Ammonium perfluorooctanoate (CF3(CF2)6CO2−NH4+) is a thermally stable synthetic surfactant manufactured for use as a polymerization aid in fluoropolymers production.1 In the presence of biologic media, ammonium perfluorooctanoate rapidly dissociates to perfluorooctanoate (PFOA, CF3(CF2)6COO−) from the perfluorooctanoic acid2,3 and can be absorbed through inhalation, ingestion, and, to a lesser extent, dermal contact. PFOA can be formed from environmental and metabolic degradation of telomers.3
High human exposure to ammonium perfluorooctanoate occurs in occupational settings, where median serum PFOA levels in the range of 100–5000 ng/mL have been reported.4,5 PFOA is also a wide-spread environmental pollutant, with exposure to the general population arising directly through ammonium perfluorooctanoate manufacturing as well as through indirect pathways of exposure.6 In 1999–2000, the United States general population had an average serum PFOA concentration of approximately 5 ng/mL (parts per billion); this declined by 25% by 2003–2004.7 The geometric mean serum half-life of elimination of PFOA is estimated at 3.5 years (95% confidence interval = 3.0–4.1) and may be the consequence of a saturable renal resorption process in humans.8
Chronic ammonium perfluorooctanoate feeding studies of Sprague Dawley rats found an increased incidence of benign testicular Leydig cell tumors,9,10 and one study reported an increased incidence of hepatocellular and pancreatic acinar cell adenomas.9 Ammonium perfluorooctanoate is an agonist for the peroxisome proliferator activated receptor alpha11–13 and a number of other receptor agonists have been shown to produce liver tumors in rats.14 However, this pathway is generally considered of low relevance to humans.14 The Leydig cell and pancreatic acinar cell tumors observed are not common to all peroxisome proliferator activated receptor alpha agonists in the rat, and other modes of action have been proposed.2,15
Occupational exposure to ammonium perfluorooctanoate at a 3M Company manufacturing facility in Cottage Grove, Minnesota has been previously associated with mortality from prostate cancer,16 and cerebrovascular disease.17 We present an updated mortality analysis of this cohort to further evaluate potential associations between occupational exposure to ammonium perfluorooctanoate and specific causes of death. Updates from the original study16 include a more complete employment roster, additional years of follow-up, and a job exposure matrix that is specific for ammonium perfluorooctanoate exposure.
METHODS
The protocol for this study was reviewed and approved by the University of Minnesota Institutional Review Board.
Study Population
This cohort included employees of a 3M Company plant located in Cottage Grove, Minnesota, where ammonium perfluorooctanoate production began in 1947. This cohort differs from the previously published analysis16 by a longer period of enrollment (1997 versus 1983) and later follow-up (2002 versus 1989), and by inclusion criteria. The current eligibility criterion was a minimum of 365 days cumulative employment prior to 31 December 1997, while the earlier study required only 6 months of cumulative employment. This change was to exclude the relatively large number of short-term workers, many of whom were summer interns. We also located employment data on an additional 169 employees who were eligible for both studies. The original study identified 398 decedents while the updated study identifies 807.
Human resource records were abstracted for demographic information, including the worker's name, Social Security number, employee identification number, date of birth, and details of work history. Demographic information and vital status were verified using consumer credit reporting sources and the Social Security Administration service for epidemiologic research studies.
Mortality Assessment
The cohort was followed until 31 December 2002. Vital record searches were performed through the National Death Index for all cohort members not employed by the company on 31 December 2002 or not previously identified as deceased.16,17 The underlying cause of death was coded in the International Classification of Disease (ICD) revision in effect at the time of death.
Exposure Assessment
The goal of the exposure assessment was to classify jobs by exposure to ammonium perfluorooctanoate. We used work history records and expert historical knowledge of the manufacturing process to classify each job held by likelihood of exposure. An expert panel of veteran workers and plant industrial hygienists reviewed job titles and administrative department codes by year to determine where the perfluorochemical production, or the development of the perfluorochemical products, took place over the history of the facility. The available information permitted classification of jobs in the work histories into 3 general categories of ammonium perfluorooctanoate exposure.
