limited subset of US facilities
Medical claims and health risk score
Medical claims	1996–2013	100778	Available for workers, spouses and dependants insured by the company PPO who used their insurance at least once (1996–2002) or who received benefits for at least 1 month (2003–13)	Series of files which includes: hospital admissions; claims related to outpatient and emergency visits; prescription use; short- and long-term disability; and workers’ compensation. Detailed eligibility files for workers, dependants and spouses are available beginning in 2003	Eligiblity data are only available beginning 1 January 2003
Health risk score	1996–2015	125455	Available for workers, spouses and dependants enrolled in the company PPO for at least 1 month	Annual actuarial risk score based on health services provided. From 1996 to 2002, risk scores are available for workers only; beginning in 2003, risk scores are available for spouses and other dependants. Risk scores based on CMS data are available from 1999 to 2015	Available for all workers through 2002 but only for those insured through the company PPO from 2003 onward. Risk scores for spouses and dependaants are only available beginning in 2003
Externally linked data
IRS	2002–12	113643	All workers	Annual income reported through IRS form W2	Includes income from employers other than Alcoa
CMS	1999–2015	84718	Eligible workers age 50 or older who worked 1 day after 1 January 1996, and spouses of eligible workers after 1 January 2012.	Includes Medicare Pars A and B claims from 1999 to 2015 and Medicare Part D claims from 2006 to 2015	Excludes eligible workers and spouses who opted out of Medicare Plan B
NDI	1996–2011	5149	All workers	Includes dates and causes of death for workers identified as deceased. based on the SSDI and reported to SSA	Not all NDI records have been successfully matched. Analysis of mortality is only valid through July 2011 due to lags in reporting to SSA

Personnel records, payroll and timeclock data

Personnel records include employee demographics as well as time-varying information on facility location, employment type, job grade and job title. Records also provide information on work trajectories through date of first hire, changes in work status (i.e. layoff, retirement and disability) and termination, and can be used to construct analytical cohorts of active employees and patterns of work for a range of dates and facility locations. Related data that link to human resources files include W-2 wages since 2002, company-supported 401(k) savings data and information on residence. In addition to personnel records, measurements of regular versus overtime hours by pay period are available for all active, hourly workers (2001–14), and daily hours worked for a subset of facilities (2003–14).

Job exposure matrices for dust and chemicals

The company has collected industrial hygiene data for 60 years. Data from sampling conducted between 1980 and 2013 were compiled in an extensive database (>300 000 samples, over 200 hazards monitored). Samples were collected under the direction of certified industrial hygienists (IHs) and analysed at an accredited IH laboratory (Clark Laboratories LLC, Jefferson Hills, PA). Sampling was performed for combinations of job, task and hazard, based on the facility IH judgement that there was a greater than 5% likelihood of that exposure exceeding 30% of the standard. Extraction of these sampling data for assessment of chemical and physical hazards has been extensively reported.^1–10

At 12 facilities, information from the IH database was used to build job exposure matrices (JEMs) for particulate matter and selected chemicals. JEMs exist for exposure to two size fractions of particulate matter—TPM and PM_2.5. TPM exposures were calculated based on historical samples collected at 12 facilities (1980–2013). Paired PM_2.5 and TPM monitoring was collected in nine of these facilities (2010–11) to determine the percentage of TPM composed of PM_2.5 for each distinct exposure group. Occupational exposures generally exceed the EPA environmental PM_2.5 annual standards by an order of magnitude.

In addition to JEMs for PM, JEMs exist for individual chemicals and groups of chemicals that represent major exposures in the workplace (e.g. metalworking fluids and fluorides) and those that have been associated with heart disease (e.g. polycyclic aromatic hydrocarbons and welding-related metals). Exposure profiles were assigned by job, based on a combination of chemical exposures and major manufacturing process.

In order to reconstruct the job-exposure history of each individual in the epidemiological study, jobs judged to have qualitatively and quantitatively similar exposures were aggregated into distinct exposure groups (DEGs) and mappings were developed to reconcile job title/department combinations between the human resources and industrial hygiene databases.

