Document RpaYjNger57oGoMxpOq60zKJa

Corporate Occupational Medicine 3M Center, Building 220-3W-05 St. Paul, MN 55144-1000 651 737 4230 Telephone 651 733 9066 Fax AR1K-OH72 4. Mortality Among Employees of a Perfluorooctanoic Acid Production Plant This published paper (Gilliland and Mandel; J Occup Med 1993;950-954) is the second update o f the original retrospective cohort mortality study (see studies # 2 and #3). This mortality analysis is the second-half o f the doctoral dissertation by Gilliland (see study # 5). The cohort consisted o f 1339 male and 245 female workers employed between 1947 and 1983 in the chemical division o f the 3M Cottage Grove (Minnesota) manufacturing site. Also analyzed for their mortality experience were 1449 male and 504 female employees who worked in the non-chemical division at this site. Vital status was ascertained through 1989. There were no significantly increased cause-specific standardized mortality ratios for either men or women. Ten years o f employment (or more) in the chemical division was associated with a 3.3 fold increase (95% Cl 1.0 10.6) in prostate cancer mortality compared to the non-chemical division. Gilliland and Mandel suggested the results be interpreted cautiously due to the fact that there were only four prostate cancer deaths in the chemical division cohort. Based on a subsequent review o f work history records, only one o f the four employees in the Chemical Division who died from prostate cancer appeared to have worked directly in the PFOA production building (Olsen etal JOEM 1998;40:614-622). This points out another limitation o f this study in that the entire work history records were not abstracted during the original study (see study # 2). This abstraction o f work history records was recently completed. An ongoing retrospective cohort mortality study will utilize this information to develop a 003166 more refined exposure matrix for PFOA. Vital status will be updated through 1997. A final report is scheduled for September 2000. C03167 950 Gilliland & Mandel Mortality in a PFOA Production Plant Mortality Am ong Employees of a Perfluorooctanoic Acid Production Plant Frank D. Gilliland, MD, PhD Jack S. Mandel, PhD, MPH Perfluorooctanoic acid (PFOA) and its salt, ammonium periluorooctanoate, are perfluorinated surfactants. Be cause of their unique surface active properties they are used in a large Perfluorooctanoic acid (PFOA) has been found at low levels (10 to 100 number of industrial applications and parts per billion) in sera o f the general population and at higher levels in consumer products including plasti occupationally exposed workers. Although PFOA has been reported to be a cizers, lubricants, wetting agents, and promoter o f rodent hepatocarcinogenesis and to alter reproductive hormones emulsifiers.'*3 Despite their wide in humans and rodents, there is little information on human health effects spread use, little is known about po associated with PFOA exposure. The present study exam ined the relationship tential adverse health effects. between PFOA and mortality using a retrospective cohort mortality design. PFOA induced marked hepatomeg The cohort consisted o f2788 m ale and 749fem ale workers employed between 1947 and 1983 at a plant that produced PFOA. The all-causes standardized mortality ratio was .75 (95% confidence interval [Cl], .56 to .99) fo r women and .77 (95% Cl, .69 to .86) fo r men. Among men the cardiovascular standardized mortality rate was .68 (95% Cl, .58 to .80) and the allgastrointestinal diseases was .57 (95% Cl, .29 to .99). There was no signifi cantly increased cause-specific standardized mortality ratio fo r either men or women. Ten years o f employment in exposedjobs was associated with a 3.3-fold increase (95% Cl, 1.02 to 10.6) in prostate cancer mortality com pared to no employment in PFOA production. There were only six prostate cancer deaths overall andfo u r am ong the exposed workers; thus, the results must be interpreted cautiously. I f prostate cancer mortality is related to PFOA. PFOA m ay increase prostate cancer mortality by altering reproductive hormones in male workers. aly and peroxisome proliferation in rodent livers.3** The chemically di verse group of xenobiotics that induce peroxisomes is of concern because of its association with nongenotoxic he patocarcinogenesis.3*10PFOA did not produce an increased number of hep atocellular carcinomas in a 2-year rat feeding study.