- “Definite occupational exposure”: Primarily jobs where electrochemical fluorination, drying, shipping, packaging, and quality-control analyses of ammonium perfluorooctanoate occurred. Workers were exposed on a regular basis with potential for high exposure.
- “Probable occupational exposure”: Jobs in other chemical division areas where ammonium perfluorooctanoate exposure was possible, but likely lower or transient.
- “No or minimal occupational exposure”: Jobs primarily in the nonchemical division of the plant. Opportunity for some exposure (more than the general population) due to contamination at the work site.
Hereafter, these job exposure subgroups will be referred to as “definite ammonium perfluorooctanoate exposure,” “probable ammonium perfluorooctanoate exposure,” and “nonexposed.”
Exposure Classification for Analysis
We incorporated 2 approaches for characterizing ammonium perfluorooctanoate exposure in the analysis. The primary analysis is based on ever attaining a minimum time in jobs with probable or definite exposure. A secondary analysis used a cumulative exposure model with a weighted exposure based on duration of employment and qualitatively-specified exposure intensity.
Exposure by Job Classification
First, we characterized the mortality experience of workers compared with that of the general population of Minnesota with respect to ever working in jobs with definite exposure, ever working in probable exposure jobs but no definite exposure jobs, or working only in nonexposed jobs. Subsequently, a more restrictive classification was developed for an analysis using an internal referent population that classified the cohort members as (1) working in a “definite exposure” job for 6 months or more (high exposure), (2) working in a “definite exposure” job for less than 6 months, or never working a “definite exposure” job but ever working in a “probable exposure” job (moderate exposure), or (3) working only in jobs not exposed to fluorochemicals (low exposure). Entry into the first 2 categories could occur at varying points in the work history.
Cumulative Exposure
Comprehensive biologic monitoring data were not available for this cohort. Estimates of exposure intensity were limited to a qualitative assessment in the form of relative exposure weights assigned to the job exposure matrix. The exposure weights were derived, in part, from serum PFOA concentrations collected in 2000 from 131 employees in the chemical division of the plant. These data provided relative ranges of serum PFOA for selected areas of the plant. Areas where jobs were classified as having definite exposure had median serum PFOA levels ranging from 2.6 to 5.2 parts per million (ppm = μg/mL), and jobs classified with probable exposure had levels ranging from 0.3 to 1.5 ppm.18 No data were available for jobs in the nonexposed areas of the plant. Serum half-life of PFOA is approximately 3.5 years,8 and thus short-term peak exposures may equate to longer-term lower exposures over time. The initial cumulative exposure assigned weights of 1 in jobs with no exposure, 30 in jobs with probable exposure, and 100 in jobs with definite exposure. These weighting factors, while somewhat arbitrary, were chosen to reflect the relative exposure intensity of jobs and long biologic half-life of PFOA. We calculated cumulative exposure for each worker as a sum of the days of employment at each level, multiplied by the exposure weighting factor (weighted exposure level × days exposed), which provides a time-dependent exposure metric. The cumulative exposure was categorized into groups selected a priori, representing the equivalent of up to 1 year (36,499 exposure-days), 1–4.9 years (36,500–182,499 exposure-days) and 5 or more years (182,500 exposure-days) of employment in a job with definite exposure. Because the true form of the cumulative exposure model is unknown, we conducted a sensitivity analysis to explore how alternative weighting schemes may affect the results. The alternate weighting schemes were 1, 10, 50 and 1, 10, 100, which would limit the extent to which workers in jobs with “probable exposure” for longer periods would be classified with workers who held jobs with “definite exposure.”
Smoking History
Occupational medical records of the cohort members were abstracted for information on smoking habit: ever smoked regularly, year started smoking, number of years smoked, and cigarettes smoked per day. We classified cohort members by their smoking history and the availability of the records as follows: smoking history available, medical record available but no information about smoking, and medical record not available.
Wage Type
Baseline socioeconomic status is a well-accepted predictor of mortality. In this cohort, there are differences in educational attainment and income between hourly and salaried workers. To explore potentially confounding effects of these differences, we classified cohort members by wage type: hourly, salaried, or both. The last was designated if the job history included earning each type of wage for at least 365 days. A dichotomization of this covariate classified workers as hourly or salaried based on the predominant wage type.
