J. Morel Symons

Evaluating uncertainty to strengthen epidemiologic data for use in human health risk assessments.

Background: There is a recognized need to improve the application of epidemiologic data in human health risk assessment especially for understanding and characterizing risks from environmental and occupational exposures. Although there is uncertainty associated with the results of most epidemiologic studies, techniques exist to characterize uncertainty that can be applied to improve weight-of-evidence evaluations and risk characterization efforts. Methods: This report derives from a Health and Environmental Sciences Institute (HESI) workshop held in Research Triangle Park, North Carolina, to discuss the utility of using epidemiologic data in risk assessments, including the use of advanced analytic methods to address sources of uncertainty. Epidemiologists, toxicologists, and risk assessors from academia, government, and industry convened to discuss uncertainty, exposure assessment, and application of analytic methods to address these challenges. Synthesis: Several recommendations emerged to help improve the utility of epidemiologic data in risk assessment. For example, improved characterization of uncertainty is needed to allow risk assessors to quantitatively assess potential sources of bias. Data are needed to facilitate this quantitative analysis, and interdisciplinary approaches will help ensure that sufficient information is collected for a thorough uncertainty evaluation. Advanced analytic methods and tools such as directed acyclic graphs (DAGs) and Bayesian statistical techniques can provide important insights and support interpretation of epidemiologic data. Conclusions: The discussions and recommendations from this workshop demonstrate that there are practical steps that the scientific community can adopt to strengthen epidemiologic data for decision making. Citation: Burns CJ, Wright JM, Pierson JB, Bateson TF, Burstyn I, Goldstein DA, Klaunig JE, Luben TJ, Mihlan G, Ritter L, Schnatter AR, Symons JM, Yi KD. 2014. Evaluating uncertainty to strengthen epidemiologic data for use in human health risk assessments. Environ Health Perspect 122:1160–1165; http://dx.doi.org/10.1289/ehp.1308062

Incomplete follow-up in the National Cancer Institute’s formaldehyde worker study and the impact on subsequent reanalyses and causal evaluations

Three of us (G.M., A.Y., and P.M.) performed reanalyses of the National Cancer Institute cohort study on nasopharyngeal cancer (NPC) risk among formaldehyde exposed workers (Hauptmann et al., 2004). Both reanalyses (Marsh and Youk, 2005; Marsh et al., 2007) were published in this journal. However, the mortality follow-up performed by the NCI working group reported in two publications by Hauptmann et al. (2003, 2004) was later stated to be incomplete (Beane Freeman et al., 2009a,b). This incomplete follow-up may impact the validity of the results of our reanalyses. At this time, corrected estimates for solid cancer mortality risks including NPC as reported in Hauptmann et al. (2004) have not been provided as an erratum by the authors or reported anywhere else to our knowledge. We would like to inform readers about these issues and ask for a prompt corrigendum of the 2004 publication by the NCI working group since this study has played such a prominent role in causal evaluations.

Cancer Risk Among Tetrafluoroethylene Synthesis and Polymerization Workers

Tetrafluoroethylene (TFE), a compound used for the production of fluorinated polymers including polytetrafluoroethylene, increases the incidence of liver and kidney cancers and leukemia in rats and mice. This is the first time the cancer risk in humans has been explored comprehensively in a cohort mortality study (1950-2008) that included all polytetrafluoroethylene production sites in Europe and North America at the time it was initiated. A job-exposure matrix (1950-2002) was developed for TFE and ammonium perfluoro-octanoate, a chemical used in the polymerization process. National reference rates were used to calculate standardized mortality ratios (SMRs) and 95% confidence intervals. Among 4,773 workers ever exposed to TFE, we found a lower rate of death from most causes, as well as increased risks for cancer of the liver (SMR = 1.27; 95% confidence interval: 0.55, 2.51; 8 deaths) and kidney (SMR = 1.44; 95% confidence interval: 0.69, 2.65; 10 deaths) and for leukemia (SMR = 1.48; 95% confidence interval: 0.77, 2.59; 12 deaths). A nonsignificant upward trend (P = 0.24) by cumulative exposure to TFE was observed for liver cancer. TFE and ammonium perfluoro-octanoate exposures were highly correlated, and therefore their separate effects could not be disentangled. This pattern of findings narrows the range of uncertainty on potential TFE carcinogenicity but cannot conclusively confirm or refute the hypothesis that TFE is carcinogenic to humans.

Mortality among workers exposed to acrylonitrile in fiber production: an update.