* However, biphasic (initiation and promotion) and tri phasic (initiation, selection, and pro motion) hepatic carcinogenesis stud ies in rodents have shown significantly increased numbers of carcinomas in the PFOA-treated ra ts ." 12It has been suggested that the marked rodent hep atomegaly produced by PFOA is a marker for carcinogenic potential.13 The observations of increased Leydig cell tumors in a 2-year rat PFOA feed ing study and of disruption of the hypothalamic-pituitary-gonad axis in PFOA-treated rats* are consistent with the hypothesis that PFOA-associated tumors are mediated by a hormonal nongenotoxic mechanism. From the Division of Environmental and Occupational Health, School of Public Health, University of Minnesota, Minneapolis, Minnesota (Dr Gilliland, Dr Mandel); and Department of Internal Medicine, Occupational and Environmental Medicine Section, St Paul Ramsey Medical Center, St Paul, Minnesota (Dr Gilliland). Address correspondence to: Frank D. Gilliland, MD, University of New Mexico School of Medicine, New Mexico Tumor Registry, 900 Camino de Salud NE, Albuquerque, NM 87131. 0096- 1736/93/3J09-0950S03.00/0 PFOA has a long half-life in hu mans. A study of occupationally ex posed workers showed that the halflife in men is greater than 1.5 years.14 Hence, accumulation of PFOA may occur from small, frequent PFOA Copyright by American College of Occupational and Environmental Medicine doses. PFOA in the serum of the gen- G03168 v \ JOM Volume 35, Number 9, September 1993 BEST COPY AVA1I ABLE 951 oral populations of industrialized countries1'"1'' is likely to be the result of an accumulation of small PFOA doses. No health problems related to PFOA exposure were observed in a cross-sectional study among workers employed at the PFOA production plant.14 Cross-sectional studies of PFOA-exposed workers at this plant have shown that PFOA was associated with decreased free testosterone and increased estradiol.20 To determine whether mortality from any cause was associated with occupational exposure to PFOA, a ret rospective cohort mortality study was conducted at a plant that has pro duced PFOA since 1947. Methods The plant consists of several divi sions, with PFOA production re stricted to the Chemical Division. A number of other specialty chemicals have been produced in this division. The study cohort consisted of workers who were employed at the plant for at least 6 months between Jan 1, 1947, and Dec 31, 1983. Data were ab stracted from plant personnel records, which were maintained on all workers ever employed at the plant. Vital status was ascertained from the Social Security Administration for the period 1947 to 1982 and from the National Death Index for the period 1979 to 1989. All workers with un known vital status were traced using a variety of tracing strategies such as directory assistance, Metronet and TRW searches, reverse directories, motor vehicle registration lists, con tacting neighbors and relatives, and the post offices. Death certificates were obtained from the appropriate state health departments for those identified as, or presumed to be, de ceased. Information concerning the data and cause of two deaths which occurred outside the United States was obtained from family members. A nosologist coded the death certifi cates for underlying cause of death according to the International Classi fication of Diseases, 8th revision. The reliability