Causes of Death of Interest
A priori causes of death of interest were cancers of the liver, pancreas, and testes and cirrhosis of the liver (selected based on results from toxicological studies2), and prostate cancer and cerebrovascular disease (CVD), (selected from prior analyses of this cohort16,17). Bladder cancer mortality was associated with perfluorooctanesulfonate (PFOS) in another occupational cohort,19 although subsequent research of incident cases offered little support for an association.20 We included ischemic heart disease as an a priori disease of interest due to the hypolipidemic effect of ammonium perfluorooctanoate in laboratory animals and the inconsistently reported and contradictory association of increased serum cholesterol levels in relation to PFOA biomonitoring data.4,5
Analysis
The mortality experience of the cohort was initially compared with the mortality rates for the state of Minnesota. We computed age-, sex-, and calendar-period-standardized mortality ratios (SMRs) and 95% confidence intervals (CIs) using the PC Life Table Analysis System.21 The all-cause and cause-specific SMRs were first computed for the full cohort and then for the exposure-specific categories and wage type.
To model the risk as a function of PFOA exposure using an internal referent population, we estimated hazard ratios (HRs) and 95% CIs were estimated with time-dependent Cox regression models.22 The time covariate was from date of entry into the cohort until death or end of follow-up. Exposure was characterized by job classification and cumulative exposure. We adjusted the models for sex and year of birth. Age at entry into the cohort, smoking status, and wage type were also examined as potential confounding covariates. To explore potential effects of latency, the exposure models were lagged by 10 years. The Cox regression analysis was conducted using the PHREG procedure in SAS 9.1.23
Because smoking data were unavailable for many of the cohort members, a multiple-imputation model was constructed using those with smoking data to predict the smoking status of those without smoking data.24 The predictors used for the imputation process were sex, year of birth, year of first employment at the facility, age at entry into the cohort, and wage type. We conducted all imputation procedures using the MI and MIANALYZE procedures in SAS 9.1.23 The imputed models were fit to further explore potential confounding by smoking status.
RESULTS
The cohort included 3993 employees, of whom 807 died in the follow-up period. The cohort was mostly male (80%), particularly in the “definite exposure” subgroup (92%) (Table 1). There was a higher prevalence of smoking in those who ever worked a job with definite ammonium perfluorooctanoate exposure (65%) compared with nonexposed workers (47%). However, smoking data were available for 66% of the definite-exposure subgroup, whereas it was available for only 20% of the nonexposed. A majority of the workers holding “definite exposure” jobs were hourly employees, while most nonexposed workers were salaried.
;)
TABLE 1: Characteristics of Ammonium Perfluorooctanoate Manufacturing Cohort by Job Exposure Subgroups
The all-cause and cause-specific SMRs were generally lower for the entire cohort and for exposure subgroups than for the general population of Minnesota (Table 2). (Results for all causes of death are presented in the online eTable, https://links.lww.com/EDE/A336). The few deaths from testicular cancer (0) and liver cancer (3) precluded further analysis for these causes of death. The number of deaths due to pancreatic cancer and cirrhosis of the liver were not more than expected for all exposure subgroups. The SMRs for cohort members ever employed in jobs with definite ammonium perfluorooctanoate exposure were elevated for prostate cancer and cerebrovascular disease, although confidence intervals are wide. By contrast, the number of deaths from prostate cancer and cerebrovascular disease were lower than expected among the never-exposed members of the cohort. Cohort members who worked in jobs with probable exposure, but who never held a job with definite exposure, had an elevated risk of death from diabetes mellitus. The number of deaths from ischemic heart disease was lower than expected.