Job Demand Survey

Job Demand Survey data consist of expert ratings of job-level physical and psychological demand, heat exposure and job control. Ratings were obtained in late 2003 from one senior health and safety professional at each of eight facility locations using a pilot job demand survey. Exposure ratings were assigned based on professional knowledge of job content rather than individual worker assessments. Overall physical job demand was rated using the U.S. Department of Labor classification scheme for physical demand characteristics of work. Raters classified each job into categories of Sedentary, Light, Medium, Heavy or Very Heavy. Psychological job demand and decision latitude (job control) were rated using questions similar to those used in the Whitehall II study,^11,12 based on the demand control/job strain model.^13,14 Heat exposure was rated using a four-point frequency scale ranging from 1 (not at all exposed) to 4 (exposed almost all the time).

Audiometry, spirometry, biometrics and injury

The AMC database contains audiometry, spirometry and biometric measurements for a subset of workers as well as data for all documented work-related incidents (Table 2). Audiometry measurements are available for workers enrolled in the company’s hearing conservation programme at facilities that purchased Occupational Health Management (OHM) software. Measurements consist of pure tone air conduction audiometric threshold testing results. Spirometry measurements were collected at least once every 3 years for employees exposed to respiratory irritants. Spirometry data include forced expiratory volume in 1 s (FEV1) and forced vital capacity (FVC). Together, these fields can be used to assess changes in lung function over time.

Biometric data include smoking status and measurements for height, weight, blood pressure, blood glucose level and blood lipids. These measurements were recorded in the OHM, as part of a corporate wellness initiative offering free biometric screenings for employees, and abstracted by researchers at Yale and Stanford Universities. The source of biometric measurements is indicated in the dataset.

Finally, injury data include the date and location of first aid and serious injuries (i.e. OSHA-reportable events) as catalogued through the company’s Incident Management System. This dataset includes over 90 000 events company-wide since 1996, and has been the subject of substantial scrutiny regarding timeliness and accuracy of data collection.

Medical claims

The company provided identical health insurance benefits to employees through local preferred provider organizations (PPO), subject to choices with respect to family coverage and deductible rates. A single firm managed its claims data. With the exception of a small number of workers who elected to receive coverage through a health maintenance organization (HMO), all claims are managed within the medical claims database.

From 1996 to 2002, the claims database includes all workers enrolled in the company PPO with at least one health insurance claim. From 2003 through 2013, complete monthly eligibility data are available for all workers enrolled in the company PPO for at least 1 month. The claims database includes monthly eligibility data for all covered workers, as well as linkable, detailed claims records for each inpatient and outpatient encounter and for each prescription filled. Claims data includes diagnostic codes from the International Classification of Diseases, Ninth Revision (ICD-9), and procedure codes from the Current Procedural Terminology (CPT). Medical claims data are available for dependent spouses and children of insured workers.

Health risk score

AMC data include an annual, continuous health risk stratification score for each worker, based on a third-party algorithm. The risk score is computed based on each beneficiary’s age, gender and use of health care services in a given calendar year, and is intended to predict future health expenditure. In these data, the score displays within-individual stability across time and is associated with multiple short-term health outcomes.¹⁵

Externally linked datasets

The AMC database includes linkages to several external sources of data. Linkages include data on: early life social context via the 1940 Census for employees born on or before that year in the USA; personal and family work experience and income across the life span, with post-employment health and mortality data through the Center for Medicare and Medicaid Services (CMS); and National Death Index (NDI) and contextual social environment information via residential geocoding. Linkage to Social Security Administration records to ascertain pre- and post-employment work history, as well as disability claiming behaviour and household total resources, is presently underway.

What has been found? Key findings and publications

To date, more than 80 articles using data from the AMC study have been published. Notable findings document:

• occupational health and safety including chemical (e.g. particulate matter, beryllium, fluoride, water-based metalworking fluid) and physical hazards (e.g. heat, noise and job demand); extensive studies for injury, noise exposure and hearing loss;

• health services research, including the impact of benefits changes over the several decades, most notably the switch from classic first-dollar coverage to deductible and co-pay insurance after 2003;

• application of G-methods to address the presence of time-dependent confounding in the study of work exposures and health;

• use of medical claims data in epidemiologica; research;

• health trajectories of workers from early life through employment, retirement and death;

• and studies of the population in relation to national economic and social events such as the recession and, more recently, the opioid epidemic.

A full list of publications can be found at [http://web.stanford.edu/group/amc].