of the coding was evaluated by resubmitting a random sample of death certificates for coding by the same nosologist. In the 25 death cer tificates from 1970 to 1989 resubmit ted to the nosologist for ICD coding, there were no changes in the major categories of cause of death. Workers were categorized as ex posed or unexposed to PFOA based on their job histories. Exposed work ers were defined as all workers em ployed for I month or more in the Chemical Division. Unexposed work ers were employees who either never worked in the Chemical Division or worked in the Chemical Division for less than 1 month. Cumulative expo sure to PFOA was estimated using the surrogate measure of months of Chemical Division employment. The observed numbers of causespecific deaths were compared to the expected numbers of deaths obtained by applying sex- and race-specific quinquennial age, calendar period, and cause-specific mortality rates for the United States and Minnesota pop ulations to the distribution of ob served person-time.2 Because less than 1% of plant employees were non white, white male and white female rates were used for comparison. For women, only United States rates were used because cause- and calendar period-specific Minnesota rates for women were not available. The effects of latency, duration of employment, and work in the Chemical Division were examined using stratified stand ardized mortality ratio (SMR) anal yses. Cause-specific mortality rates were compared between exposed and unexposed workers using stratified SMRs.23 SMRs were calculated for men based on US and Minnesota white male mortality rates for three latency intervals (10, 15, and 20 years) and three categories of duration of employment (5, 10, and 20 years). The SMRs were calculated using the program developed by Monson.22 The relative risk (RR) and 95% confidence interval (Cl) for deaths from all causes, cancer, cardiovascular diseases, and other selected causes were estimated using proportional hazard models.24,25 The time to event or censoring was defined as time from first employment to event or to De cember 31, 1989. In models for spe cific causes of death, deaths from other causes were censored at the time of death. Age at first employment, year of first employment, and dura tion of employment were included as covariates in the model. The analyses were stratified by gender. The appro priateness of the proportional hazard assumptions was tested using strati fied models with graphical analysis of log (-logfsurvivalj) versus follow-up time relationships and models that tested the significance of a product term between exposure and log fol low-up time.23,26 Proportional hazard calculations were conducted using SAS.23 Results A total of 3537 workers employed at the plant between Jan 1, 1947 and Dec 31, 1983 were identified from company records. Six workers who had incomplete employment records were excluded from the study. The cohort consisted of 2788 (79%) men and 749 (21%) women (Table 1). Men TABLE 1 Characteristics of Female and Male Employees. 1947-1989 Chem ical Non-Chem ical Division D iv isio n Number of workers Person-years of ob- servation Mean follow-up (y) Mean age at employ- ment (y) Mean year of death Mean year of death Mean age at death (y) F em ale 245 6029.0 24.6 28.8 1965.0 1981.3 58.7 M ale 1339 33385.3 24.8 25.6 1963.8 1978.3 54.2 Fem ale 504 13280.4 26.4 26.9 1962.8 1979.2 54.4 M ale 1449 37732.4 26.0 28.9 1962.3 1978.1 58.1 _. _______________ F em ale 749 19309.4 M ale 2788 71117.7 25.8 27.6 25.5 27.3 1963.5 1979.6 55.4 1963.0 1978.2 56.4 CG3169 952 Gilliland & Mandel Mortality in a PFOA Production Plant contributed 71.117.7 person-years of SMR was .36 (95% Cl. .07 to 1.05). tate cancer were 2.03 (95% Cl. .55 to observation, which were equally di The all-causes SMR for the non- 4.59) in the Chemical Division group vided between the Chemical Division Chemical Division women was .91 and .58 (95% Cl. .07 to 2.09) in the and non-Chemical Division. Women (95% Cl, .64 to 1.24) and the cancer non-Chemical Division cohort. In the contributed 19,309.4 person-years, SMR was .91 (95% Cl, .49 to 1.52) Chemical Division group, there were two-thirds of which were in the non- (data not shown). 