TABLE 2: Standardized Mortality Ratios for Selected Causes by Ammonium Perfluorooctanoate Exposure Category
The SMRs for salaried workers (data not shown in tables) indicated a decreased risk of death for all cancers combined (SMR = 0.7 [95% CI = 0.6–0.8]), respiratory cancers (0.6 [0.4–0.9]), prostate cancer (0.5 [0.2–1.2]), diabetes (0.2 [0.02–0.7]), cerebrovascular disease (0.6 [0.4–1.0]), and heart disease (0.6 [0.5–0.7]). The results were somewhat different for hourly employees: all cancers combined (1.0 [0.9–1.2]), respiratory cancers (1.2 [0.9–1.6]), prostate cancer (0.9 [0.4–1.6]), cerebrovascular disease (0.7 [0.4–1.0]), and heart disease (0.9 [0.8, 1.1]). The SMR for diabetes (2.1 [1.3–3.1]) was elevated for the hourly workers.
In the time-dependent Cox regression models, moderate or high exposure work history, compared with working only in low-exposure jobs, was associated with an increased risk for prostate cancer and cerebrovascular disease. A work history of only moderate-exposure was associated with the risk of dying from diabetes mellitus (Table 3). Due to the rarity of some outcomes, the moderate and high exposure categories were combined; these are presented in Table 3 as well. Including wage type and smoking habit in the models did not alter the results. We further explored the models for prostate cancer, cerebrovascular disease, ischemic heart disease and diabetes by stratifying by wage type. There was no evidence that the observed associations were limited to either hourly or salaried workers. Lagging exposures by 10 years made unremarkable differences in the hazard ratio estimates.
TABLE 3: Hazard Ratios From Time-dependent Cox Regression Analysis for Cause-specific Mortality as a Function of Ammonium Perfluorooctanoate Exposure Characterized by Job Classification
Hazard ratios comparing the highest with the lowest cumulative exposure category indicated an increased risk for prostate cancer and cerebrovascular disease (Table 4). The results combining the 2 higher exposure categories are also presented. There was no association between exposure and risk of pancreatic or bladder cancer, cirrhosis of the liver, and diabetes. The risk of dying from ischemic heart disease was lower among those with increased exposure. The sensitivity analysis using alternate weighting schemes did not change the overall conclusions.
TABLE 4: Hazard Ratios From Time-dependent Cox Regression Analysis for Cause-specific Mortalities as a Function of Ammonium Perfluorooctanoate Exposure Characterized by Cumulative Exposure (Years)
DISCUSSION
We observed no association between ammonium perfluorooctanoate exposure and liver, pancreatic, and testicular cancer or cirrhosis of the liver. Exposure was associated with prostate cancer and cerebrovascular disease within the cohort but not when compared with the general population. Diabetes-related deaths were elevated among workers with moderate exposure.
Interpreting these results requires consideration of several limitations. Most notably, this is a relatively small cohort with limited power for studying deaths from rare diseases. However, it is one of very few occupational populations exposed to this chemical. The associations of ammonium perfluorooctanoate exposure with prostate cancer and cerebrovascular disease were apparent with the internal referent population. While an internal referent population may provide a more valid comparison (assuming similar social and demographic determinants of disease), the interpretation of this internal analysis should consider the stratum-specific prostate cancer and cerebrovascular disease SMRs. The SMRs for the exposed categories were modestly above unity, while the nonexposed members of the cohort were markedly below. This difference of the nonexposed and other men in Minnesota with respect to baseline prostate cancer and cerebrovascular disease risk may be related, in part, to socioeconomic status. Wage status was the only available proxy for socioeconomic status, which does not fully capture the complexities of socioeconomic status and its relation to health.