Occupational health and safety

Several studies report on multiple risk factors for workplace injuries, including BMI, sex, comorbidities, ergonomic hazards, work hours and measures of job demand.^16–22 Examples of more novel findings include reports on the risk of hearing loss, hypertension and occupational injury in association with noise exposure, as well as the role of sex and gender differences in job status and health.^23–27 Claims data has also been leveraged to characterize work-related asthma³ and beryllium disease,^28,29 as well as PM_2.5 exposure and risk of ischaemic heart disease (IHD).^5,30–33

Application of G-methods in the context of time-dependent confounding

The AMC includes time-varying health data not usually available in occupational studies, which allow for the application of advanced methodological approaches to address issues of confounding affected by previous exposure in the form of the healthy worker survivor bias. Examples have included applications of marginal structural models³² as well as targeted minimum loss-based estimation³³ approaches to estimate the long-term effects of occupational PM_2.5 exposures on incidence of IHD. AMC is also unique in the availability of sufficient data for estimation of incident morbidity in a cohort of actively employed workers.

Use of medical claims data in epidemiological research

The AMC data have also been leveraged to assess the use of medical claims for epidemiological research on various health endpoints.^34,35 Of particular significance is the finding that algorithms initially developed to forecast and manage expenditures have been shown to be highly predictive of multiple health outcomes, including occupational injury, incident morbidity and mortality, as well as health care use and retirement.¹⁵ Claims data have also been used to assess effects of job demand on mental health.³⁶

Health trajectories from early life through employment, retirement and death

Several studies leverage the rich, longitudinal nature of the AMC data in order to examine health trajectories throughout the life course. One study found evidence that characteristics of workers’ early life state of residence—including income inequality, percentage non-White population and education—were associated with health outcomes such as hypertension, diabetes and IHD.³⁷ Several studies consider the importance of the workplace social environment and associated psychosocial hazards for workers’ subsequent health,^27,38 with a particular focus of the effects of the Great Recession on worker’s mental and physical health. Studies of the Great Recession indicate increased mental healthcare use, incident hypertension and diabetes, particularly among workers employed at facilities that experienced the most severe layoffs.^39–41

Finally, studies have examined the relationship between worker health and the transition to retirement. These studies suggest that increased job demand is associated with earlier retirement, offer evidence of concordance between chronic disease status before and after retirement and some evidence of null or beneficial effects of retirement on workers’ subsequent health.^42–44

Strengths and weaknesses of the AMC study database

The AMC database provides detailed, longitudinal records for an economically and geographically diverse occupational cohort. These data have helped investigators establish medical claims as a proxy for true health outcomes. Investigators have also incorporated proprietary risk scores to approximate time-varying overall health status and account for the healthy worker survivor effect (HWSE) in aetiological analyses. Finally, external linkages at the person, job, and facility levels have allowed us to characterize employment, health and economic trajectories for large swathes of the over 200 000 employees, their spouses and their dependants.

Despite notable strengths, several limitations are associated with AMC data. First, whereas medical claims data provide appropriate measurement of disease status for acute onset conditions for which workers are almost certain to seek care (i.e. stroke, myocardial infarction), measurement of underlying health status becomes imprecise for disorders with more indolent onset or for which services may be perceived as elective (i.e. mental health outcomes). Additionally, outcomes for many chronic disease endpoints remain limited. Until recently, this limitation has stymied work with cancer endpoints save for bladder cancer, for which workers in this industry are known to be at higher risk. Given the recent linkage to CMS, a large fraction of the cohort, including many for whom there were no earlier claims data, have claims data on later life, rendering cancer endpoints increasingly available for study.

Third, the availability of data for each worker varies, based on a number of factors (including job type, facility, duration of employment and choice of health plan). For example, measurements of exposure to PM_2.5 are currently available for a limited subset of facilities, and educational attainment is missing for nearly half of all employees. Complete health care eligibility data are only available for PPO beneficiaries beginning in 2003. Fourth, several key covariates of interest, including direct measures of household composition, previous work histories and biomarker data, are absent from the database altogether. Other covariates of interest—job type and educational attainment, for example—are sparsely populated and are missing for a large subset of workers. Finally, the study data may have limited generalizability as they describe employees at a single US firm in a single industry. The extent to which findings from the AMC database are generalizable to other working populations remains an area of ongoing investigation.

Nevertheless, AMC data represent a comprehensive and detailed longitudinal record of an employed, geographically diverse, universally insured US population. The ultimate contribution of this database is that it enables not only the study of the aetiological effects of occupational hazards on worker health, but also the investigation of social and economic determinants of health in a working population throughout the life course.

Can I gain access to the data? Where can I find out more?