4 observed and 2 expected deaths Chemical Division. Using Minnesota rates for compar from prostate cancer. There was no Vital status was obtained for 100% ison, the SMR for men for all causes, significant association between any of the cohort (Table 2). There were 50 for cardiovascular diseases, and for all cause of death and latency in either deaths among the women (11 in the gastrointestinal diseases was signifi exposure group. For the Chemical Di Chemical Division cohort and 39 in cantly less than 1 (Table 4). None of vision cohort, the prostate cancer the non-Chemical Division cohort) the cause-specific SMRs was large nor SMR was 1.61 (95% Cl, .32 to 4.70) and 348 deaths among the men (148 was any significantly different from 1. in the greater than 15-year latency deaths in the Chemical Division The results were similar when the ex group. group and 200 in the non-Chemical pected numbers of male deaths were Table 5 presents the final propor Division group). Death certificates based on US mortality rates. For the tional hazard model for all-causes, all were obtained for 99.5% of deaths. three latency intervals, the SMRs for cancer, and prostate-cancer mortality For women, the SMR for all causes deaths from all causes ranged from among the 2788 male workers em of death (SMR = .75; 95% Cl, .56 to .75 to .77. For all cancers, the SMRs ployed for more than 6 months. The .99) was significantly lower than ex ranged from 1.06 to 1.12 and were estimated relative risk for all-cause pected (Table 3). There was no asso nonsignificant. mortality for a 1-year increase in age ciation with duration of employment Among men, there was no associa at first employment was 1.08 (95% or latency for deaths from all causes, tion between any cause of death and Cl, 1.07 to 1.09). Year of first em cancer, and cardiovascular diseases duration of plant employment. The ployment and duration of employ (data not shown). Mortality among all-causes SMRs were .86 (95% Cl, ment were negatively associated with Chemical Division women was less .72 to 1.01) for the Chemical Division deaths from all causes. The risk asso than expected. In Chemical Division group and .69 (95% Cl, .59 to .79) for ciated with months employed in the women, the all-causes SMR was .46 the non-Chemical Division group Chemical Division was small and (95% Cl, .23 to .86) and the cancer (data not shown). The SMRs for pros- nonsignificant. In the final prostate cancer mortal ity model, length of employment in TABLE 2 Vital Status and Cause of Death Ascertainment among Female and Male Employees, 1947-1989 Chem ical Division Non-Chem ical Division Total the Chemical Division was positively and significantly associated with pros tate cancer risk. The relative risk for a 1-year increase in Chemical Divi sion employment time was 1.13 (95% Fem ale M ale Fem ale M ale Fem ale M ale Cl, 1.01 to 1.27) For 10 years' em AJrve Dead Total No. % No. % No. % No. % No. % No. % 234 95.3 1191 88.9 465 91.6 1249 86.2 699 93.3 2440 87.5 11 4.7 148 11.1 39 8.4 200 13.8 50 6.7 348 12.5 245 100.0 1339 100.0 504 100.0 1449 100.0 749 100.0 2788 100.0 ployment in the Chemical Division, the relative risk was estimated to be 3.3 (95% Cl, 1.02 to 10.6) compared with workers never employed in the Chemical Division. Age at first em ployment was positively associated TABLE 3 Observed (Obs) and Expected (Exp) Deaths, Standardized Mortality Ratios (SM R ) and 9 5 % Confidence Intervals (Cl) for 749 Female Employees C ause of Death_______ O bs_______ Exp SM R 95% Cl All causes Cancer 50 66.74 0.75 0.56-.99 17 23.04 0.71 0.42-1.14 with prostate cancer mortality. Length of time employed