Our findings for the association between prostate cancer and work in an exposed job are similar to the results of Gilliland and Mandel,16 who analyzed the same population over a shorter period of follow-up. They reported (based on 6 cases) a 3.3-fold increase (95% CI = 1.0–10.6) in prostate cancer mortality associated with working 10 years in the chemical division compared with nonchemical division workers; only one of these workers was directly involved in the production of ammonium perfluorooctanoate.25 A cohort mortality study that included about half of the workers potentially exposed to ammonium perfluorooctanoate during the production of fluoropolymers did not report an elevated SMR for prostate cancer based on 3 referent populations.26 However, no exposure-specific estimates were provided. A prospective cohort study of cancer risk in the Danish general population reported no apparent association between prostate cancer risk and plasma levels of PFOA.27 It is important to note, however, that the mean plasma concentrations in this general population were 0.007 μg/mL compared mean exposures ranging from 0.3 to 5.2 μg/mL in this occupationally exposed population. The biologic mechanism for an association between PFOA and prostate cancer is not clear. There was no histologic evidence of prostate neoplasia associated with administered ammonium perfluorooctanoate doses of 0, 30, and 300 μg/kg in a 2-year chronic feeding study of Sprague Dawley rats.2 Doses of 0, 3, 10, and 30 (reduced to 20) mg/kg/day of ammonium perfluorooctanoate administered by oral capsule to male cynomolgus monkeys for 26 weeks resulted in prostate glands that were microscopically normal.28 Nevertheless, nongenotoxic mechanisms of carcinogenesis are possible. An effect of PFOA on the endocrine system in the rat has been described, involving the mode of action of Leydig cell tumors that might involve induction of CYP19A1 (aromatase), resulting in the conversion of testosterone to estradiol. In occupationally exposed populations, PFOA biomonitoring data were not clearly associated with changes in circulating levels of reproductive hormones.4,25 Inhibition of gap junction intracellular communication has also been associated with peroxisome proliferators such as PFOA.29
Deaths from heart disease and cerebrovascular disease are often below unity in epidemiologic studies of chemical workers,30 and thus our finding of an increased risk of cerebrovascular disease death associated with higher exposure was unexpected. The risk of stroke is related to diabetes, hypertension, and life-style factors, including diet and smoking.31–33 In this cohort, risks of death from life-style-associated diseases (eg, lung cancer, diabetes and heart disease) were not consistent across exposure groups. In the internal analysis, adjusting for smoking status and wage type did not alter the association between working in an ammonium perfluorooctanoate-exposed job and death from cerebrovascular disease. Diet is also a potential factor in the risk of stroke. In the same working population, body mass index (BMI)34 of the almost 50% of the workers for whom these data were available ranged from 25 to 30 kg/m2, which is considered overweight.35 However, the BMI distribution did not correlate with PFOA levels.34,36
Any findings in a mortality study related to diabetes should be interpreted with caution due to poor reporting of prevalent diabetes on death certificates.37 Leonard et al26 reported an elevated SMR for diabetes of a cohort of employees of a plant that manufactured ammonium perfluorooctanoate compared with other company workers in regional plants, but there was no association when compared with the general population. No specific estimates of PFOA exposure were made. A more comprehensive assessment of diabetes morbidity is required to fully evaluate any potential relationship with PFOA exposure.
In addition to the limits of mortality analyses characterizing diseases that do not uniformly cause death, the following limitations are acknowledged. Some exposure misclassification is unavoidable when using work history records. The extent of exposure misclassification and the effects on the study results remain unknown, as no additional data were available to further verify these assumptions. Although information on race was not available for the cohort, the impact is likely to be limited, as most residents of Minnesota over the decades have been white (97% in the 1980 census, 94% in 1990, and 89% in 2000).38 Our analysis considered potential confounding by age, sex, wage type, and (to some extent) smoking. The smoking data were sparse, and though sophisticated methods to impute the missing data were applied, the validity of these imputations is not clear. Finally, the mean age at follow-up was 60 years, and thus the relatively small number of deaths limits the ability of the study to examine exposure responses.
This study also has several notable strengths, including the complete enumeration of the cohort from employment records. A detailed review of the ammonium perfluorooctanoate production history by veteran workers and industrial hygienists was integrated with the biologic monitoring data, which helped to reduce exposure misclassification. Also, the comprehensive follow-up of the cohort found an underlying cause of death for 99.6% of the known deaths (804/807); all deaths with unknown causes were from cohort members who worked in the nonchemical division of the plant.
In summary, this study did not show ammonium perfluorooctanoate exposure to be associated with liver, pancreatic, and testicular cancer or cirrhosis of the liver. Elucidating the observed associations between exposure and prostate cancer, cerebrovascular disease, and diabetes will require study methods that include nonfatal cases.
ACKNOWLEDGMENTS
We thank Diane Kampa, Nancy Pengra, Allison Iwan, and Richard Hoffbeck for assistance with data management and analysis, and Harvey Checkoway and Jeffery Mandel for constructive comments on earlier versions of the manuscript.