Presently, AMC Study data are available as a supported access resource. Based on the complexities described here, orientation and initial feasibility analyses are required for all new researchers. Members of the AMC Research Consortium (Appendix 1) review initial proposals for feasibility as well as overlap with existing Consortium research initiatives and published work to date. Once initial proposals are reviewed and approved, access to AMC data is available to qualified researchers contingent upon the following:

• documented approval of research question and protocol from an institutional review board;

• completion of human subjects research ethics training;

• engagement in a Data Use Agreement with the Stanford Center for Population Health Sciences (SPHS), which includes conducting all analyses via VPN to a secure server environment maintained by Stanford University;

• sharing of research output using AMC data for review before submission, for the purpose of accuracy and attribution;

• and citation of the Stanford Center for Population Health Sciences as a data source.

As part of the ongoing effort to improve this database for research use, we encourage those accessing data to produce appropriately detailed information about new variables, intermediate datasets and additional supportive code at the close of the project. Currently, there are no charges associated with data access. Analytical support, if needed, is contingent on the staff resources available, and related fees will be determined based on the estimated effort required to complete the project. To enquire about working with AMC data and the associated Consortium, please e-mail the Stanford Center for Population Health Sciences Data Core team at [phsdatacore@stanford.edu].

Profile in a nutshell

• The AMC database comprises multiple, linked datasets that collectively provide a resource for associations between work-life exposures and health and economic outcomes in a cohort of US light and specialty metalworkers over the life course.

• Study follow-up spans from 1996 to 2013. As of 1 January 2003, company personnel files include records for 49 641 active US workers, 28 458 spouses, 44 424 dependants and 25 071 retirees.

• The database comprises a wide range of measures extracted from human resources, medical claims, company exposure and job demand assessments, injury surveillance, biometrics, health risk scores and linkages to external datasets not derived from data initially provided by the company.

Funding

This research was supported by: the National Institute of Health National Institute on Aging grant R01-AG026291; the National Institute for Occupational Health and Safety grant R01-OH009939; and by the National Institute of Mental Health grant F31-MH112246. The conclusions expressed are solely those of the authors.

Conflict of interest: None declared.

Appendix 1.

Active AMC Consortium members and current affiliations

Name	Division and School	Institution
Cabral, Marika	Department of Economics	University of Texas at Austin
Cantley, Linda F	Occupational and Environmental Medicine Program	Yale University School of Medicine
Combs, Mary	Division of Biostatistics	University of California, Berkeley School of Public Health
Costello, Sadie	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Cullen, Mark R	Center for Population Health Sciences	Stanford University School of Medicine
Dufault, Suzanne	Division of Biostatistics	University of California, Berkeley School of Public Health
Einav, Liran	Department of Economics	Stanford University
Eisen, Ellen A	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Eisenberg, Michael	Department of Urology	Stanford University School of Medicine
Elser, Holly	Center for Population Health Sciences;	Stanford University School of Medicine;
	Division of Epidemiology	University of California, Berkeley School of Public Health
Falconi, April	Center for Population Health Sciences	Stanford University School of Medicine
Ferguson, Jacqueline	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Galusha, Deron	Occupational and Environmental Medicine Program	Yale University School of Medicine
Hamad, Rita	Department of Family Community Medicine	University of California, San Francisco
Hammond, S Katharine	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Harrati, Amal	Center for Population Health Sciences	Stanford University School of Medicine
Liu, Sa	Department of Occupational Health Sciences	Purdue University School of Health Sciences
Lutzker, Elizabeth	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Meausoone, Valerie	Center for Population Health Sciences	Stanford University School of Medicine
Modrek, Sepideh	Health Equity institute	San Francisco State University
Neitzel, Richard	Department of Environmental Health Sciences	University of Michigan School of Public Health
Neophytou, Andreas M	Department of Environmental and Radiological Health Sciences; Division of Environmental Health Sciences	Colorado State University;University of California, Berkeley School of Public Health
Noth, Elizabeth M	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Picciotto, Sally	Division of Environmental Health Sciences	University of California, Berkeley School of Public Health
Rabinowitz, Peter	Department of Environmental and Occupational Health Sciences	University of Washington School of Public Health
Rehkopf, David	Division of Primary Care and Population Health	Stanford University School of Medicine
Tamang, Suzanne	Department of Biomedical Data Science	Stanford University School of Medicine
Tribett, Erika	Center for Population Health Sciences	Stanford University School of Medicine