in the Chemical Division was not significantly related to mortality from lung cancer, gas trointestinal cancer, pancreatic can cer, or diabetes mellitus. Gastrointestinal 2 4.54 0.44 0.05-1.59 Respiratory Breast Genital Lymphopoietic Cardiovascular Cerebrovascular Gastrointestinal Injuries Suicide 4 4.72 0.95 0.26-2.43 3 5.87 0.51 0.10-1.49 2 3.37 0.59 0.07-2.14 3 2.04 1.47 0.30-4.29 10 12.39 0.81 0.49-1.29 3 3.51 0.86 0.01-4.80 3 3.41 0.88 0.18-2.57 4 6.23 0.64 0.17-1.64 1 1.78 0.56 0.01-3.13 Discussion This was the first retrospective co hort mortality study of workers em ployed in a PFOA production plant. Mortality from all causes in both men and women was significantly less than expected. Because of the healthy worker effect, internal comparisons C03170 JOM Volume 35, Number 9, September 1993 BEST COPY AVAII ABLE 953 TABLE 4 D eath s and Standirdized M ortality R atio s (S M R ) B a se d on M inn esota W hite M ale R ates, A m ong 278 8 M ale Em p lo ye e s, 1947-1989, and 1339 M en Ever Em ployed in the Chem ical Divvision, 1 9 4 7 -1 9 8 9 Causes of Death All causes Cancer Gastrointestinal Colon Pancreas Respiratory Lung Prostate Testis Bladder Lymphopoietic Cardiovascular CHD* Cerebrovascular A l gastrointestinal A l respiratory Diabetes Injuries Suicide All M ale Em ployee. " en Chem ical Division Obs Exp SM R 95% Cl Obs Exp SMR 95% Cl 347 450.79 0.77 0.69-0.86 148 172.96 0.86 0.72-1.01 103 97.29 1.05 0.86-1.27 40 36.31 1.10 0.79-1.50 24 26.78 0.90 0.57-1.33 9 9.77 0.92 0.42-1.75 9 9.42 0.96 0.44-1.81 4 3.46 1.15 0.31-4.01 8 5.58 1.43 0.62-2.83 4 2.04 1.96 0.53-5.01 31 30.42 1.02 0.69-1.45 12 11.26 1.07 0.55-1.86 29 28.94 1.00 0.67-1.44 11 10.70 1.03 0.51-1.84 6 1 6.07 0.92 0.99 1.09 0.36-2.15 0.01-6.05 "sC4~ " 1.97 0.44 2.03 2.28 0.55-4.59 0.03-12.66 3 2.18 1.37 0.28-4.01 1 0.75 1.33 0.02-7.40 13 12.07 1.09 0.57-1.84 5 4.76 1.05 0.34-2.45 145 212.19 0.68 0.58-0.80 54 76.65 0.70 0.53-0.92 110 159.09 0.69 0.57-0.83 43 57.74 0.74 0.54-1.00 10 24.66 0.60 0.32-1.02 4 6.53 0.47 0.13-1.20 12 21.13 0.57 0.29-0.99 8 8.27 0.97 0.42-1.91 13 21.75 0.60 0.32-1.06 7 7.77 0.91 0.36-1.87 8 6.52 1.23 0.53-2.42 3 2.55 1.18 0.24-3.44 38 47.74 0.80 0.56-1.08 31 31.72 0.98 0.66-1.39 12 15.09 0.79 0.41-1.39 10 6.99 1.43 0.68-2.63 *CHD, coronary and atherosclerotic heart disease. were made between Chemical Divi sion and non-Chemical Division em ployees. There were no significantly elevated SMRs in Chemical Division or non-Chemical Division employees. However, prostate cancer mortality was associated with length o f employ ment in the Chemical Division in pro portional hazard analysis. Ten years of employment in the Chemical Di vision was associated with an esti mated 3.3-fold increase (95% Cl, 1.02 to 10.60) in prostate cancer mortality. The use of prostate cancer mortality to assess the association between PFOA and prostate cancer occurrence is problematic. Age-adjusted prostate cancer mortality rates from 1983 to 1989 (949 per 100,000) were only 25% of the incidence rates (99.4).27 This low proportion of deaths among cases attributed to prostate cancer re flects the high risk of death for com peting causes for this disease of elderly men. Given the small number of ob served deaths from prostate cancer in the study, and the observed difference in incidence and mortality rates, the suggested association between PFOA exposure and prostate cancer must be viewed as hypothesis generating and should not be overinterpreted. The association may be real, may have been a chance finding, or may be the result of an unrecognized environ mental factor. However, the biologic plausibility for any association be tween PFOA employment and pros tate cancer is provided by animal tox icologic and human epidemiologic data that show an association between PFOA and reproductive hormone changes.20 The all-causes, all-cancer, and allcardiovascular mortality among women was less than expected in the overall