REFERENCES
1. Begley TH, White K, Honigfort P, Twaroski ML, Neches R, Walker RA. Perfluorochemicals: potential sources of and migration from food packaging.
Food Addit Contam. 2005;22:1023–1031.
2. Kennedy GL, Butenhoff JL, Olsen GW, et al. The toxicology of perfluorooctanoate.
Crit Rev Toxicol. 2004;34:351–384.
3. Lau C, Anitole K, Hodes C, et al. Perfluoroalkyl acids: a review of monitoring and toxicological findings.
Toxicol Sci. 2007;99:366–394.
4. Olsen GW, Zobel LR. Assessment of lipid, hepatic, and thyroid parameters with serum perfluorooctanoate (PFOA) concentrations in fluorochemical production workers.
Int Arch Occup Environ Health. 2007;81:231–246.
5. Sakr CJ, Leonard RC, Kreckmann KH, Slade MD, Cullen MR. Longitudinal study of serum lipids and liver enzymes in workers with occupational exposure to ammonium perfluorooctanoate.
J Occup Environ Med. 2007;49:872–879.
6. Houde M, Martin JW, Letcher RJ, Solomon KR, Muir DC. Biological monitoring of polyfluoroalkyl substances: A review.
Environ Sci Technol. 2006;40:3463–3473.
7. Calafat AM, Wong LY, Kuklenyik Z, Reidy JA, Needham LL. Polyfluoroalkyl chemicals in the US population: data from the National Health and Nutrition Examination Survey (NHANES) 2003–2004 and comparisons with NHANES 1999–2000.
Environ Health Perspect. 2007;115:1596–1602.
8. Olsen GW, Burris JM, Ehresman DJ, et al. Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical production workers.
Environ Health Perspect. 2007;115:1298–1305.
9. Biegel LB, Hurtt ME, Frame SR, O'Connor JC, Cook JC. Mechanisms of extrahepatic tumor induction by peroxisome proliferators in male CD rats.
Toxicol Sci. 2001;60:40–55.
10. Riker.
Two Year Oral (Diet) Toxicity/Carcinogenicity Study of Fluorochemical FC-143 in Rats. Saint Paul, MN: Riker Laboratories, Inc; 1987. Experiment No. 0281CR0012 US EPA Docket AR-226–0437.
11. Maloney EK, Waxman DJ. trans-Activation of PPARalpha and PPARgamma by structurally diverse environmental chemicals.
Toxicol Appl Pharmacol. 1999;161:209–218.
12. Takacs ML, Abbott BD. Activation of mouse and human peroxisome proliferator-activated receptors (alpha, beta/delta, gamma) by perfluorooctanoic acid and perfluorooctane sulfonate.
Toxicol Sci. 2007;95:108–117.
13. Vanden Heuvel JP, Thompson JT, Frame SR, Gillies PJ. Differential activation of nuclear receptors by perfluorinated fatty acid analogs and natural fatty acids: a comparison of human, mouse, and rat peroxisome proliferator-activated receptor-alpha, -beta, and -gamma, liver X receptor-beta, and retinoid X receptor-alpha.
Toxicol Sci. 2006;92:476–489.
14. Klaunig JE, Babich MA, Baetcke KP, et al. PPARalpha agonist-induced rodent tumors: modes of action and human relevance.
Crit Rev Toxicol. 2003;33:655–780.
15. Liu RC, Hurtt ME, Cook JC, Biegel LB. Effect of the peroxisome proliferator, ammonium perfluorooctanoate (C8), on hepatic aromatase activity in adult male Crl:CD BR (CD) rats.
Fund Appl Toxicol. 1996;30:220–228.
16. Gilliland FD, Mandel JS. Mortality among employees of a perfluorooctanoic acid production plant.
J Occup Med. 1993;35:950–954.
17. Alexander BH.
Mortality Study of Workers Employed at the 3M Cottage Grove Facility. Minneapolis, MN: University of Minnesota; 2001. US EPA Docket AR-226–1030a018.