Objective: The investigation updates the mortality experience through 2002 for a cohort of workers exposed to acrylonitrile (AN). Methods: Standardized mortality ratios (SMR) were estimated based on two reference populations: the US population and a regional employee population. Exposure–response analyses were conducted using Cox regression models for cumulative and mean intensity exposure measures. Results: In the cohort of 2548 workers, 839 deaths have occurred with 91 deaths due to respiratory system cancer. Most standardized mortality ratio estimates are at or near no-effects levels. Hazard ratio (HR) estimates indicate no increased mortality risk for respiratory system cancer (adjusted HR = 0.96, 95% confidence interval: 0.74, 1.25). Conclusions: In summary, no mortality outcome of a priori interest, principally respiratory system cancer, is associated with increased AN exposure among fiber production workers over five decades of follow-up.

Carcinogenicity of Perfluoroalkyl Compounds

This chapter reviews the information available on the carcinogenic potential of perfluoroalkyl acids in both animals and humans. Historically, perfluorooctanoic acid (PFOA) and perfluorooctanesulfonate (PFOS) have been the most widely used members of this chemical class making these the subject of the largest proportion of the reported studies. Caution needs to be exercised in projecting the biological activities of any of the chemicals in this family based on results from others. For example, considering the three chemicals for which lifetime studies in rats are available, the outcomes were different with no increase in tumors seen with perfluorohexanoic acid (PFHxA), liver adenomas seen with PFOS, and adenomas of the liver, testis, and pancreas seen with PFOA. Mechanistic studies suggest that the liver tumors seen with PFOA reflect the activation of PPARα while the mechanism for tumor formation in the testis and pancreas is less clear. Epidemiologic studies have been reported for several levels of population exposure. Limited evidence of associations with kidney and testicular cancer has been reported in studies among community members exposed to drinking water contaminated by PFOA. Studies in workers exposed to higher levels of both PFOA and PFOS have not shown consistent evidence for an association with any specific cancer type. Studies in populations exposed to low levels of PFOA and PFOS have shown equivocal results for a variety of cancers with no consistent associations. Based on the evidence reported to date, the prospect for developing a carcinogenic outcome following exposure to PFOA and PFOS is remote. For other perfluoroalkyl acids, there is not sufficient evidence regarding their potential carcinogenicity. It should be noted that human exposures to these chemicals is currently quite low and appears to be decreasing.

0114 A Bayesian approach to account for the healthy worker selection effect

Objectives We propose a Bayesian method to adjust for the component of the healthy worker effect that arises from selection of healthier individuals into workforce to allow correct estimation of the standardised mortality ratio (SMR) and associated credible intervals. Method Information on general populations is typically used to generate expected counts for outcomes in SMR calculations but an occupational cohort is not a random sample of the general population. The alternative is to use the expected number of outcomes from industrial cohorts known to experience the outcome of interest but free of the exposures that defined the observed cohort. In Bayesian terms, we can view “expected counts of outcomes given the observed age-sex-period structure” as the target of inference for which we seek a posterior distribution. We show that the problem reduces to elucidation of a prior distribution: we propose using expert opinions about relative rates of mortality outcomes of interest in the observed cohort relative to general population rates and direct estimation of reference rates from occupational cohort studies. Results Data from DuPont on 320 000+ active and former employees with work histories in the US from 1955 will be used. This registry allows for the calculation of expected mortality counts using adjusted rates for national and regional DuPont worker populations. Robust specification of priors will be sought. Implementation of the calculations will be developed in common software. Conclusions We plan to develop a method for SMR calculation that accounts for the healthy worker selection effect both in the point estimate and uncertainty interval.

0112 Components of the Healthy Worker Effect with Quantification for Different Referent Comparisons

Objectives The healthy worker effect (HWE) is widely known to bias standardised risk estimates from occupational cohort studies. Multiple factors contribute to HWE bias that is commonly characterised as confounding due to the selection of individuals with “better health status” who are more likely to gain and retain employment relative to a general population including non-employed persons. Comparisons between standardised mortality ratios (SMRs) estimated from reference population rates with different characteristics allow for quantitative evaluation of different components of the HWE. Method Data from over five decades for a company-wide mortality registry comes from life insurance claims, and deaths are validated against the U. S. National Death Index. Average person-years at risk during five-year calendar periods for the occupational cohort population are estimated. The expected mortality counts are specific to age, sex, race, and calendar-time period strata. SMRs are calculated based on the mortality rates for the general U. S. population and the company-wide population. Results From 1956 through 2012, the annual US employee population has ranged from 29 000 to 108 000 workers. The mortality registry includes over 80 000 deaths validated through 2010, 25% due to malignant neoplasms and 37% due to cardiovascular diseases. Conclusions The HWE influences the interpretation of standardised estimates from occupational studies. Comparisons for different reference populations can evaluate differential HWE bias of associations between occupational exposure and mortality. Analyses based on company reference rates identify contributions from components of the HWE based on comparable demographic characteristics, a similar likelihood of obtaining and retaining employment, and an equivalent potential for ascertainment of mortality outcomes.