cohort. The low SMRs are most likely to be a result of the healthy worker effect. Latency and duration of plant employment did not have a strong relationship with the healthy worker effect. The interpretation of this study re quires consideration of methodologi cal issues. SMRs for the subgroups of workers are not strictly comparable. We attempted to calculate standard ized rate ratios; however, the rates were based on small numbers and pro duced unstable ratios. Estimates of PFOA exposure were based on job history, and categorization of workers into ever versus never employed in the Chemical Division may not reflect the biologic effective dose of PFOA. PFOA exposure was apparently wide spread among employees not directly exposed to PFOA,14and the exposure categorization may misclassify work ers as unexposed when they were ex- TABLE5 Proportional Hazard Regression Model of Factors Predicting Mortality among All Male Em ployees* V iriable All C a u s a l of Death SEW ) P RRf Cancer Deaths SEW ) P RRt Prostate Cancer Deaths SEW ) P RRf Year of first employment Age at first employment (y) Duration of employment (y) Months in chemical divi sion -0 .5 5 0.079 -0.34 0.001 0.009 0.006 0.001 0.001 0.0001 0.0001 0.0001 0.24 0.946 1.08 0.967 1.001 -0.031 0.078 -0 .0 2 8 0.002 0.019 0.011 0.009 0.001 0.11 0.0001 0.002 0.2 0.969 1.081 0.972 1.002 0.010 0.082 - n 07 0.01 0.081 0.045 0.052 0.005 0.9 0.06 0.18 0.03 1.011 1.085 0.932 1.01 * Abbreviations used are: 0. regression parameter SE(/J), standard error of the slope parameter: RR. relative nsk. t Relative risk for one unit change in independent variable. C03171 954 Gilliland & Mandel Mortality in a PFOA Production Plant posed. Such misclassification would be expected to bias the effect estimates toward the null if increased exposure increases death rates. Months em ployed in the Chemical Division may better reflect the biologic effective dose because cumulative exposure re flects the bioaccumulation of PFOA. Workers were exposed to many other xenobiotics, such as benzene and as bestos, during their employment at the plant. However, none of these ma terials has been associated with pros tate cancer. Although the mean age at first em ployment and mean year of first em ployment are similar in the Chemical Division and non-Chemical Division cohorts of men and women, the com parisons of the rates of disease are confounded by differences in the dis tribution of age at risk. The use of an internal comparison group may re duce, but not eliminate, confounding if the internal comparison groups have different distributions of these time factors. Because the disease oc currence relationship is defined in terms of cumulative exposure, the true effect of PFOA exposure may have been biased toward or away from the null by uncontrolled confounding by time factors.28,29 Further research is needed to eval uate and confirm the association be tween PFOA and prostate cancer. The findings in this study are based on a small number of cases and could have resulted from chance or unrecognized confounding from exposure to other factors. Studies of prostate cancer in cidence in this and other PFOA-exposed work forces may clarify the sug gested increase in prostate cancer risk. Acknowledgments This work was supported in part by Na tional Institute for Occupational Safety and Health Grant T150H07098-16. and the 3M Corporation. References I. Guenthner R, Vietor M. Surface active materials from perfluorocarboxylic and pcrtluorosull'omc acids. Ul EC Product Research and Development. 1962; I: 165- 169. 2. Griffith F. Long J. Animal toxicity stud ies with ammonium perfluorooctanoate. Am Ind Hyg Assoc J 1980;41:576-583. 3. Olson C, Andersen M. The acute toxicity of perfluorooctanoic and perfluorodecanoic acids in male rats and their effects on fatty tissue. Toxicol Appt Pharmacol. 1983;70:362-372. 