18. Olsen GW, Butenhoff JL, Mandel JH.
Assessment of Lipid, Hepatic, and Thyroid Function in Relation to an Occupational Biologic Limit Value for Perfluorooctanoate. Saint Paul, MN: 3M Company; 2003. USEPA Public Docket AR-226–1351.
19. Alexander BH, Olsen GW, Burris JM, Mandel JH, Mandel JS. Mortality of employees of a perfluorooctanesulphonyl fluoride manufacturing facility.
Occup Environ Med. 2003;60:722–729.
20. Alexander BH, Olsen GW. Bladder cancer in perfluorooctanesulfonyl fluoride manufacturing workers.
Ann Epidemiol. 2007;17:471–478.
21. National Institute for Occupational Safety and Health.
PC LTAS: Life Table Analysis System for Use on the PC. Cincinnati, OH: US Department of Health and Human Services; 1998.
22. Breslow NE, Day NE.
Statistical Methods in Cancer Research. Vol. 2: the design and analysis of cohort studies. Lyon: International Agency for Research on Cancer; 1987.
23. SAS Institute Inc. SAS System for Windows Version 9.1 SAS Institute. Cary, NC 2003.
24. Allison PD.
Missing Data. Quantitative Applications in the Social Sciences. Vol. 136. Thousand Oaks, CA: Sage Publications; 2001.
25. Olsen GW, Gilliland FD, Burlew MM, Burris JM, Mandel JS, Mandel JH. An epidemiologic investigation of reproductive hormones in men with occupational exposure to perfluorooctanoic acid.
J Occup Environ Med. 1998;40:614–622.
26. Leonard RC, Kreckmann KH, Sakr CJ, Symons JM. Retrospective cohort mortality study of workers in a polymer production plant including a reference population of regional workers.
Ann Epidemiol. 2008;18:15–22.
27. Eriksen KT, Sorensen M, McLaughlin JK, et al. Perfluorooctanoate and perfluorooctanesulfonate plasma levels and risk of cancer in the general Danish population.
J Natl Cancer Inst. 2009;101:605–609.
28. Seacat AM, Thomford PJ, Hansen KJ, Olsen GW, Case MT, Butenhoff JL. Subchronic toxicity studies on perfluorooctanesulfonate potassium salt in cynomolgus monkeys.
Toxicol Sci. 2002;68:249–264.
29. Upham BL, Deocampo ND, Wurl B, Trosko JE. Inhibition of gap junctional intercellular communication by perfluorinated fatty acids is dependent on the chain length of the fluorinated tail.
Int J Cancer. 1998;78:491–495.
30. Greenberg RS, Mandel JS, Pastides H, Britton NL, Rudenko L, Starr TB. Chemical workers in the US and Western Europe: a meta-analysis of cohort studies describing mortality and cancer incidence.
Epidemiology. 2001;12:727–740.
31. Desai J, Devlin H. 2002. Diabetes in Minnesota. Minnesota Department of Health. Available at:
http://www.health.state.mn.us/diabetes/diabetesinminnesota.
32. Gorelick PB, Sacco RL, Smith DB, et al. Prevention of a first stroke: a review of guidelines and a multidisciplinary consensus statement from the National Stroke Association.
JAMA. 1999;281:1112–1120.
33. Kuller LH. Epidemiology and prevention of stroke, now and in the future.
Epidemiol Rev. 2000;22:14–17.
34. Gilliland FD, Mandel JS. Serum perfluorooctanoic acid and hepatic enzymes, lipoproteins, and cholesterol: a study of occupationally exposed men.
Am J Ind Med. 1996;29:560–568.
35. CDC.
BMI: Body Mass Index. Atlanta, GA: Centers for Disease Control and Prevention; 2006.
36. Olsen GW, Burris JM, Burlew MM, Mandel JH. Plasma cholecystokinin and hepatic enzymes, cholesterol and lipoproteins in ammonium perfluorooctanoate production workers.
Drug Chem Toxicol. 2000;23:603–620.
37. Bild DE, Stevenson JM. Frequency of recording of diabetes on US death certificates: analysis of the 1986 National Mortality Followback Survey.
J Clin Epidemiol. 1992;45:275–281.