4. Pastoor T. Lee K, Pern M. Gillies, P. Biochemical and morphological studies of ammonium perfluorooctanoate-induced hepatomegaly and peroxisome proliferation. Exp Mol Pathol. !987;47: 98-109. 5. Just W, Gorgas K, Haiti F, Heinemann P, Salazar M, Schimassek H. Biochemi cal effects and zonal heterogeneity of peroxisome proliferation induced by per fluorocarboxylic acids in rat liver. Hepatology. 1989;9:570-581. 6. Kennedy G. Dermal toxicity of ammo nium perfluorooctanoate. Toxicol Appl Pharmacol. 1985;81:348-355. 7. Kennedy G, Hall G, Brittelli J, Chen H. Inhalation toxicity of ammonium per fluorooctanoate. Food Chem Toxicol. 1986;24:1325-1329. 8. Cook J, Murray S, Frame S, H unt M. Induction of Leydig cell adenomas by ammonium perfluorooctanate: A possi ble endocrine related mechanism. Toxi colAppl Pharmacol. 1991;113:209-213. 9. Reddy J, Azamoff D, Hignite C. Hypo lipidemic hepatic peroxisom proliferators form a novel class of chemical car cinogens. Nature. 1980;283:397-398. 10. Reddy J, Lalwani N. Carcinogenesis by hepatic peroxisome proliferators: evalu ation of the risk of hypolipidemic drugs and industrial plasticizers to humans. Crit Rev Toxicol. 1983; 12:1-85. 11. Nilsson R, Beije B, Preat V, Erixon K, Ramel C. On the mechanism of the hepatocarcinogenicity of peroxisome proliferators. Chem Biol Interact. 1991; 78:235-250. 12. AbdellatifA, Preat V, Vamecq J, Nilsson R, Roberfroid M. Peroxisome prolifera tion and modulation of rat liver carci nogenesis by 2,3-dichlorophenoxyacetic acid, 2,4,5-trichlorophenoxyacetic acid, perfluorooctanoic acid and nafenopin. Carcinogenesis. 1990; 11:1899-1902. 13. Takagi A, Sai K, Umemura T, Hasegawa R, Kurokawa Y. Short-term exposure to the peroxisome proliferators. perfluo rooctanoic acid and perfluorodecanoic acid, causes significant increases of 8hydroxydeoxyguanosine in liver DNA of rats. Cancer Lett. 1991 ;57:55--60. 14. Ubel F. Sorenson S. Roach D. Health status of plant workers exposed to fluorochemicals: a preliminary report. Am Ind Hyg .I.vsw J 1980:41:584-589. 15. Guy W. Fluorocompounds of human plasma: analysis, prevalence, purifica tion. and characterization, PhD thesis. Rochester, NY: University of Rochester; 1972. 16. Guy W, Taves D, Brey W. Organic fluo rocompounds in human plasma: preva lence and characterization. In: Filler R, ed. Biochemistry involving carbon-fluo rine bonds. ACS Symposium Series. New York: American Chemical Society; 1976:117-134. 17. Guy W. Ionic and organic fluorine in blood. In: Johansen E, Taves D, Olsen T, ed. Continuing evaluation o f the use offluoride. Boulder, CO: Westview Press; 1979. 18. Taves D. Evidence that there are two forms of fluoride in human serum. Na ture. 1968;217:1050-1051. 19. Taves D. Comparison of "organic" flu oride in human and nonhuman serums. J Dent Res. 197!;50:783. 20. Gilliland F. Fluorocarbons and human health: studies in an occupational cohort, PhD thesis. Minneapolis, MN: Univer sity of Minnesota; 1992. 21. Berry G. The analysis of mortality by the subject-years method. Biometrics. 1983; 39:173-184. 22. Monson R. Analysis of relative survival and proportional mortality. Comput BiomedRes. 1974;7:325-332. 23. Miettinen O. Standardization of risk ratios. Am J Epidemiol. 1972;96: 383-388. 24. Cox D. Regression models and life tables. J R Stat Soc (B). 1972:34: 187-220. 25. SAS Institute. SAS User's Guide: Statis tics. 6th ed. Cary, NC: SAS Institute Inc; 1990. 26. Kalbfleisch J, Prentice R. The statistical analysis o ffailure time data. New York: John Wiley & Sons; 1980. 27. Miller R. Cancer statistics review 19731989. Bethesda, Maryland. US Depart ment of Health and Human Services, National Institutes of Health, National Cancer Institute. NIH Pub. No. 92-2789, 1992. 28. Checkoway H, Pierce N, CrawfordBrown D. Research methods in occupa tional epidemiology. New York: Oxford University Press; 1989. 29. Robins J. A new approach to causal in ference in mortality studies with a sus tained exposure period. Math Modeling. 1986;7:1393-1512. C03172